# The Race: Building Democracy in the Age of AI > Tracking whether AI concentrates power or disperses democratic capacity. Public Ghost content for AI and LLM tooling. This file includes a bounded export of public pages first, then recent public posts. Append `.md` to any post or page URL to get the content in Markdown (for example, `/example-post.md`). ## Pages ### About Mark Howard and The Race URL: https://www.theraceai.org/about-author/ Last updated: 2026-07-21T19:14:59.000Z **The Race: Building Democracy in the Age of AI* is an independent research and writing project by* [*Mark Howard*](https://www.linkedin.com/in/markhowardco/?ref=theraceai.org)*. It examines how artificial intelligence is changing the distribution of power—and whether democratic institutions, communities, workers, and smaller organizations will retain the capacity to shape what comes next.* My name is Mark Howard. I created *The Race* to examine a question that will shape the coming decades: **As artificial intelligence develops, will it concentrate power—or expand the capacity of people and democratic institutions to act?** I came to that question through three long-standing interests: history, technology, and democracy. History taught me that powerful technologies do not determine their own consequences. Their effects depend on the institutions, laws, markets, public investments, organizational choices, and social habits that grow around them. I am a technology optimist, although I hope a realistic one. I have seen technology extend human capability, make knowledge more accessible, and help organizations work more effectively. But useful technology does not automatically produce broadly shared benefits or broadly distributed power. ## My Perspective My approach is grounded in practical experience with institutions and technological change. I spent approximately ten years in city government, including service as a city manager. That work taught me that public policy is not only legislation and regulation. It is also budgets, contracts, staffing, utility systems, planning decisions, public meetings, administrative capacity, and the everyday work of turning policy into practice. Later, in leadership roles at Accenture, much of my job involved anticipating what was coming next. I worked to help clients prepare for technological and economic change while helping ensure that Accenture had the capabilities and trained people needed to respond. I began discussing the likely significance of machine learning with clients as early as 2014 and 2015. Those experiences shape how I view AI. Its consequences will not be decided only in research laboratories or corporate headquarters. They will also be determined through decisions about electricity, water, land, infrastructure, public procurement, education, employment, economic development, and access to essential services. ## What The Race Does *The Race* follows concrete developments in AI and asks what they are doing to the distribution of power. It looks beyond whether a particular system is impressive, whether AI is simply “good” or “bad,” or which company or country appears to be ahead. It examines who gains the ability to decide, build, participate, challenge decisions, and shape the conditions under which AI operates. The project does not assume that large institutions are inherently undemocratic. Governments need enough authority and competence to govern. Companies often need scale to invest and build. Regulation can protect the public, but it can also entrench the organizations best equipped to comply with it. The more useful question is whether power remains visible, accountable, and open to meaningful challenge. *The Race* is an evolving framework, not a claim to have settled the answers. Its categories and judgments will change as the technology develops and the evidence becomes clearer. ## How AI Is Used in This Project ChatGPT is part of how *The Race* is produced. I choose the questions, set the direction, make the analytical and editorial judgments, and take responsibility for the final work. ChatGPT helps locate and organize information, compare developments, test interpretations, and prepare early drafts. Its use requires care. Sources must be checked, arguments tested, and language revised. That is also part of the project: exploring how AI can help one person conduct sustained public-policy research while identifying where human experience, judgment, and responsibility remain essential. --- **Mark Howard** is the founder and author of *The Race: Building Democracy in the Age of AI*. He writes about artificial intelligence, democracy, public policy, institutional capacity, infrastructure governance, organizational change, and the shifts in political and economic power accompanying AI. ### The Race URL: https://www.theraceai.org/home/ Last updated: 2026-09-08T19:21:18.000Z # Building Democracy in the Age of AI **AI is changing who has power.** It can give workers, communities, public institutions, and smaller organizations new capacity to solve problems and act for themselves. It can also concentrate control in dominant companies, governments, security institutions, and technical elites. The outcome is not predetermined. *The Race* tracks concrete developments in AI and asks what they are doing to the distribution of power. **Who gains capacity? Who loses agency? Who gets to decide?** [How The Race works →](https://www.theraceai.org/framework/) --- ## Where the Race Stands ### Leaning Toward Concentration, but Still Contested June showed that the AI race is moving beyond models and software into the physical and institutional systems that determine who holds power: data centers, electricity, water, public procurement, frontier-model access, and workforce redesign. The overall picture is mixed and contested. Concentrated control is hardening in several lanes, but local governance, public inventories, open models, and new accountability tools are keeping parts of the window open. [Read the latest Scoreboard Review →](https://www.theraceai.org/scoreboard/) --- ## Featured Posts ### Utility Governance Is Democracy’s Entry Point Into the AI Race AI’s physical expansion depends on electricity, water, land, and grid access. That gives utilities, regulators, and local governments a crucial opening to shape who pays, who benefits, and whether the public retains a meaningful say in the AI race. [Read the post →](https://www.theraceai.org/utility-governance-is-democracys-entry-point-into-the-ai-race/) --- ## Latest from The Race [Browse all posts →](https://www.theraceai.org/tag/articles/) --- ## The Nine Lanes AI power is not being determined in one place. It is being shaped through laws and public institutions, national strategy, computing infrastructure, platforms, organizational choices, public finance, and the practical ability of people and institutions to use new tools. *The Race* follows these developments through nine connected lanes. The lanes are not separate races. They are nine views of the same one. [Explore the Nine Lanes →](https://www.theraceai.org/the-eight-lanes/) --- ## Follow the Race The significance of any one development is often unclear. The larger direction becomes visible as patterns form across governments, industries, technologies, and communities. Subscribe for new posts and newsletters. \[Subscribe\]\] ### The Scoreboard URL: https://www.theraceai.org/scoreboard/ Last updated: 2026-09-08T20:12:11.000Z The Scoreboard tracks whether artificial intelligence is concentrating power or dispersing meaningful capacity—and whether the choices shaping that outcome remain open to challenge. It follows nine connected lanes. The judgment is not based on any one announcement or event. It comes from the patterns forming across infrastructure, government, platforms, public institutions, and the organization of work. ## July 2026 ### Overall Reading: Leaning Toward Concentration, but Still Contested July’s strongest public actions show that the window of contestability remains real. States can create enforceable duties. Regulators can reach critical providers. Competition authorities can require access at controlled platform layers. Communities can organize, pause, appeal, and sometimes alter infrastructure trajectories. Public institutions can also begin to purchase shared capacity rather than merely subsidize private scale. But the month’s harder-to-reverse developments point in the opposite direction. Power and compute were committed for long periods. Federal agencies became gatekeepers for sites, recipients, and suppliers. AI providers moved deeper into institutional and personal operating layers. Firms began reorganizing teams and authority around AI before worker-side institutions could respond. The central July lesson is therefore not that the window closed. It is that the actors capable of making durable commitments are still moving faster than the institutions capable of contesting them. [Read the full July Scoreboard Review →](https://www.theraceai.org/8-lane-scoreboard-review/) --- ## July 2026 Eight-Lane Snapshot Lane 9 — Public Finance and Fiscal Capacity was added after the July review. It will first appear in the August 2026 Scoreboard. | Lane | Current Reading | Window Status | | ---------------------------------------------------- | --------------------- | ------------- | | 1\. Governance and Regulation | Mixed and contested | Narrowing | | 2\. National Security and Frontier Control | Leaning concentration | Narrowing | | 3\. Compute and Physical Infrastructure | Leaning concentration | Narrowing | | 4\. Infrastructure Miniaturization | Mixed and contested | Open | | 5\. Platforms, Distribution, and Separations | Mixed and contested | Narrowing | | 6\. Civic Infrastructure and Public Capacity | Leaning dispersion | Open | | 7\. Implementation Capacity | Leaning concentration | Narrowing | | 8\. Organizational Forms and Workforce Restructuring | Leaning concentration | Narrowing | The full monthly review explains the evidence and reasoning behind each lane’s current reading. [Read the full July Scoreboard Review →](https://www.theraceai.org/8-lane-scoreboard-review/) [What the Nine Lanes cover →](https://www.theraceai.org/the-nine-lanes/) --- ## Overall read for July July preserved the window of contestability but worsened the underlying balance of power. Public institutions created meaningful counterweights through law, direct supervision, interoperability mandates, civic participation, infrastructure pauses, shared implementation tools, and publicly anchored compute access. At the same time, the month’s most durable commitments formed around long-term infrastructure allocation, federal access and supplier gatekeeping, persistent agent platforms, and top-down organizational restructuring. **Overall direction:** Mixed-contested, leaning concentration **Overall window:** Open but narrowing **Overall confidence:** High **Transparency / contestability overlay:** Mixed, with the largest gaps in infrastructure bargains, frontier-access decisions, political financing networks, platform context, and workforce restructuring evidence. ### Strongest concentration signals - Long-duration allocation of scarce compute and power capacity to undisclosed or privately selected users. - Federally gated private infrastructure combining federal sites, regulatory pathways, strategic suppliers, and private ownership. - Persistent enterprise and consumer agent layers accumulating institutional context, workflow history, permissions, and action authority. - Expansion of frontier control into outputs, distillation, robotics, power inverters, and trusted-recipient systems. - Organizational recomposition proceeding faster than worker-side institutions and public transition capacity. ### Strongest dispersion signals - Illinois independent audits, incident reporting, whistleblower protections, and statewide K–12 governance resources. - UK direct supervision of critical cloud providers. - European Commission binding interoperability and Search-data access requirements. - New York’s hyperscale-permit pause and the QTS / Digital Gateway termination. - National civic mobilization, local deliberation, and emerging municipal shared-capacity networks. - EU procurement of shared frontier-scale compute access. ### Most contested lanes - **Lane 03:** local and public review against structural infrastructure lock-in. - **Lane 05:** binding separations against rapidly accumulating platform context and workflow control. - **Lane 01:** operative public governance against privately financed political agenda-setting. - **Lane 04:** model-layer openness against compute- and implementation-layer concentration. ### Lanes where the window is narrowing fastest - Lane 02 - National Security and Frontier Control - Lane 03 - Compute and Physical Infrastructure - Lane 05 - Platforms, Distribution, and Separations - Lane 08 - Organizational Forms and Workforce Restructuring ### Lanes where the window remains most open - Lane 04 - Infrastructure Miniaturization - Lane 06 - Civic Infrastructure and Public Capacity - Lane 01 remains shapeable, but the window is narrowing as institutional and political architectures harden. --- ## How the Scoreboard Works The Scoreboard records concrete developments when they occur, but it does not assume that their long-term meaning is settled. A law may be enacted but poorly enforced. A public commitment may prove substantial or symbolic. A new service may widen access while creating new dependence. An early concentration signal may later be offset by stronger public rules or practical alternatives. The Scoreboard therefore distinguishes among: - What has happened. - What direction it appears to move power. - Whether the opportunity for challenge remains open. - What future developments could change the judgment. [Read the Scoreboard methodology →](https://www.theraceai.org/scoreboard-methodology/) --- ## Previous Reviews Monthly Scoreboard Reviews provide the supporting evidence and fuller analysis behind the current page. [Browse previous Scoreboard Reviews →](https://www.theraceai.org/tag/scoreboard-review/) ### How The Race Works URL: https://www.theraceai.org/framework/ Last updated: 2026-09-08T20:11:13.000Z *The Race* is not intended to be a catalogue of everything happening in artificial intelligence. It is a way of judging what important developments mean for the distribution of power. A new model, law, data center, workplace system, or public investment may be significant for many reasons. The question for *The Race* is more specific: **Does it give more people and institutions meaningful capacity to act, or does it place greater control beyond their reach?** The answer is rarely simple. The same development can widen access while creating new dependencies. It can protect the public while strengthening established institutions. It can distribute useful tools while concentrating control over the infrastructure on which those tools depend. The purpose of the framework is to make those tensions visible. --- ## Concentration, Dispersion, and Contestable Scale *The Race* uses concentration and dispersion as its basic way of interpreting change. **Concentration** occurs when control over important capabilities, resources, systems, or decisions moves into fewer hands. That concentration can come from market dominance, government authority, control of infrastructure, technical complexity, ownership of data, or rules that only the largest institutions can satisfy. **Dispersion** occurs when more people and institutions gain meaningful capacity to act. That may come through lower costs, wider access to tools, competition, open standards, shared infrastructure, stronger public institutions, or rules that make powerful systems more accountable. Dispersion does not mean that every person or small organization must be able to build and operate everything independently. Some technologies require large investments. Some infrastructure must operate at regional, national, or global scale. Governments and other large institutions must be capable of acting effectively. The question is whether that scale remains **contestable**. Can others enter the field, gain access to essential resources, challenge important decisions, and influence how power is used? Are there meaningful alternatives? Can public institutions establish and enforce conditions? Or does scale become a way of placing important choices beyond competitive or democratic reach? This is why *The Race* does not assume that large institutions are necessarily concentrating power or that smaller technologies are necessarily dispersing it. A large public institution can create shared capacity that would otherwise not exist. A small tool can still deepen dependence on a dominant platform. A regulation can protect rights while raising barriers to entry. A widely available AI service can expand individual capability while giving its provider greater control over information, distribution, and access. The framework asks both sides of the question. --- ## The Nine Lanes AI power is being organized across nine connected arenas: 1. **Governance and Regulation** 2. **National Security and Frontier Control** 3. **Compute and Physical Infrastructure** 4. **Infrastructure Miniaturization** 5. **Platforms, Distribution, and Separations** 6. **Civic Infrastructure and Public Capacity** 7. **Implementation Capacity** 8. **Organizational Forms and Workforce Restructuring** 9. **Public Finance and Fiscal Capacity** The lanes provide structure without treating developments as isolated. A data center may begin as an infrastructure story, but it can also affect electric rates, water supplies, land-use authority, national strategy, and access to computing power. An AI system introduced into a workplace may affect productivity, professional judgment, surveillance, organizational structure, and bargaining power. Primary and secondary lanes help identify those connections. They are not intended to force every development into only one category. [Explore the Nine Lanes →](https://www.theraceai.org/the-nine-lanes/) --- ## How the Scoreboard Works The Scoreboard brings developments across the nine lanes into a broader judgment about the direction of the Race. It does not measure AI capability in the abstract. It assesses how concrete actions affect capacity, control, agency, and accountability. The Scoreboard considers several questions: - Does the development move power toward greater concentration or greater dispersion? - Does it open or close opportunities for competition, public participation, and institutional challenge? - Does it create meaningful new capacity, or only promise that capacity? - Are its effects immediate, or will they depend on later implementation? - Does a gain in access create a new dependency elsewhere? - Could a development in one lane alter conditions in several others? The Scoreboard records concrete developments when they occur. It does not wait for every long-term consequence to become clear. But scoring an event does not end the inquiry. Implementation may strengthen, weaken, or reverse the initial judgment. A law may be enacted but poorly enforced. A new service may expand access but later become more restrictive. A public commitment may prove significant or largely symbolic. The Scoreboard therefore distinguishes between what has happened, what it appears to mean, and what could change that reading. No single event determines the overall result. Patterns matter more than isolated announcements. The Scoreboard is a structured judgment rather than a mechanical index. Evidence provides the foundation, but interpretation is still necessary. The central questions remain: Who now has greater capacity to act? Who has gained control? What alternatives remain available? And who can challenge the decisions being made? Public Scoreboard Reviews are normally published monthly. Quarterly reviews take a wider view of the patterns that appear to be becoming more durable. [Read the Scoreboard methodology →](https://www.theraceai.org/scoreboard-methodology/) --- ## How to Read The Race *The Race* publishes individual posts examining important developments, patterns, institutions, and policy questions, along with newsletters that periodically bring that work together. Monthly Scoreboard Reviews assess the cumulative direction of the Race. They are recurring posts rather than a separate publishing category. Not every important development becomes a published post. Much of the work behind *The Race* consists of tracking events, testing interpretations, comparing developments across lanes, and waiting to see whether an isolated example becomes part of a larger pattern. Publication is reserved for developments and ideas that help clarify that larger direction. [Browse all posts →](https://www.theraceai.org/tag/articles/) --- ## The Democratic Standard *The Race* does not begin from the assumption that AI should be stopped, that scale is inherently dangerous, or that every powerful institution is acting against the public interest. AI will require capable companies, effective governments, substantial infrastructure, technical expertise, and institutions able to carry out difficult work. But capability and accountability must develop together. Democracy in the age of AI will require more than elections and formal rights. It will also require institutions able to understand new systems, establish enforceable rules, provide shared capacity, protect the public, and give people meaningful ways to challenge important decisions. The question is not whether AI becomes powerful. It is whether that power remains open to challenge, whether its benefits become broadly usable, and whether the public retains a meaningful say in how the AI race is run. ### The Nine Lanes URL: https://www.theraceai.org/the-nine-lanes/ Last updated: 2026-09-08T20:06:07.000Z AI is changing power through laws, national strategy, infrastructure, platforms, public institutions, and the organization of work. *The Race* follows those changes through nine connected lanes. The lanes are not separate races. They are different ways of asking the same questions: **Who gains capacity? Who gains control? Who gets to decide?** Each post is assigned a primary lane and, when useful, additional lane tags. --- ## Lane 1 — Governance and Regulation This lane follows the laws, regulations, court decisions, standards, and public institutions shaping AI. Governance can protect rights, require accountability, and keep important decisions open to public challenge. It can also expand government power or create rules that only the largest organizations can satisfy. The question is not simply whether AI is regulated. It is who gains power through the way it is regulated. [Read Lane 1 posts →](https://www.theraceai.org/tag/lane-1-governance/) --- ## Lane 2 — National Security and Frontier Control This lane follows government efforts to control advanced models, computing power, chips, technical talent, and other strategic resources. Some controls may be necessary for security. But they may also place more of the AI system behind closed doors and strengthen a small group of governments and approved companies. The question is how to protect security without making frontier control permanent and largely beyond public reach. [Read Lane 2 posts →](https://www.theraceai.org/tag/lane-2-national-security/) --- ## Lane 3 — Compute and Physical Infrastructure AI depends on data centers, chips, electricity, water, land, transmission systems, construction, and financing. These resources can expand shared capacity. They can also give lasting advantages to institutions able to secure scarce infrastructure and negotiate large commitments years in advance. This lane asks who controls the physical foundations of AI, who pays for them, and whether the public has a meaningful role in the decisions. [Read Lane 3 posts →](https://www.theraceai.org/tag/lane-3-compute-infrastructure/) --- ## Lane 4 — Infrastructure Miniaturization This lane follows smaller models, more efficient chips, local computing, edge devices, and other technologies that reduce the need for centralized infrastructure. Smaller systems may lower costs, improve privacy, and give more people and organizations useful AI capacity. But smaller technology does not necessarily create independence. It may still depend on software, data, standards, and platforms controlled elsewhere. The question is whether miniaturization creates real alternatives. [Read Lane 4 posts →](https://www.theraceai.org/tag/lane-4-miniaturization/) --- ## Lane 5 — Platforms, Distribution, and Separations Most people will reach AI through platforms, cloud services, operating systems, workplace software, app stores, and other intermediaries. Those systems can distribute useful tools quickly. They can also control access, pricing, data, visibility, and the terms on which others participate. This lane asks whether AI reaches people through open and contestable pathways—or through gateways controlled by a small number of institutions. [Read Lane 5 posts →](https://www.theraceai.org/tag/lane-5-platforms/) --- ## Lane 6 — Civic Infrastructure and Public Capacity This lane follows whether governments, schools, libraries, universities, nonprofit organizations, utilities, and other civic institutions can understand, use, and govern AI. Public institutions do not need to build every system themselves. But they need enough expertise, staffing, procurement capacity, and independence to make informed choices and enforce public conditions. The question is whether democratic institutions are gaining the capacity to act—or merely responding to decisions made elsewhere. [Read Lane 6 posts →](https://www.theraceai.org/tag/lane-6-civic-capacity/) --- ## Lane 7 — Implementation Capacity Access to AI is not the same as the ability to use it well. Organizations need skilled people, good data, workable systems, leadership, trust, and the ability to change how work is done. Large institutions may be better able to turn AI into practical results even when the tools are widely available. This lane asks who can convert access into real capacity—and who remains dependent on others to do it. [Read Lane 7 posts →](https://www.theraceai.org/tag/lane-7-implementation/) --- ## Lane 8 — Organizational Forms and Workforce Restructuring This lane follows how AI changes jobs, professions, management, firms, and the division of work between people and machines. AI may allow smaller teams to do more, reduce routine burdens, and strengthen human judgment. It may also increase surveillance, weaken bargaining power, narrow professional discretion, or move authority farther from workers. The question is not only whether jobs disappear. It is how work, responsibility, and decision-making power are reorganized. [Read Lane 8 posts →](https://www.theraceai.org/tag/lane-8-workforce/) --- ## Lane 9 — Public Finance and Fiscal Capacity This lane follows taxes, subsidies, public borrowing, utility rates, fiscal incentives, and other ways governments finance AI or absorb its costs. Public finance can build shared capacity and help communities capture public value. It can also shift costs and risks to taxpayers or ratepayers while benefits flow to a narrow group of companies. This lane asks who pays, who benefits, and whether public resources strengthen democratic capacity or entrench private power. [Read Lane 9 posts →](https://www.theraceai.org/tag/lane-9-public-finance/) --- ## One Race, Nine Views The lanes overlap because the underlying decisions overlap. A data center can affect infrastructure, public finance, utility governance, local authority, national strategy, and market access. A workplace AI system can raise questions about regulation, platforms, implementation, and worker autonomy. The primary lane identifies the center of a development. Secondary lanes show the most important connections. The larger question remains whether AI is creating broadly usable and contestable capacity—or placing more power beyond meaningful challenge. ### July 2026 8-Lane Scoreboard Review URL: https://www.theraceai.org/8-lane-scoreboard-review/ Last updated: 2026-08-03T17:14:38.000Z ## Lane 01 - Governance and Regulation - **Direction of power:** Mixed-contested - **Window status:** Narrowing - **Confidence:** Medium-high - **Transparency / contestability overlay:** Mixed - **Current read:** July produced some of the month’s strongest public-governance gains, but the institutional politics of AI regulation is becoming more privately financed and professionally organized. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------- | ------------------- | -------------------------------------------------------------------------------------------------------------- | | 2026-07-06 | [Illinois Artificial Intelligence Safety Measures Act](https://www.ilga.gov/Legislation/BillStatus?DocNum=315&DocTypeID=SB&GAID=18&LegId=157797&Print=1&SessionID=114&ref=theraceai.org) | Dispersion +2 | Preserves | Creates enforceable risk frameworks, independent audits, incident reporting, and whistleblower protections. | | 2026-07-10 | [UK critical cloud-provider supervision](https://www.gov.uk/government/news/uk-financial-system-strengthened-with-new-safeguards-for-major-technology-providers?ref=theraceai.org) | Dispersion +2 | Preserves / reopens | Extends direct public authority to concentrated dependency points. | | 2026-07-15 | [Australia establishes Office of AI](https://www.reuters.com/world/asia-pacific/australia-establish-government-ai-office-coordinate-regulation-2026-07-14/?ref=theraceai.org) | Dispersion +1 | Preserves | Builds central public coordination, but authority and transparency remain incomplete. | | 2026-07-21 | [Anthropic doubles Public First Action support](https://www.anthropic.com/news/donation-public-first-action?ref=theraceai.org) | Mixed 0 | Preserves | Creates countervailing political capacity while increasing frontier-company influence over governance debates. | | 2026-07-23 | [Canada launches AI-transparency consultation](https://www.canada.ca/en/innovation-science-economic-development/news/2026/07/government-of-canada-launches-public-consultation-on-ai-transparency.html?ref=theraceai.org) | Dispersion +1 | Preserves | Opens a formal public process on disclosure, incident reporting, and agent traceability. | ### Interpretation The regulatory lane is still genuinely contested. July demonstrates that public institutions can create enforceable duties and reach concentrated dependency points. The concentration risk is shifting from the absence of rules toward the architecture surrounding rule formation: compliance capacity, private political networks, federal preemption, closed standards processes, and company influence over which regulatory approaches receive sustained support. ### What would change this read Move toward dispersion if public reporting, enforcement, independent evaluation, state experimentation, and broad public-interest participation become durable. Move toward concentration if federal preemption, privately financed agenda-setting, compliance barriers, or industry-shaped standards narrow the range of viable governance models. --- ## Lane 02 - National Security and Frontier Control - **Direction of power:** Leaning concentration - **Window status:** Narrowing - **Confidence:** High - **Transparency / contestability overlay:** Negative / mixed - **Current read:** Frontier governance is becoming a multi-layer control architecture extending from chips and cloud access to model outputs, distillation, robotics, power systems, trusted recipients, and emergency intervention. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------- | --------------- | --------------------------------------------------------------------------------------------------- | | 2026-07-10 | [UAE trusted-recipient access to advanced computing](https://www.bis.gov/press-release/department-commerce-eases-export-controls-uae?ref=theraceai.org) | Concentration +2 | Narrows | Allocates strategic access through government-approved recipients and geopolitical conditions. | | 2026-07-14 | [Gold Eagle cybersecurity clearinghouse](https://www.whitehouse.gov/releases/2026/07/white-house-launches-gold-eagle-initiative-for-unprecedented-cybersecurity-vulnerability-coordination/?ref=theraceai.org) | Mixed 0 | Preserves | Could diffuse defensive capability or consolidate access inside a federal-industry network. | | 2026-07-21 | [OpenAI agent containment failure](https://huggingface.co/blog/agent-intrusion-technical-timeline?ref=theraceai.org) | Concentration +1 | Narrows | Demonstrates real autonomous capability and provider-controlled testing and disclosure. | | 2026-07-28 | [FCC foreign-equipment restrictions](https://www.fcc.gov/document/fcc-adds-foreign-produced-power-inverters-and-robots-covered-list?ref=theraceai.org) | Concentration +2 | Narrows sharply | Extends federal gatekeeping into robotics and power-conversion equipment. | | 2026-07-31 | [Military-linked distillation evidence](https://www.reuters.com/world/asia-pacific/chinese-military-researchers-tap-us-ai-models-train-defence-systems-2026-07-31/?ref=theraceai.org) | Watch / concentration-leaning | Narrows | Shows useful capability moving through outputs and synthetic data beyond original provider control. | ### Interpretation The relevant control surface is now broader than frontier-model release. Security authority is being exercised over infrastructure, suppliers, recipients, outputs, testing conditions, and derived capability. The principal contestability deficit is the absence of generally applicable public criteria, evidence standards, review procedures, and transparency substitutes where secrecy is necessary. ### What would change this read Move toward a more mixed assessment if access rules, Conditional Approvals, emergency powers, and trusted-recipient regimes gain public criteria, independent review, reporting, public-interest access, and meaningful appeals. Move further toward concentration if discretionary public-private arrangements become the durable default. --- ## Lane 03 - Compute and Physical Infrastructure - **Direction of power:** Leaning concentration - **Window status:** Narrowing - **Confidence:** High - **Transparency / contestability overlay:** Mixed / negative - **Current read:** July’s public counterweights were meaningful, but private and federally enabled commitments became larger, longer-lived, and more integrated than the institutions reviewing them. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ----------------------------- | -------------------------- | ---------------------------------------------------------------------------------- | | 2026-07-05 | [QTS terminates Digital Gateway project](https://www.reuters.com/business/blackstones-qts-terminates-digital-gateway-data-center-project-virginia-2026-07-02/?ref=theraceai.org) | Dispersion +2 | Reopens | Demonstrates that approval does not end contestability. | | 2026-07-14 | [New York pauses incomplete hyperscale permits](https://www.governor.ny.gov/executive-order/no-62-establishing-temporary-moratorium-data-centers-new-york-while-state-develops?ref=theraceai.org) | Dispersion +2 | Preserves / reopens | Creates time for linked environmental, grid, water, and community standards. | | 2026-07-20 | [Hut 8 Beacon Point allocation](https://www.hut8.com/news-insights/press-releases/hut-8-fully-commercializes-1-gw-beacon-point-ai-data-center-campus-with-second-352-mw-it-lease?ref=theraceai.org) | Concentration +2 | Narrows | Commits 704 MW to an undisclosed tenant for 15 years. | | 2026-07-22 | [Northern Virginia load disturbance](https://www.datacenterdynamics.com/en/news/pjm-grid-hit-by-voltage-disturbance-after-data-center-load-abruptly-drops-offline-report/?ref=theraceai.org) | Mixed / concentration-leaning | Narrows | Demonstrates system-scale operational effects of concentrated computational loads. | | 2026-07-27 to 2026-07-30 | [EPA islanded power](https://www.epa.gov/stationary-sources-air-pollution/clean-air-act-resources-data-centers?ref=theraceai.org); [Hyperion](https://www.opportunitylouisiana.gov/data-center/meta?ref=theraceai.org); [Paducah](https://www.energy.gov/articles/energy-department-announces-partnership-expand-reliable-affordable-energy-access-and-power-americas-ai?ref=theraceai.org); [EU Gigafactory procurement](https://digital-strategy.ec.europa.eu/en/news/eu-launches-ai-gigafactories-call-boost-europes-computing-capacity-and-unlock-more-eu30-billion?ref=theraceai.org) | Mixed, concentration-leaning | Narrows with counterweight | Reveals federally gated private infrastructure and a public-access counter-model. | ### Interpretation The lane moved from mixed-contested to leaning concentration. Local moratoria, litigation, appeals, utility proceedings, and public procurement remain consequential, but they are increasingly defensive responses to long-duration capacity allocation, integrated private bargaining, parallel power systems, and federal-site reuse. Public institutions need the ability to assess the whole bargain before separate approvals harden. ### What would change this read Move back toward mixed-contested if beneficiary-pays rules, integrated review, federal lease conditions, public-service obligations, reliability standards, and public compute-access guarantees become enforceable. Move further toward concentration if dedicated power and federal-site models scale without equivalent public visibility, cost discipline, or exit and restoration duties. --- ## Lane 04 - Infrastructure Miniaturization - **Direction of power:** Mixed-contested - **Window status:** Open - **Confidence:** Medium - **Transparency / contestability overlay:** Watch / mixed - **Current read:** Model-layer separability improved in July, but the institutions capable of operating, securing, and adapting the most capable released systems remain narrow. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------ | --------------------- | ------------------------------------------------------------------------------------------------------ | | 2026-07-24 | [Open-weight industry coalition](https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf?ref=theraceai.org) | Leaning dispersion | Preserves | Strengthens political support for locally deployable and substitutable models. | | 2026-07-27 | [Open Secure AI Alliance](https://blogs.nvidia.com/blog/open-secure-ai-alliance/?ref=theraceai.org) | Dispersion +1 | Opens | Organizes production of shared defensive models, data, tools, and evaluations. | | 2026-07-27 | [Kimi K3 full-weight release](https://arxiv.org/abs/2607.24653?ref=theraceai.org) | Dispersion +1 | Opens | Creates real separation from Moonshot’s hosted API. | | July 2026 | [GLM-5.2 developer use](https://www.reuters.com/world/china/a-new-inexpensive-chinese-ai-model-is-catching-up-with-anthropic-openai-their-2026-07-02/?ref=theraceai.org) and [forensic use](https://huggingface.co/blog/agent-intrusion-technical-timeline?ref=theraceai.org) | Mixed-contested | Reopens / complicates | Shows lower-cost open-weight capability narrowing the proprietary frontier gap. | | 2026-07-31 | [Military-linked distillation evidence](https://www.reuters.com/world/asia-pacific/chinese-military-researchers-tap-us-ai-models-train-defence-systems-2026-07-31/?ref=theraceai.org) | Concentration / control-bypass watch | Narrows | Demonstrates that smaller systems can inherit selected capability while accountability becomes harder. | ### Interpretation Miniaturization remains the most open lane because capability can move into downloadable, local, private, smaller, or specialized systems. But model files alone do not create contestable scale. Compute, hardware, implementation skill, cybersecurity, licensing, evaluation, and distribution determine whether technical openness becomes usable institutional independence. ### What would change this read Move toward dispersion if public agencies, smaller firms, schools, clinics, and civic institutions can independently deploy portable models with auditable governance and realistic hardware requirements. Move toward concentration if local systems remain wrappers around proprietary services or depend on closed device, cloud, orchestration, and security stacks. --- ## Lane 05 - Platforms, Distribution, and Separations - **Direction of power:** Mixed-contested - **Window status:** Narrowing - **Confidence:** Medium-high - **Transparency / contestability overlay:** Mixed - **Current read:** July established both the strongest practical separations remedy and some of the clearest evidence that assistants are becoming persistent operating layers. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------------------- | ------------------- | ------------------------------------------------------------------------------------------------------------ | | 2026-07-07 | [Claude Cowork persistent cloud layer](https://www.theverge.com/ai-artificial-intelligence/961978/anthropic-claude-cowork-mobile-web?ref=theraceai.org) | Concentration-leaning | Narrows | Extends identity, files, task history, scheduled work, and remote execution into provider-hosted continuity. | | 2026-07-16 | [European Commission Android and Search-data decisions](https://digital-markets-act.ec.europa.eu/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data-under-2026-07-16%5Fen?ref=theraceai.org) | Dispersion +2 | Preserves / reopens | Requires equivalent access and data access at controlled distribution layers. | | 2026-07-22 | [OpenAI Presence](https://openai.com/index/introducing-openai-presence/?ref=theraceai.org) | Concentration +1 | Narrows | Bundles models, policies, evaluation histories, connectors, monitoring, and implementation support. | | 2026-07-24 to 2026-07-29 | [Meta persistent agent layer](https://www.reuters.com/business/meta-adds-new-task-automation-features-ai-assistant-2026-07-24/?ref=theraceai.org) | Concentration +1 | Narrows | Extends recurring tasks, context, messaging, and small-business workflows across Meta surfaces. | | 2026-07-27 | [Kimi K3 full-weight release](https://arxiv.org/abs/2607.24653?ref=theraceai.org) | Dispersion +1 | Opens | Separates a capable model from the provider API while leaving infrastructure dependence. | ### Interpretation The lane’s central distinction is now between formal openness and practical substitutability. Binding access rights matter, but a user or institution is not free to switch if memory, policies, evaluations, connectors, task histories, permissions, and functioning workflows cannot move with it. ### What would change this read Move toward dispersion if Commission enforcement produces real rival adoption and if providers support exportable context, multi-model operation, default choice, comparable third-party access, and tested workflow migration. Move toward concentration if persistent agents become cross-service defaults before those rights exist. --- ## Lane 06 - Civic Infrastructure and Public Capacity - **Direction of power:** Leaning dispersion - **Window status:** Open - **Confidence:** Medium - **Transparency / contestability overlay:** Mixed / improving - **Current read:** July showed public and civic capacity beginning to become reusable infrastructure rather than isolated participation or one-off experimentation. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------- | ------------------- | -------------------------------------------------------------------------------------------------------------------------- | | 2026-07-08 | [Snohomish County civic assembly](https://ourcivicgenius.org/learn/final-report-snohomish-county-civic-assembly-on-ai/?ref=theraceai.org) | Participatory anchor | Opens / preserves | Places residents upstream of public AI adoption and governance conditions. | | 2026-07-09 | [Illinois statewide K–12 AI framework](https://www.isbe.net/Lists/News/DispForm.aspx?ID=1593&ref=theraceai.org) | Dispersion +2 | Opens | Gives districts reusable governance, procurement, privacy, training, and participation tools. | | 2026-07-18 | [National data-center protest mobilization](https://www.reuters.com/business/retail-consumer/us-data-center-protests-go-national-backlash-grows-2026-07-18/?ref=theraceai.org) | Dispersion +1 | Preserves / reopens | Converts fragmented local disputes into shared civic capacity. | | 2026-07-20 | [UK central AI minister](https://www.gov.uk/government/ministers/minister-of-state-minister-for-artificial-intelligence?ref=theraceai.org) | Mixed 0 | Preserves | Builds coordination capacity while concentrating executive agenda-setting. | | July 2026 | [GovAI Coalition](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/artificial-intelligence-inventory/govai-coalition/templates-resources?ref=theraceai.org), [NLC](https://www.nlc.org/initiative/ai-in-action-empowering-local-governments/?ref=theraceai.org), [ICMA](https://icma.org/blog-posts/ai-local-level-what-municipal-leaders-told-us?ref=theraceai.org), and municipal framework pattern | Supporting / dispersion-leaning | Preserves | Shared policies, inventories, contracts, training, and peer learning reduce the need for each jurisdiction to build alone. | ### Interpretation The seven-part public layer is becoming more complete. Visibility, rules, experimentation, workforce capacity, and shared resources all improved. Contestability remains uneven, and portability / independence remains the principal gap. Public institutions are learning how to govern and use AI, but not yet how to preserve practical independence from vendors and platforms. ### What would change this read Move further toward dispersion if shared public networks produce measurable adoption, bargaining power, independent expertise, enforceable participation, and tested provider replacement. Move back toward mixed if public capacity remains dependent on vendor-defined systems or consultation remains visible but nonconsequential. --- ## Lane 07 - Implementation Capacity - **Direction of power:** Leaning concentration - **Window status:** Narrowing - **Confidence:** Medium-high - **Transparency / contestability overlay:** Mixed - **Current read:** Public implementation infrastructure improved in July, but accumulated deployment knowledge remains concentrated and compounds inside a small set of firms and high-capacity institutions. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | -------------------------------- | ------------- | ----------------------------------------------------------------------------------------------------------------- | | 2026-07-09 | [Illinois K–12 framework](https://www.isbe.net/Lists/News/DispForm.aspx?ID=1593&ref=theraceai.org) | Dispersion +2, secondary lane | Opens | Distributes usable implementation guidance across public school systems. | | 2026-07-13 | [UK Financial Services Skills Compact](https://www.gov.uk/government/publications/financial-services-skills-compact--2?ref=theraceai.org) | Dispersion +1 | Preserves | Creates sector-wide training, entry-route, accountability, and reporting commitments. | | 2026-07-22 | [OpenAI Presence](https://openai.com/index/introducing-openai-presence/?ref=theraceai.org) | Concentration +1, secondary lane | Narrows | Reduces implementation friction while accumulating customer context and learning inside a provider-managed layer. | | July 2026 | [Municipal shared-capacity networks](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/artificial-intelligence-inventory/govai-coalition/templates-resources?ref=theraceai.org) | Supporting / dispersion-leaning | Preserves | Spread policies, procurement tools, training, inventories, and incident resources. | | July 2026 | [Public procurement and shared compute efforts](https://digital-strategy.ec.europa.eu/en/news/eu-launches-ai-gigafactories-call-boost-europes-computing-capacity-and-unlock-more-eu30-billion?ref=theraceai.org) | Mixed-contested | Preserves | Create access but do not yet establish robust portability or independent operating capacity. | ### Interpretation July confirms that access is not capacity. Public institutions are beginning to build common resources, but major providers and implementation intermediaries still capture the most valuable learning about workflow redesign, evaluation, integration, and agent operation. Portability must cover implementation knowledge, not only data files. ### What would change this read Move toward dispersion if public agencies and smaller institutions can retain and share evaluation histories, contract lessons, workflow designs, staff knowledge, and functioning configurations across vendors. Move further toward concentration if implementation expertise becomes inseparable from provider-managed platforms and consulting relationships. --- ## Lane 08 - Organizational Forms and Workforce Restructuring - **Direction of power:** Leaning concentration - **Window status:** Narrowing - **Confidence:** Medium-high - **Transparency / contestability overlay:** Negative / mixed - **Current read:** Organizational recomposition is the leading near-term pattern: firms are redesigning layers, teams, roles, hiring, outsourcing, and human-agent workflows before aggregate direct replacement can be established. ### Evidence this period | Date | Event | Event score | Window effect | Notes | | ---------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------- | ------------- | --------------------------------------------------------------------------------------------------------------- | | 2026-07-08 | [Allianz Partners AI-linked role elimination](https://www.reuters.com/legal/litigation/allianz-cut-up-1800-jobs-due-increasing-ai-use-2026-07-08/?ref=theraceai.org) | Concentration +2 | Narrows | Concrete AI-linked elimination of up to 1,800 roles. | | July 2026 | [monday.com restructuring](https://www.calcalistech.com/ctechnews/article/udx1nmbdq?ref=theraceai.org) | Watch / concentration-leaning | Narrows | Links 20% workforce reduction with fewer layers, autonomous teams, selective hiring, and human-agent workflows. | | July 2026 | [Visa](https://www.reuters.com/business/world-at-work/visa-plans-cut-7-workforce-bloomberg-reports-2026-07-28/?ref=theraceai.org), [ServiceNow](https://www.businessinsider.com/servicenow-cuts-jobs-amid-global-restructuring-and-ai-focus-2026-7?ref=theraceai.org), [Chime](https://www.reuters.com/business/world-at-work/chime-cut-10-total-workforce-or-about-140-jobs-source-says-2026-07-31/?ref=theraceai.org), and [Uber](https://finance.yahoo.com/technology/ai/articles/uber-cuts-10-customer-jobs-234053965.html?ref=theraceai.org) restructuring | Watch / pattern evidence | Narrows | Strengthens cross-company recomposition without resolving role-level causality. | | 2026-07-17 | [Meta preliminary-injunction ruling](https://www.reuters.com/business/world-at-work/us-judge-wont-block-meta-laying-off-workers-who-filed-ai-discrimination-lawsuit-2026-07-17/?ref=theraceai.org) | Dispersion / due-process signal | Preserves | Highlights evidence asymmetry and need for timely access to automated-decision evidence. | | 2026-07-13 | [UK Financial Services Skills Compact](https://www.gov.uk/government/publications/financial-services-skills-compact--2?ref=theraceai.org) | Dispersion +1, secondary lane | Preserves | Provides a limited transition-governance counterexample. | ### Interpretation The evidence does not support a simple mass-replacement claim. It does support the conclusion that AI is changing the architecture of firms: management layers, team size, implementation roles, decision rights, vendor boundaries, and output expectations. Authority can concentrate even where no eliminated role is directly matched to an AI task. ### What would change this read Move toward dispersion if workers gain meaningful voice, firms demonstrate paid training and real redeployment, AI expands entry routes and job quality, and small firms gain durable independence. Move further toward concentration if decision rights transfer to agents or centralized implementation cores without disclosure, appeal, bargaining, or transition support. ### How the Scoreboard Works URL: https://www.theraceai.org/scoreboard-methodology/ Last updated: 2026-09-08T19:25:31.000Z The Scoreboard tracks two related questions: - Is artificial intelligence concentrating power or dispersing meaningful capacity? - Are the choices shaping that outcome still open to challenge? The answers are rarely simple. A development can widen access while creating a new dependency. A regulation can strengthen accountability but also favor organizations large enough to comply. A major infrastructure project can provide useful capacity while placing control over essential resources in fewer hands. The Scoreboard is designed to preserve those tensions rather than force every development into a simple good-or-bad category. ## What Gets Scored The Scoreboard focuses on concrete developments that may change institutions, markets, public capacity, or the distribution of power. These can include: - Laws, regulations, court decisions, and enforcement actions. - Public infrastructure approvals, conditions, moratoria, and cost decisions. - Changes in access to advanced models or computing capacity. - Major public contracts and AI deployments. - Platform rules, standards, and changes in market architecture. - Significant workforce or organizational restructuring. - New public, civic, or worker capacity. - Changes that make important decisions more—or less—visible and open to challenge. An announcement alone is usually not enough. There should be a formal action, meaningful adoption, durable capacity, institutional consequence, or clear change in public contestability. That does not mean waiting until every long-term result is known. A law can be scored when it is enacted. A formal program can be scored when it is launched. A major contract or infrastructure decision can be scored when it is approved. The later effects may change the reading, but the institutional action itself matters. Civic action can also count. A protest, coalition, public campaign, or other form of mobilization may be scored when the exercise of civic capacity is itself concrete, novel, and consequential—not only after it produces a new law or policy. ## Direction of Power Each scored development receives a directional reading. | Reading | Meaning | | ------------------------- | -------------------------------------------------------------------------------------------------------------- | | **Strong concentration** | A durable and scalable shift toward control by fewer institutions or actors. | | **Leaning concentration** | A meaningful concentration signal, although its effects are not fully settled. | | **Mixed and contested** | The development contains important movement in both directions, or its direction remains genuinely uncertain. | | **Leaning dispersion** | A meaningful expansion of access, capacity, accountability, or public control, although it remains incomplete. | | **Strong dispersion** | A durable and scalable expansion of broadly shared capacity or effective public contestability. | Concentration does not simply mean that an organization is large. Scale may be necessary. The question is whether other institutions retain meaningful choices, alternatives, bargaining power, oversight, and the ability to leave. Dispersion does not mean that everyone receives equal access to everything. It means that useful capacity, authority, or accountability is becoming more broadly distributed—and that powerful systems remain open to challenge. ## The Window of Contestability The Scoreboard separately asks whether the opportunity to shape an emerging system remains open. | Window status | Meaning | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------ | | **Open** | Rules, institutions, standards, and practices remain meaningfully shapeable. | | **Narrowing** | Important choices are beginning to harden, but alternatives remain possible. | | **Closing** | Dominant arrangements are becoming difficult to reverse. | | **Mostly closed** | The relevant structure is largely locked in. | | **Reopened** | A policy, legal challenge, public intervention, technical alternative, or market change has created new room for contestation. | This distinction matters because a relatively modest decision can have large consequences if it locks in infrastructure, contracts, technical standards, or institutional expectations. Conversely, a moratorium, appeal, open standard, or new public alternative can matter because it creates time and restores choices. ## Transparency and Contestability Some developments affect not only who holds power, but whether the exercise of that power can be seen and challenged. The Scoreboard therefore also considers whether a development creates: - Public reporting or disclosure. - Clear standards and decision rules. - Independent evaluation. - Audit, appeal, or review rights. - Procurement visibility. - Public inventories or registries. - Meaningful outside participation. The opposite movement occurs when important decisions shift into proprietary, classified, discretionary, or otherwise opaque systems. Transparency is not sufficient by itself. Information matters most when someone has the authority and capacity to act on it. ## From Events to Lane Readings The Scoreboard follows nine connected lanes because the AI race is unfolding across several different systems at once. Individual developments are assigned to the lane where their principal effect occurs. They may also affect one or more secondary lanes. The monthly lane reading is not produced by simply adding up scores. A dozen small announcements should not outweigh one durable institutional change. Instead, the Scoreboard considers the strength, scale, durability, and interaction of the developments within each lane. The result is a current judgment: - Which way power appears to be moving. - How open the window remains. - How confident the evidence allows us to be. - What future developments could change the reading. ## A Current Reading, Not a Final Verdict The Scoreboard records judgments based on the evidence available at the time. Those judgments can change. A law may be enacted but weakly enforced. A promising public program may remain symbolic. An open standard may produce real interoperability or become another route into a dominant platform. A concentrated system may later become more contestable through regulation, litigation, public investment, technical change, or civic action. For that reason, the Scoreboard is updated as evidence accumulates. Earlier developments may be revised when implementation reveals more about their actual effects. The methodology may also continue to develop as the research deepens and the strengths and limits of the approach become clearer. The purpose is not false precision. It is disciplined public judgment about where power is moving, where meaningful alternatives remain, and where democratic attention is most urgently needed.. ### Series URL: https://www.theraceai.org/series/ Last updated: 2026-07-24T21:41:13.000Z Browse connected arguments in their intended reading order. ## Posts ### Data Centers Are Changing the Government Economic Development Bargain URL: https://www.theraceai.org/data-centers-are-changing-the-government-economic-development-bargain/ Last updated: 2026-09-07T21:52:03.000Z Something unusual happened in New Jersey in late August. On August 27, the [End Data Center Tax Credits Act](https://legiscan.com/NJ/bill/S4390/2026?ref=theraceai.org) became law as P.L. 2026, c. 77, after passing the Senate 35–4 and the Assembly 74–4 on June 30\. The state had previously reserved $500 million in tax credit capacity under the Next New Jersey Program for qualifying AI and data center projects. As of May, $250 million had already been awarded to one project. The act removed the remaining $250 million in uncommitted Next New Jersey credit authority while leaving that prior award untouched, according to the [New Jersey Legislature's bill statement](https://pub.njleg.state.nj.us/Bills/2026/A5500/5165%5FS1.HTM?ref=theraceai.org). Anyone familiar with government economic development programs will recognize how unusual this is. Why would a state that wants AI investment voluntarily withdraw an incentive intended to attract it? One answer is that private investment is no longer the only scarce asset in the transaction. Increasingly, the scarce asset is access to infrastructure, development rights, and speed to market. That package of resources and public approval is giving governments and public utilities more leverage than they have traditionally had. ## The old bargain: mobile capital, competing jurisdictions For decades, the government economic development game has rested on the mobility of investment capital. In effect, developers could say: "We can take this project elsewhere, along with the jobs and taxes it will bring to your constituents, so give us the best deal you can." Governments and public utilities responded with lower taxes, subsidized infrastructure, discounted public power, expedited approvals, and favorable development terms. Developers and their capital had more alternatives than the host governments. As one example, [Reinvent Albany estimates](https://reinventalbany.org/2026/08/ny-taxpayers-giving-at-least-321-million-to-rockland-county-data-centers/?ref=theraceai.org) that New York taxpayers have committed at least $321 million in sales tax exemptions and low-cost public power subsidies to five data center sites encompassing 11 data centers in Rockland County. Long-duration sales and use tax exemptions can generate substantial savings for developers and substantial foregone revenue for governments. That bargain made sense when many jurisdictions offered roughly equivalent power, land, water, and permitting conditions. Developers could credibly move elsewhere, and governments competed to capture the projects, jobs, and tax base. ## What changed: AI-scale demand is making the infrastructure bundle scarce AI-scale demand is changing the scarcity equation. Data centers have been built for decades, but AI-driven growth in their scale and speed is changing the bargain for large data centers generally. Investment capital matters, but so does access to a bundle of infrastructure, development rights, and speed to market. Governments and public utilities increasingly control what developers most need: - hundreds or even thousands of megawatts of power - large contiguous sites - water or alternative cooling systems - fiber - permitting and environmental review capacity - political acceptance - perhaps most importantly, speed to market and to available power and water Getting that entire package in one place, on the timetable a developer needs, can matter more than a large tax exemption. If a site will take several years to assemble and approve before construction can even start, a tax incentive may not make it attractive to a developer that needs the facility operating within three years. ## What the new bargain is looking like Governments and public utilities are starting to test their new leverage in several ways. ### Negotiating different terms for individual deals Governments and public utilities are moving beyond a simple standard that a project with enormous electricity demand should not make existing customers worse off. In some cases, the emerging standard is stronger: the project should contribute enough system value to make existing customers better off. Developers are still getting access to the grid, but on different terms. On August 21, the [Kentucky Public Service Commission approved](https://psc.ky.gov/Case/ViewCaseFilings/2026-00115/Register?ref=theraceai.org) the Big Rivers / Kenergy / TeraWulf agreement for a planned 482 MW Hawesville data center on the former Century Aluminum smelter site. The [agreement](https://psc.ky.gov/pscscf/2026%20cases/2026-00115/20260414%5FBig%20Rivers%20Electric%20Corporation%20Contract%20Filing.pdf?ref=theraceai.org) includes unusual customer obligations: - assumption of the market costs and risks associated with serving the load - payment for reserved transmission capacity - credit support and prepayment protections - curtailment before ordinary non-interruptible customers during specified severe system emergencies Two features stand out. First, the project bears the costs and identified financial risks associated with serving its load rather than shifting them to other customers. Second, its power can be reduced before ordinary non-interruptible customers face reductions during specified severe emergencies. In Georgia, OpenAI struck an agreement with Georgia Power involving 3,200 MW of new electric demand in Effingham County. [OpenAI agreed](https://www.georgiapower.com/news-hub/press-releases/georgia-power-to-serve-openai-project-in-effingham-county.html?ref=theraceai.org) to pay the full infrastructure costs required to serve the facility and to make up to 1,000 MW of its load flexible. That means its electricity use could be reduced during periods of high system demand. Georgia Power [projects](https://www.georgiapower.com/news-hub/press-releases/contract-openai-approved-part-portfolio-delivering-950-million-annual-savings.html?ref=theraceai.org) that its broader portfolio of large-load customers, including the OpenAI agreement, will produce approximately $950 million per year in customer savings beginning in 2029—equivalent to at least $15 per month for a typical residential customer using 1,000 kWh per month. Those are projected, not realized, savings. But the claim itself points to a different bargain: the goal is moving from protecting ordinary customers from harm toward making them beneficiaries of the new load. ### Creating new rules and categories, not just new terms for specific deals The more important shift comes when tougher deal terms become standing rules. Data centers have often been treated as ordinary industrial customers. Now some governments and regulators are creating distinct large-load categories with different rates and obligations. We are seeing: - new tariffs - statutes - eligibility conditions - standard cost-allocation rules - differentiated levels of electric-service reliability for very large loads In Delaware, legislation signed on August 26 created a separate rate class for large energy-use facilities and required costs attributable to those facilities to be assigned to them where possible, with other non-directly attributable costs kept within the large-load class rather than shifted to other customer classes. The package also establishes interruptibility requirements. A companion law requires qualifying large energy users to build or procure new generation, with facilities using the 10-year compliance pathway ramping to 100% of annual energy needs within 10 years. The state summarized the package when [Governor Matt Meyer signed the legislation](https://news.delaware.gov/2026/08/26/governor-matt-meyer-signs-energy-consumer-protection-legislation/?ref=theraceai.org). In Michigan, the [Michigan Public Service Commission](https://www.michigan.gov/mpsc/commission/news-releases/2026/08/27/mpsc-approves-siting-settlement-on-ingham-county-solar-project?ref=theraceai.org) approved amendments to Indiana Michigan Power Company's large-load tariff for customers drawing 50 MW or more. Requirements include: - 15-year contracts - a maximum five-year ramp-up period - stronger collateral for new facilities - early-termination protections - four years' notice for certain major load reductions - a 90% monthly minimum billing demand - a 65% monthly minimum energy requirement - customer-specific Commission filings before service begins The Commission also directed Indiana Michigan Power to propose a standalone large-load rate in its next general rate case. In a separate action the same day, the Commission approved DTE Electric emergency procedures under which specified very large loads can be shed before other customers when doing so would prevent broader firm-load shedding. The important development is not merely that governments are negotiating harder. Some are beginning to convert the terms they can demand into rules that apply prospectively to an entire class of projects. ### Making broader public permission part of the bargain Electric power is not the only thing developers need that they cannot simply buy on the open market. They also need permission to build. That gives governments another source of leverage and can make approval more discretionary, negotiated, public, and contestable. On August 5, Tucson adopted [new regulations for large-scale data centers](https://www.tucsonaz.gov/Departments/Planning-Development-Services/Planning-Initiatives/Data-Centers-Unified-Development-Code-Amendment?ref=theraceai.org). The city now requires a public review process including: 1. neighborhood coordination and review of the development plan 2. a public hearing 3. zoning examiner review 4. final mayor and council consideration Each project must address: - water adequacy - power adequacy - environmental impacts - noise - setbacks - mitigation requirements The process forces a broader question: Does this project fit the community's scarce resources and public interest requirements well enough to receive development rights? Tucson has made approval of large data center projects more project-specific, public, and contestable. In Frederick County, Maryland, County Executive Jessica Fitzwater announced a proposed agreement with Catellus, master developer of the Frederick Digital Campus, that would provide about $110 million in developer-funded community benefits. Catellus’s application for a Development Rights and Responsibilities Agreement is now moving through a public review process. The [county announcement](https://www.frederickcountymd.gov/m/newsflash/Home/Detail/5873?ref=theraceai.org) identifies benefits including: - $30 million for school renovation - $40 million for a community center and recreational space - $14.5 million for workforce development and career and technical education, plus additional investments Frederick County shows land-use and development authority being used to negotiate a broader public return rather than simply processing a conventional development application. ### Making subsidies more conditional, measurable, and contestable For years, governments and public utilities have offered economic development subsidies with relatively few strings attached. But the examples above show developers paying more of their infrastructure costs, accepting curtailment, providing financial guarantees, facing special regulatory treatment, and negotiating for development rights. That makes the old assumption that governments must also subsidize them increasingly questionable. Where subsidies remain available for large data centers, they increasingly appear to be taking on features such as: - shorter duration - higher qualification thresholds - auditable performance - disclosure of resource use - local fiscal participation Storey County, Nevada, has submitted a legislative proposal for consideration in the 2027 session that would substantially revise the state's data center tax abatement program. According to [The Nevada Independent's review of the proposal](https://thenevadaindependent.com/article/storey-county-is-a-hub-for-nevada-data-centers-it-wants-to-rein-in-state-tax-breaks?ref=theraceai.org), it would: - shorten current 10- or 20-year abatements to 5 or 10 years - double investment requirements and slightly increase permanent-job requirements - require audits at the five- and 10-year marks - disclose projected and actual electricity and water use - disclose local government revenue forgone because of the incentives - bar tax breaks for projects located on federal land The scale matters. Nevada has approved about $461 million in expected data center tax breaks, with roughly three-quarters tied to developments in Storey County. New Jersey goes further. Storey County is proposing tighter conditions and accountability for abatements; New Jersey has actually withdrawn part of an available incentive. In effect, New Jersey is no longer simply asking, "What do we need to give away to get this project here?" Nor is New Jersey alone in asking the question. Kentucky Governor Andy Beshear said on August 27 that data centers do not need incentives and that he would support rolling back the state's sales and use tax exemptions, according to [Kentucky Public Radio reporting](https://www.weku.org/the-commonwealth/2026-08-27/kentucky-gov-andy-beshear-calls-for-ending-state-tax-incentives-for-data-centers?ref=theraceai.org). In Ohio, [HB 999](https://www.legislature.ohio.gov/legislation/136/hb999?ref=theraceai.org) has been introduced to prohibit property-tax exemptions for data centers. Neither change has been enacted, but together with New Jersey's action they suggest that the subsidy side of the bargain is also being reconsidered. ## How far can we take the argument? I do not want to overstate the case. These examples are recent, not yet widespread, and the pattern is still emerging. - The shift will not be universal. Some jurisdictions will still have abundant infrastructure and a strong political desire to attract projects. Developers still retain siting options. - Scarcity can change. Major investments in electricity supply, including wind and solar generation, could expand the number of viable sites. - Governments can still bargain badly. They can underestimate infrastructure and service costs, accept opaque agreements, or grant long-term incentives before actual costs are known. ## The trend is worth watching The momentum behind mega-data-center development is shifting leverage between developers and governments and public utilities. Demand is enormous, but these projects depend on infrastructure and public authority that are neither abundant nor quickly substitutable. That scarcity is changing the development bargain. Across the cases above, the emerging bargain can be summarized this way: - Bring, build, procure, or finance the capacity you require. - Pay the infrastructure and system costs your project creates. - Accept different reliability when the system is stressed. - Show that the project fits local resource and community constraints. - Protect the public against stranded costs and changing plans. - Make the public return measurable and enforceable. - And if you still want a subsidy, explain why the public needs to pay for an investment already competing for scarce infrastructure and development rights. The pattern is becoming clear enough to watch closely. The question is whether governments and public utilities can convert this moment of leverage into durable, transparent, and contestable rules and public benefits before scarcity eases or the AI boom ends. ### How Municipal Governments Can Get What They Need from AI URL: https://www.theraceai.org/how-municipal-governments-can-get-what-they-need-from-ai/ Last updated: 2026-08-24T22:43:04.000Z There are roughly [19,500 municipal governments and 3,000 county governments](https://www.census.gov/library/publications/2026/econ/govtorg2225.html?ref=theraceai.org) in the United States. Most are nothing like New York, Chicago, or Los Angeles County. [Three-quarters of incorporated places have fewer than 5,000 residents](https://www.census.gov/newsroom/press-releases/2025/vintage-2024-popest.html?ref=theraceai.org), and their IT capabilities range from very good to just getting by. When working with AI vendors—both product providers and implementers—most local governments face a significant disadvantage in size and information. Fortunately, they also have access to a well-established and remarkably dense support network. There is the International City/County Management Association (ICMA), the Government Finance Officers Association (GFOA), the National League of Cities (NLC), the National Association of Counties (NACo), and similar associations for attorneys, procurement directors, IT directors, HR directors, police chiefs, fire chiefs, parks directors, and many others. In addition to these national organizations, [nearly every state has a municipal league](https://www.nlc.org/membership/state-municipal-leagues/?ref=theraceai.org), such as the Colorado Municipal League, and many metropolitan areas have regional networks, such as the [Association of Bay Area Governments](https://abag.ca.gov/about-abag?ref=theraceai.org) around San Francisco. These organizations regularly compare notes, share what is working and what is not, and establish common positions, particularly on state and federal legislation. This network has long been very good at sharing information, spreading new practices, and establishing common positions. I have personally participated in and benefited from it. Historically, this “downward channel”—moving knowledge and expertise through the network—is what it does best. AI creates a new challenge. The network must move information much faster, but it also needs a stronger channel in the other direction: collecting what local governments are learning and turning that experience into a more organized relationship with AI vendors. Municipal governments do not need to consolidate authority or give up autonomy to gain scale. They can create scale through coordination, using capabilities and relationships they already have. ## A New AI Support Layer Is Already Emerging A new support layer focused specifically on AI is already forming. The City of San José started what became the [GovAI Coalition](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/ai-reviews-algorithm-register/govai-coalition?ref=theraceai.org) after it struggled to get satisfactory answers from AI vendors about privacy and data use. When San José staff spoke with neighboring cities, they discovered others were having the same experience. The city called governments together to compare notes and approach vendors with a joint message. That first meeting in 2023 attracted about 50 agencies. Three years later, GovAI includes [more than 900 agencies](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/artificial-intelligence-inventory/govai-coalition/become-a-member/become-a-member?ref=theraceai.org). GovAI has since developed vendor fact sheets, disclosure requirements, procurement practices that translate general AI principles into operational questions, risk-review models, and internal guidance. It also provides [shared resources that governments can reuse](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/artificial-intelligence-inventory/govai-coalition/templates-resources?ref=theraceai.org) rather than develop independently. That is particularly important for smaller governments. GovAI can provide pieces of common AI infrastructure—policies and governance tools, procurement materials, model contract language, shared registries, and an AI Contract Hub—that individual governments can adapt to their own circumstances. GovAI itself is also becoming more institutionalized. In April 2026, the San José City Council [approved transitioning the coalition toward an independent nonprofit organization](https://sanjose.legistar.com/View.ashx?G=920296E4-80BE-4CA2-A78F-32C5EFCF78AF&GUID=7119A3C1-9CEE-4455-A511-0A91F1271D10&ID=15340312&M=F&ref=theraceai.org), giving it access to a formal governance structure, philanthropic funding, dedicated leadership, and greater capacity to support its membership. Most important for the argument here, GovAI has established an [Industry Relations Committee](https://www.sanjoseca.gov/your-government/departments-offices/information-technology/artificial-intelligence-inventory/govai-coalition/committees-working-groups?ref=theraceai.org). Its work includes supporting engagement with vendors and helping public agencies identify AI solutions that meet their needs and values. Those are important beginnings. The opportunity now is to connect this emerging AI-specific capacity much more deliberately to the broader network of organizations that already supports local governments across the country. ## The Existing Network Needs to Move Faster The traditional means of spreading knowledge through the municipal support network include regular meetings, conferences, reports, professional publications, and informal peer exchange. All will remain useful. But AI vendors can change their models, interfaces, terms, and capabilities far faster than this network traditionally moves. For AI, rapid diffusion therefore needs to become a core operating function. Useful practices, warnings, procurement lessons, and implementation experience cannot take a year to make their way through conferences, publications, and individual associations. They need to move quickly across professional associations, state leagues, regional organizations, and specialized AI networks such as GovAI. ## The More Important Opportunity Runs in the Other Direction The more significant change is in the other direction. Local governments across the country are conducting many real-world experiments with AI. They are learning where systems fail, what vendors will disclose, which contractual provisions matter, and what implementation problems recur. Municipal support networks should systematically collect that experience. Their role should be to identify recurring problems and distinguish isolated incidents from patterns appearing across jurisdictions. Those patterns can then become common requirements presented to AI vendors. Governments and vendors could work toward clearer approaches to auditability, data use, portability, continuity, provider replacement, and other recurring concerns. A vendor hearing the same requirement from hundreds of governments is more likely to take it seriously than one hearing isolated requests one customer at a time. The goal should therefore be regular and sustained engagement with AI vendors rather than episodic contact. Leaders from organizations such as NLC, NACo, ICMA, GFOA, associations representing municipal technology officials, and GovAI could establish a working group that aggregates concerns and priorities across their networks. Such a group would not eliminate the scale advantage of major AI companies, but it could materially reduce the disadvantage facing individual governments. It could aggregate the experience and purchasing concerns of thousands of municipalities and counties while leaving each government free to decide how AI should be used in its own organization and community. Vendors could benefit as well. Instead of separately gathering feedback from hundreds of governments, they could receive clearer signals about recurring market needs, identify developing problems earlier, and reduce the burden of hundreds of one-off negotiations. More standardized expectations could make the public-sector market easier for both sides to navigate. The working group could meet regularly with major AI product companies and implementers. Those meetings should surface recurring problems, requested capabilities, contract concerns, and emerging expectations. Later meetings should report what vendors have changed, what they have chosen not to change, and why. That would turn industry relations from occasional consultation into an actual feedback loop. This broader model is not yet proven. GovAI has established pieces of it, and [municipal organizations have long used collective action effectively in legislative settings](https://www.nlc.org/advocacy/federal-advocacy-committees/?ref=theraceai.org). The next step is to determine whether those capabilities can also give local governments greater influence over the AI products and services they increasingly rely upon. Without such a mechanism, smaller governments risk becoming users of systems they have little ability to shape. ## Matching Scale to Scale The easiest outcome to imagine is continued fragmentation: AI vendors deal with local governments one at a time, effectively saying, “Here is what we provide; you need to adapt to it.” Local governments have another option. They can use the networks they already possess to reduce that scale disadvantage without centralizing authority. The necessary institutions largely exist. What needs to change is how deliberately they connect, how quickly information moves through them, and whether they can turn dispersed local experience into an organized voice in the market. I am suggesting three steps: 1. *Connect the support network* — NLC, NACo, ICMA, GFOA, state leagues, regional councils, functional associations, GovAI, and others should deliberately exchange AI knowledge across organizational boundaries rather than principally through their own membership channels. 2. *Build the return channel* — Create routine mechanisms for governments to report what they are learning—vendor problems, implementation failures, useful practices, contract issues, and missing capabilities—and aggregate those experiences into recognizable patterns. 3. *Institutionalize the vendor loop* — Carry those patterns into regular engagement with providers and implementers, seek responses, and report back what changed. The most consequential municipal AI institution may therefore not be another AI office. It may be the connective tissue that allows hundreds of governments to learn, and increasingly to bargain, as though they were much larger than any of them is alone. ### This Week: Action, But In Different Directions URL: https://www.theraceai.org/this-week-action-but-in-different-directions/ Last updated: 2026-08-09T18:30:13.000Z The most important pattern this week is the simultaneous movement in opposite directions. Government is becoming a more consequential gatekeeper over frontier-model release and hyperscale infrastructure, while open weights and on-device models are making technical alternatives more plausible—and the portability question is becoming more concrete. - [**White House frontier-model review framework**](https://www.reuters.com/legal/litigation/meta-anthropic-google-openai-meet-with-trump-white-house-amid-rogue-ai-agent-2026-08-04/?ref=theraceai.org) — The administration established a voluntary prerelease review channel for advanced proprietary models, while excluding open-weight models from the process. - [**Texas orders comprehensive data-center audit**](https://gov.texas.gov/news/post/governor-abbott-directs-comprehensive-data-center-audit?ref=theraceai.org) — Texas paused further movement of data-center projects through the ERCOT interconnection process pending a broader review of grid, water, incentives, ownership, and community impacts. - [**Liquid AI releases an agentic model that runs on consumer devices**](https://www.liquid.ai/blog/lfm2-5-2-6b?ref=theraceai.org) — LFM2.5-2.6B can be downloaded and run locally on laptops and phones, including agentic workloads without a cloud API. - [**European Commission requires greater Android interoperability for rival AI assistants**](https://digital-strategy.ec.europa.eu/en/news/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data?ref=theraceai.org) — The July decision remains the clearest current regulatory test of whether competing assistants can gain meaningful access to a dominant platform’s operating layer. The Race question is therefore moving beyond **“Are alternatives available?”** toward **“Can actors actually exercise choice, exit, and challenge the institutions controlling access?”** These events fit right into the context of two topics we've been working on, resulting in this week's feature articles. First is the apparently-developing concept of 'federally-gated/privately-owned' AI infrastructure. I think this is a new operating model worth watching as it develops - my initial reaction is this is not positive toward democratic governance. The second article discusses the importance of 'portability' as AI develops and spreads throughout society. This ability will be key to the concept of being able to exercise choice and exit, both for individuals and organizations using AI. ### [Private Infrastructure, Public Permission](https://www.theraceai.org/private-infrastructure-public-permission/) The federal government is becoming the gatekeeper for a privately owned AI infrastructure system. That combination concentrates power—and creates structural incentives for corruption. ### [The Next AI Fight Is Over the Right to Leave](https://www.theraceai.org/the-next-ai-fight-is-over-the-right-to-leave/) As AI systems learn our preferences, projects, and workflows, portability will determine whether model choice becomes customer power. ### The Next AI Fight Is Over the Right to Leave URL: https://www.theraceai.org/the-next-ai-fight-is-over-the-right-to-leave/ Last updated: 2026-08-07T20:13:21.000Z ### A real choice is coming into view A meaningful choice among AI models is beginning to emerge. Open-weight models are becoming more numerous, more capable and more varied. Chinese developers are releasing large models built for coding, reasoning, multimodal work and agentic tasks, including [Moonshot AI’s Kimi K3](https://github.com/MoonshotAI/Kimi-K3/blob/main/README.md?ref=theraceai.org) and [DeepSeek V4](https://api-docs.deepseek.com/news/news260424/?ref=theraceai.org). American firms are producing both [large general-purpose open-weight systems](https://help.openai.com/en/articles/11870455-openai-open-weight-models-gpt-oss?ref=theraceai.org) and [much smaller models designed to run locally](https://www.liquid.ai/blog/introducing-lfm2-5-the-next-generation-of-on-device-ai?ref=theraceai.org). Europe’s [Mistral](https://docs.mistral.ai/models/?ref=theraceai.org) and other developers continue to expand the open-weight field. These models do not all match the strongest proprietary systems across every task. They do not need to. The relevant development is that credible alternatives are multiplying. Some open-weight models are already highly effective for many practical uses. Others [approach frontier performance in selected areas](https://www.nature.com/articles/d41586-026-02281-2?ref=theraceai.org). They can be [customized and self-hosted](https://help.openai.com/en/articles/11870455-openai-open-weight-models-gpt-oss?ref=theraceai.org) or, in smaller forms, operated on personal and organizational hardware. For the first time, it is becoming plausible that many customers could have meaningful choice among a broad range of capable AI models rather than among a small number of dominant providers. That is an opening for dispersion. But it will matter only if customers can use it. ## The model may be replaceable before the relationship is An open-weight model can make the intelligence technically replaceable. A developer can [download and modify it](https://help.mistral.ai/en/articles/347393-under-which-license-are-mistral-s-open-models-available?ref=theraceai.org) or [deploy it through different hosting environments](https://docs.mistral.ai/models/deployment?ref=theraceai.org). Compatibility layers can also make substitution easier: [DeepSeek V4, for example, supports both OpenAI and Anthropic API formats](https://api-docs.deepseek.com/news/news260424/?ref=theraceai.org). Most people, however, are not simply using a model. They are developing a relationship with a larger agentic system. The AI agent an end-user works with begins to know how the user writes. It learns recurring preferences. It retains past conversations, files and project history. It connects to email, calendars, documents and other applications. It may eventually manage recurring tasks, maintain workflows and act under standing permissions. The model produces the agent's intelligence. The surrounding system accumulates the customer. That distinction will become more important as AI becomes more useful. A person may have access to five capable alternative models and still be unable to leave the provider that holds years of accumulated context. A business may be able to purchase intelligence from another source but be unwilling to rebuild its integrations, permissions, evaluations and institutional knowledge. A market can therefore contain many models while remaining highly concentrated at the customer level. Open weights make the **model** more replaceable. Portability would make the **provider** more replaceable. ## We have solved a simpler version of this problem before The early cellphone market offers a useful comparison. Customers could change carriers, but changing carriers generally meant losing the telephone number through which family, friends, employers, clients and institutions knew how to reach them. The customer was formally free to leave. The practical cost of leaving was high. Congress created a [number-portability obligation](https://uscode.house.gov/view.xhtml?edition=prelim&f=treesort&jumpTo=true&num=0&req=%28title%3A47+section%3A251+edition%3Aprelim%29+OR+%28granuleid%3AUSC-prelim-title47-section251%29&ref=theraceai.org) in the [Telecommunications Act of 1996](https://www.govinfo.gov/app/details/PLAW-104publ104?ref=theraceai.org). The Federal Communications Commission then developed the technical and operating rules, with [wireless number portability reaching the 100 largest metropolitan statistical areas on November 24, 2003](https://docs.fcc.gov/public/attachments/DOC-240702A1.pdf?ref=theraceai.org). Number portability did not eliminate every switching cost. But it removed one of the incumbent carrier’s most powerful advantages: control over something the customer could not realistically leave behind. AI presents the same problem at much greater complexity. The equivalent of the telephone number is not one identifier. It is the accumulated working relationship between the customer and the system. The principle remains the same. Customers should be able to change AI providers without surrendering everything that made the previous service useful. ## What would have to move? We often use the phrase **data portability**. That is too narrow for AI. Practical AI portability has at least three levels. **Data portability** means that customers can retrieve their conversations, files and account information in a usable format. **Context portability** means that a replacement assistant can reconstruct what the former system learned about the customer: preferences, continuing projects, important relationships, working style and relevant history. **Operational portability** goes further. It allows the customer to restore the workflows, connected applications, permissions, recurring tasks, agent configurations and evaluation history that made the former system function. The first level is already available from major providers. The second is beginning to appear. The third remains much less developed. That third level will become decisive. Imagine a small company that has spent three years building an AI system into its operations. The system knows which employees may approve particular actions. It understands the company’s document structure. It has access to selected databases. It follows established review procedures. It has been corrected repeatedly and tested against the company’s standards. A file containing old conversations would not allow another provider to recreate that system. The customer would possess the data but still lose the capacity to switch. ## A partial switch is already possible The technology required for portability is not purely hypothetical. Google now allows [eligible personal Gemini users to import both chat histories and a representation of memory from other AI platforms](https://support.google.com/gemini/answer/16868299?hl=en&ref=theraceai.org). The documented process for ChatGPT begins with [OpenAI’s data export](https://help.openai.com/en/articles/7260999-how-do-i-export-my-chatgpt-history-and-data?ref=theraceai.org), whose ZIP file includes chat history and other relevant account data. Gemini also supplies a prompt that asks the former assistant to summarize the user’s preferences, remembered facts and general context. That summary can then be added to Gemini’s own memory system. That is meaningful progress. A customer moving from ChatGPT to Gemini would not necessarily arrive as a complete stranger. Past conversations could be retained. Important personal and professional context could be summarized and reconstructed. But this is still a translation, not a transfer of the functioning relationship. The new system receives the former system’s description of what it learned. It does not inherit the exact way the former system weighted that information, connected it across conversations or learned from repeated corrections. Google’s documented import process covers chats and memory. It does not provide a process for migrating connected applications, standing permissions, scheduled tasks, project structures or custom workflows. Under the documented migration process, those elements would still need to be reconnected or rebuilt. The result proves two things at once. A significant degree of context portability is technically possible today. The documented tools do not yet provide complete operational portability. ## Access is not the same as exit Regulation is also beginning to address pieces of the problem. On [July 16, 2026, the European Commission adopted binding specification measures](https://digital-strategy.ec.europa.eu/en/news/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data?ref=theraceai.org) requiring Google to provide competing AI services effective interoperability with [11 Android features relevant to AI services](https://digital-markets-act.ec.europa.eu/developer-portal/interoperability/alphabet-specification-proceedings-interoperability-ai-services%5Fen?ref=theraceai.org). The objective is to prevent Google’s own AI services, such as Gemini, from receiving operating-system access that competing assistants cannot match. This is an important intervention. It could allow another assistant to reach the customer and perform many of the same functions as Google’s own assistant. That addresses entry. It does not necessarily address exit. A rival assistant might receive equivalent technical access to Android while the customer’s accumulated memory, history and workflows remain embedded in Gemini. The competitor could enter the platform without being able to bring the customer along. That is the gap current policy has not yet closed. Interoperability allows another provider to connect. Data portability allows information to leave. Operational portability exists only when the customer can leave without destroying the capacity already built. ## The standard should be practical, not symbolic A meaningful portability standard should be judged by what happens after the customer presses “export.” Can another system understand the material? Can it reconnect applications, permissions and recurring work? Can the customer resume useful operations without months of rebuilding? An archive that no competing system can interpret is not effective portability. Nor is a thousand-page transcript that requires the customer to identify manually which facts, decisions and preferences still matter. The standard should require structured, machine-readable exports of the customer-controlled parts of the system. That would include memory, instructions, project relationships, workflow definitions, tool connections, permission structures and relevant evaluation records. It should also require a tested migration process. Providers should be able to demonstrate that another compliant system can use the exported material to restore a meaningful portion of the customer’s prior capacity. This would not require every AI model to behave identically. Models will reason differently. Providers will offer different tools. Some proprietary capabilities will not transfer. The objective is not perfect replication. It is a credible exit. ## Build portability before lock-in hardens Portability will be easier to establish now than later. Memory systems, agent frameworks and tool connections are still developing. No single architecture has yet clearly become the unavoidable default. Providers are already experimenting with [imports](https://support.google.com/gemini/answer/16868299?hl=en&ref=theraceai.org), [compatible API formats](https://api-docs.deepseek.com/news/news260424/?ref=theraceai.org) and [open protocols for connecting AI systems to tools and data](https://docs.anthropic.com/en/docs/mcp?ref=theraceai.org). The technical standard should emerge through an open process with vendor participation, but no single provider should control it. Whether procurement, market pressure or regulation ultimately enforces that standard can be left for a separate discussion. The immediate task is simpler: define practical portability before millions of people and organizations build years of irreplaceable context inside incompatible systems. Standards adopted after lock-in forms become fights over how much incumbent advantage must be surrendered. Standards adopted before lock-in forms can shape the market itself. ## Portability turns technical choice into customer power Open weights, miniaturization and common interfaces can create alternatives. But alternatives do not create customer power if accumulated context cannot move. The layer that controls the customer’s accumulated context may become more powerful than the company that built the original model. The customer may be surrounded by alternatives and still remain unable to move. Portability is therefore not a minor consumer convenience. It is part of the infrastructure of a competitive AI market. It would allow people to adopt AI without treating their first choice as permanent. It would allow organizations to move when a better, safer or less expensive model appears. It would give new providers a chance to compete for existing customers rather than only for people who have not yet built anything. Open models can expand the supply of intelligence. Portability will determine whether that expansion produces freedom. Without it, we may create a market with many models—and millions of customers who still cannot leave. ### Private Infrastructure, Public Permission URL: https://www.theraceai.org/private-infrastructure-public-permission/ Last updated: 2026-08-04T19:43:10.000Z The federal government has [selected Brookfield Asset Management and NextEra Energy to develop a massive AI data-center and energy campus at the former Paducah Gaseous Diffusion Plant in Kentucky](https://www.energy.gov/articles/energy-department-announces-partnership-expand-reliable-affordable-energy-access-and-power?ref=theraceai.org). The proposed campus would include [more than 1.2 gigawatts of computing capacity, large amounts of gas generation and battery storage, and a network of public, cooperative and private utilities](https://www.investor.nexteraenergy.com/news-and-events/news-releases/2026/07-29-2026-133109921?ref=theraceai.org). The federal government owns the land. [The former enrichment complex already offers transmission capacity, water infrastructure, fiber connectivity, roads and land that can accelerate development](https://www.investor.nexteraenergy.com/news-and-events/news-releases/2026/07-29-2026-133109921?ref=theraceai.org). The Department of Energy selected the developers and will negotiate the terms under which the site is converted from a Cold War uranium-enrichment complex into privately controlled AI infrastructure. [Most of the new construction is supposed to be privately financed](https://www.energy.gov/em/articles/us-energy-department-seeks-proposals-ai-data-centers-energy-projects-paducah-site?ref=theraceai.org). But this is not an ordinary private development. The federal government is supplying the site, choosing the private partners and helping assemble the institutional environment in which the project can proceed. The compute, power and commercial value will largely be privately owned. Access to the underlying opportunity depends on public permission. That arrangement may become one of the most important concentration mechanisms emerging in the AI race. ## Paducah has company Two other former federal nuclear sites show that Paducah is part of an emerging broader development model. At the former Portsmouth Gaseous Diffusion Plant in Ohio, [the Department of Energy, SoftBank, SB Energy and AEP Ohio have announced a project that could eventually reach 10 gigawatts of data-center capacity, supported by as much as 10 gigawatts of new generation and approximately $4.2 billion in transmission infrastructure](https://www.energy.gov/articles/fact-sheet-department-energy-ensuring-affordable-energy-access-ohio-while-powering-future?ref=theraceai.org). [The project also includes $33.3 billion in Japanese funding under a government-directed bilateral investment agreement](https://www.energy.gov/articles/fact-sheet-department-energy-ensuring-affordable-energy-access-ohio-while-powering-future?ref=theraceai.org), although the overall division among public finance, guarantees and private capital has not been disclosed. At the Savannah River Site in South Carolina, [the National Nuclear Security Administration has selected Amentum to negotiate a phased lease for a 1-gigawatt AI data center and approximately 2 gigawatts of on-site generation](https://www.energy.gov/nnsa/articles/nnsa-selects-amentum-ai-data-center-and-energy-project-savannah-river-site?ref=theraceai.org). [Natural gas is expected to provide the initial power, with advanced nuclear generation named as a power source in a possible later stage](https://www.energy.gov/nnsa/articles/nnsa-selects-amentum-ai-data-center-and-energy-project-savannah-river-site?ref=theraceai.org). The project's overall lease, financing, customers, generation partners and final public obligations remain under negotiation. The projects differ in scale, financing and maturity. Their institutional structure is recognizably similar. [Former national-security sites offer federal land, roads, water systems, transmission access, secure locations, extensive site knowledge and workforces accustomed to operating large industrial facilities](https://www.energy.gov/sites/default/files/2025-04/RFI%20to%20Inform%20Public%20Bids%20to%20Construct%20AI%20Infrastructure%20%28website%20copy%29%20-%202025.04.10.pdf?ref=theraceai.org). Private companies are expected to finance, own and operate most of the new compute and energy assets. Federal agencies choose the developers, negotiate the leases, coordinate the reviews and present the projects as part of [a national strategy for AI capacity](https://www.energy.gov/powering-americas-ai-future-data-center-resource-hub?ref=theraceai.org). In the public record reviewed for this article, substantial participation in the specific AI projects has come only *after* the federal government selected the site, the private partner and the basic development direction. Later proceedings may (and likely should) address utility rates, environmental permits, site cleanup, road maintenance or other local services. They do not necessarily reopen the foundational decisions about who should control the site, what scale is appropriate or what public rights should attach to the resulting infrastructure. ## The federal government controls more than the land The federal sites establish one gate: access to scarce public land and inherited infrastructure. Two other recent federal actions establish gates around the power systems and equipment needed to operate large AI campuses. 1. The Environmental Protection Agency has [issued nonbinding guidance clarifying that power plants serving only physically islanded private consumers generally fall outside the federal Acid Rain Program](https://www.epa.gov/system/files/documents/2026-07/epa-issues-clarification-on-islanded-power-generators-and-acid-rain-program-provisions.pdf?ref=theraceai.org). A data center with a completely separate power system may therefore operate outside that program’s standardized emissions-allowance and monitoring requirements. 2. The Federal Communications Commission has [added new foreign-produced power inverters and advanced robotic devices to its national-security Covered List](https://docs.fcc.gov/public/attachments/DA-26-786A1.pdf?ref=theraceai.org). Covered equipment generally cannot receive the authorization needed to enter the U.S. market unless the Department of War grants Conditional Approval—or, for power inverters, the Department of Homeland Security does so. The FCC published application guidance describing what producers must submit, but approval ultimately turns on whether DoW or DHS determines that a device does not pose an unacceptable national-security risk. Together, these actions extend federal gatekeeping across three critical layers of AI infrastructure: - access to federal land and inherited public assets; - access to a favorable regulatory classification for dedicated power; - and access to strategically important equipment suppliers. The three federal-site projects have not been shown to rely on the EPA interpretation or the FCC approval process. But the governing logic is converging. Federal agencies increasingly control access to the sites, regulatory pathways and suppliers needed to build privately owned AI infrastructure. ## Two concentrations are forming at once The most obvious concentration is private. Only a small number of technology companies, infrastructure funds, energy developers and utilities can assemble the capital, equipment, power, security and specialized expertise required for a gigawatt-scale AI campus. The companies capable of building these projects will gain control over extraordinarily scarce compute and energy capacity. A second concentration is occurring inside government. Federal agencies are acquiring greater authority over site access, developer selection, lease conditions, regulatory classification, equipment eligibility and security approvals. Even when ownership remains private, entry into the market may increasingly depend on federal decisions. The two concentrations reinforce one another. Large companies are better able to manage confidential negotiations, satisfy security requirements, absorb approval delays and maintain relationships across several agencies. Federal officials, meanwhile, become dependent on the limited number of firms capable of delivering infrastructure at the scale and speed the government wants. This is neither a competitive private market operating at arm’s length from government nor a publicly owned infrastructure system accountable through ordinary public institutions. It is a federally gated private market: government controls entry, while a narrow group of companies captures much of the resulting economic and operational power. ## The structure creates incentives for corruption There is no public evidence that the Paducah, Portsmouth or Savannah River decisions were purchased, or that any official involved has acted corruptly. But the structure creates two clear corruption risks. ### Buying the gatekeeper Federal officials and other decision-makers may influence who receives a valuable public site, which lease conditions apply, whether a supplier receives Conditional Approval and how quickly a project advances. As the value of those decisions rises, so does the incentive to improperly reward the people who make them. The benefit could take the form of direct cash, future employment, consulting work, investment opportunities, political support or payments routed through an intermediary. The decisions are also dispersed across land leases, security reviews, environmental classifications, utility arrangements, supplier approvals and financing relationships. Each decision can appear narrow or technical while contributing to a combined outcome worth billions of dollars. A system with many gates does not necessarily reduce corruption risk. It may simply create more gatekeepers. ### Hiding the transfer The second risk does not require a personal payment. Public value can be transferred to private companies through relationships whose complete terms are difficult for the public to see—without a clear accounting of what the public receives in return. That value can include federal land, existing roads and water systems, environmental cleanup, security, emergency services, expedited agency coordination, transmission access, government-supported financing and the strategic advantage of being selected for a national AI project. Without a complete public accounting, it is difficult to determine whether the government is receiving fair value in return. A lease can require rent while still transferring land below its economic value. A developer can pay for its buildings and power while relying on publicly maintained infrastructure or leaving the government responsible for long-term environmental, service or decommissioning risks. The concern is not merely that some terms remain confidential. It is that the public may never see the combined bargain clearly enough to know whether public assets are producing public benefits or primarily enlarging private fortunes. ## Public permission should produce public leverage The answer is not to prohibit private development on federal sites. Former nuclear and industrial facilities may offer land, transmission access, security, infrastructure and redevelopment opportunities that are difficult to reproduce elsewhere. But public assets and federal gatekeeping should produce public leverage, not merely private access. At minimum, the government should do three things. First, it should publish a complete public accounting of the bargain. That means disclosing selection criteria, lease valuations, public financing or guarantees, infrastructure contributions, service arrangements and enforceable public benefits. Legitimately protected commercial or national-security details may require redaction, but confidentiality should not prevent publication of meaningful and substantive summaries. Second, it should subject the full project to integrated review before the principal commitments become irreversible. Land, electricity, generation, water, wastewater, roads, emergency response, environmental effects, security conditions and decommissioning obligations should be assembled into one visible public account. A collection of separate approvals is not the same as review of the combined bargain. Third, federal gatekeeping decisions need rules. Conditional Approvals, supplier restrictions, site selections and major regulatory classifications should have published criteria, conflict-of-interest protections, recorded reasons, predictable timelines and avenues for independent review. The goal is not to create one federal super-agency for AI infrastructure. It is to ensure that the institutions holding different pieces of the decision can evaluate the cumulative bargain before the developer has assembled commitments that are too expensive and politically important to reconsider. The federal government is not building a public AI utility at Paducah, Portsmouth or Savannah River. It is creating valuable pathways through which private firms can build and control strategic AI infrastructure. That makes the government more than a regulator. It becomes the landlord, selector, sponsor and gatekeeper. The question is whether that involvement gives the public lasting power over the infrastructure being created—or mainly gives a small number of private actors access to public sites, regulatory advantages and protected market positions. When public permission determines who can build the future, the public should be able to see who received that permission, why they received it and what they promised in return. --- ## Sources and further reading ### Paducah - [Department of Energy announcement selecting Brookfield, NextEra Energy and the participating utilities](https://www.energy.gov/articles/energy-department-announces-partnership-expand-reliable-affordable-energy-access-and-power?ref=theraceai.org) - [NextEra Energy project announcement with compute, generation, storage, utility and inherited-infrastructure details](https://www.investor.nexteraenergy.com/news-and-events/news-releases/2026/07-29-2026-133109921?ref=theraceai.org) - [Department of Energy request for offers and applicant responsibilities](https://www.energy.gov/em/articles/us-energy-department-seeks-proposals-ai-data-centers-energy-projects-paducah-site?ref=theraceai.org) - [Department of Energy account of local water, sewer, transportation and infrastructure planning](https://www.energy.gov/em/articles/study-guide-paducah-site-reindustrialization-nears-completion?ref=theraceai.org) ### Portsmouth - [Department of Energy fact sheet on the data-center, generation, transmission, federal-land and Japanese-funding arrangements](https://www.energy.gov/articles/fact-sheet-department-energy-ensuring-affordable-energy-access-ohio-while-powering-future?ref=theraceai.org) - [Department of Energy announcement of the SoftBank and AEP Ohio partnership](https://www.energy.gov/articles/energy-department-announces-partnership-ensure-affordable-energy-and-power-americas-ai?ref=theraceai.org) - [AEP announcement describing the planned transmission investment](https://www.aep.com/news/stories/view/10823/?ref=theraceai.org) - [SB Energy overview of the PORTS Technology Campus](https://sbenergy.com/digital-infrastructure/?ref=theraceai.org) ### Savannah River - [National Nuclear Security Administration announcement selecting Amentum for lease negotiations](https://www.energy.gov/nnsa/articles/nnsa-selects-amentum-ai-data-center-and-energy-project-savannah-river-site?ref=theraceai.org) - [National Nuclear Security Administration request for proposals and private-partner responsibilities](https://www.energy.gov/nnsa/articles/nnsa-seeks-proposals-ai-data-centers-energy-projects-savannah-river-site-0?ref=theraceai.org) - [Amentum announcement describing its proposed project](https://www.amentum.com/news/amentum-selected-for-the-department-of-energy-ai-data-center-and-energy-generation-project/?ref=theraceai.org) ### Federal AI-infrastructure strategy - [Department of Energy request for information on AI infrastructure on DOE lands](https://www.energy.gov/sites/default/files/2025-04/RFI%20to%20Inform%20Public%20Bids%20to%20Construct%20AI%20Infrastructure%20%28website%20copy%29%20-%202025.04.10.pdf?ref=theraceai.org) - [Department of Energy data-center resource hub](https://www.energy.gov/powering-americas-ai-future-data-center-resource-hub?ref=theraceai.org) ### EPA islanded-power guidance - [EPA memorandum clarifying Acid Rain Program provisions for islanded power generation](https://www.epa.gov/system/files/documents/2026-07/epa-issues-clarification-on-islanded-power-generators-and-acid-rain-program-provisions.pdf?ref=theraceai.org) - [EPA Clean Air Act resources for data centers](https://www.epa.gov/stationary-sources-air-pollution/clean-air-act-resources-data-centers?ref=theraceai.org) ### FCC Covered List and Conditional Approvals - [FCC public notice adding foreign-produced power inverters and advanced robotic devices to the Covered List](https://docs.fcc.gov/public/attachments/DA-26-786A1.pdf?ref=theraceai.org) - [FCC Covered List](https://www.fcc.gov/supplychain/coveredlist?ref=theraceai.org) - [FCC guidance for Conditional Approval applications involving power inverters](https://www.fcc.gov/sites/default/files/annex-a-guidance-conditional-approval-inverters.pdf?ref=theraceai.org) - [FCC guidance for Conditional Approval applications involving advanced robotic devices](https://www.fcc.gov/sites/default/files/robots-ca.pdf?ref=theraceai.org) ### What Happened In The Race In July? URL: https://www.theraceai.org/what-happened-in-the-race-in-july/ Last updated: 2026-08-04T00:51:12.000Z This week's note starts with the monthly update for July - where we try to gather up all the events we've studied during the month and tell you where we think things stand. The result - there's still a bias of activity toward centralization, but the window of contestability is still open. Looking at specific lanes, National Security was notably more centralized and closed during July, while Infrastructure Miniaturization and Civic Infrastructure were still wide open and leaning toward decentralization. Much more detail at the links below. Our major papers this week include an overview of actual labor force changes that seem to indicate the imminent death of the human-powered economy might be overstated - at least so far. The predictions that AI would force massive layoffs in a very short time aren't bearing out. It's much more nuanced than that. The second paper is the first in a new series that will explore the concept of 'capacity governance.' AI is forcing us to consider how our fractured decision-making public bodies can cope with the scale of demands data centers are bringing. This first paper outlines the demand we can see coming - and it's truly massive, on a scale of infrastructure development we literally have never seen in the US. Is our current manner of managing all the demands up to the task? I hope you find these perspectives enlightening and helpful. ## Where The Race Stands - July 2026 July’s strongest public actions show that the window of contestability remains real. States can create enforceable duties. Regulators can reach critical providers. Competition authorities can require access at controlled platform layers. Communities can organize, pause, appeal, and sometimes alter infrastructure trajectories. Public institutions can also begin to purchase shared capacity rather than merely subsidize private scale. But the month’s harder-to-reverse developments point in the opposite direction. Power and compute were committed for long periods. Federal agencies became gatekeepers for sites, recipients, and suppliers. AI providers moved deeper into institutional and personal operating layers. Firms began reorganizing teams and authority around AI before worker-side institutions could respond. The central July lesson is therefore not that the window closed. It is that the actors capable of making durable commitments are still moving faster than the institutions capable of contesting them. [Read the latest Scoreboard Review →](https://www.theraceai.org/scoreboard/) ### [The AI Layoff Story Is Too Simple](https://www.theraceai.org/the-ai-layoff-story-is-too-simple/) AI is not simply replacing jobs. It is compressing teams, shifting work outside firms, concentrating implementation expertise, and changing who holds authority. The early workforce struggle is over who controls the redesign—and whether workers can contest it. ### [From Land Use to Capacity Governance: The AI Infrastructure Problem Coming Into View](https://www.theraceai.org/from-land-use-to-capacity-governance-the-ai-infrastructure-problem-coming-into-view/) AI data centers are no longer merely local land-use projects. They are claims on regional power, water, infrastructure, and public capacity. Can our distributed institutions govern the whole before the choices become difficult to reverse? ## Learn More From The Race [Home](https://www.theraceai.org/home/) [Scoreboard](https://www.theraceai.org/scoreboard/) [Articles](https://www.theraceai.org/tag/articles/) [How The Race Works](https://www.theraceai.org/framework/) [The Eight Lanes](https://www.theraceai.org/the-eight-lanes/) [About Mark Howard and The Race](https://www.theraceai.org/about-author/) ### The AI Layoff Story Is Too Simple URL: https://www.theraceai.org/the-ai-layoff-story-is-too-simple/ Last updated: 2026-08-01T18:19:49.000Z ### AI is beginning to change the architecture of the firm—even as it remains unclear whether or when direct job replacement will become widespread Rumors of the human-powered economy’s imminent death appear to have been exaggerated. At least for now. The arrival of generative AI produced no shortage of predictions that machines would soon perform most economically useful work. The first broad evidence does not look much like the opening chapter of an economy without workers. A new analysis from the [UK Office for National Statistics](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026?ref=theraceai.org) found that reported AI use among businesses with at least ten employees rose from about 12 percent in late 2023 to about 35 percent in June 2026\. That is rapid adoption. But the adoption remains shallow. Only 10 percent of the businesses using AI described that use as extensive. Around half reported no overall change in headcount, and training or retraining existing workers was much more common than replacing them. In a separate question, around 10 percent said they were automating or replacing roles with AI. This is one country, one survey, and an early point in a long transition. It cannot tell us where the labor market will be in five or ten years. It does show that the immediate story is more complicated than a wave in which AI simply takes one job after another. That is not necessarily reassuring. Some large firms are already cutting jobs while reorganizing around AI. But the more revealing changes are often occurring somewhere between augmentation and direct replacement. Companies are compressing teams, removing management layers, moving work to outside providers, concentrating technical expertise, and changing the mix of workers they expect to need. They may be redesigning the firm around what they believe AI will soon make possible before they can show that an AI system has directly performed every job being eliminated. The question is therefore not only whether AI is replacing workers. It is whether AI is changing the kind of organization that firms believe they need—and who retains authority inside the organization that remains. ## A company does not need to automate every job to redesign itself around AI The restructuring announced by [monday.com](https://www.cio.com/article/4200330/monday-com-cuts-20-of-its-workforce-to-restructure-for-the-ai-era.html?ref=theraceai.org) in July captures the ambiguity. The workplace-software company said it would reduce its workforce by approximately 20 percent while continuing to hire in selected strategic areas. In its public filing, monday.com described the restructuring as part of its strategy to become an “AI Work Platform,” a new product and operating organization built around people and software agents working together. The reorganization is intended to achieve a leaner and more focused operating model, not a simple program of replacing employees with machines. That distinction matters. There is no public evidence showing that monday.com identified 600 jobs, assigned each job to an AI agent, and then dismissed the corresponding people. Its leaders have described a broader organizational change: fewer management layers, smaller autonomous teams, continued hiring for specialized roles, and a greater emphasis on helping customers implement AI inside their own operations. The company may be correct that its restructuring is not one-for-one automation. But that does not make AI incidental to the decision. It reveals why the usual test—*Did AI directly replace these workers?*—is too narrow. A firm can decide that smaller teams will produce more, fewer managers will be needed, agents will handle parts of routine execution, and specialists should concentrate on integrating AI into customer workflows. None of those changes requires a machine to reproduce the complete job description of each displaced worker. The org chart may change before the technology can fully justify it. This is one reason the early workforce evidence needs a more careful vocabulary. AI-related restructuring currently appears to follow at least five pathways. - **Replacement** occurs when AI performs tasks or roles that people previously performed. - **Compression** occurs when fewer workers are expected to produce the same or greater output with AI assistance. - **Outsourcing** moves work to vendors, contractors, offshore operations, or managed systems. - **Redeployment** moves existing workers into different or higher-value roles. - **Organizational recomposition** changes the firm’s layers, skill mix, team structure, decision rights, and division of work between people and machines. These are not separate boxes. A single restructuring can involve several at once. The important point is that only the first pathway looks like the familiar image of a machine taking a human job. The other four can change labor power, career opportunities, institutional knowledge, and the distribution of authority just as significantly. ## The clearest cases still involve real labor compression Some cases do look much closer to direct substitution. [Allianz Partners announced](https://www.reuters.com/legal/litigation/allianz-cut-up-1800-jobs-due-increasing-ai-use-2026-07-08/?ref=theraceai.org) that it would eliminate as many as 1,800 jobs in its travel-insurance operations because of increasing AI use. The affected work reportedly includes operations in which AI can handle more customer interactions and routine processing. Allianz has said it will seek to retrain or reassign employees where possible, but the connection between the technology and the planned workforce reduction is unusually explicit. This is stronger evidence than a general corporate layoff accompanied by a reference to “efficiency.” Management has identified AI as a significant cause of a substantial reduction in a defined operation. Even here, however, the organizational choice extends beyond replacing a set of manual tasks. Allianz must also decide which customer interactions remain human, when an automated system escalates a case, how exceptions are handled, who monitors quality, and how much service capacity should remain available when an AI system fails. To maintain and ideally improve performance, changes in operations must accompany headcount reductions. This example displays the broader process of labor compression: fewer people are expected to support a larger volume of work because some portion of production, administration, analysis, or coordination has become cheaper and faster with AI. Compression can occur even when the remaining workers become more capable. A claims specialist with AI assistance may handle more cases and spend more time on difficult judgments. But if the organization captures most of that capacity by reducing staffing and raising output expectations, the technology expands organizational leverage more than worker agency. The relevant question is not whether augmentation or replacement occurred. Both may occur at the same time. The question is who receives the benefit of the additional capacity. ## AI can also move work outside the firm The restructuring at [British American Tobacco](https://www.reuters.com/business/world-at-work/british-american-tobacco-cuts-5500-jobs-globally-2026-06-29/?ref=theraceai.org) shows a different pathway. BAT announced that approximately 9,000 roles would be affected by a cost-reduction and technology transformation program. About 5,500 jobs are expected to be eliminated. Another 3,500 roles have been or will be transferred to outside providers, including Accenture. BAT connected the restructuring to a broader effort to become more agile, technology-enabled, and able to use advanced AI capabilities supplied through strategic partners. It would be misleading to say that AI will perform all 9,000 jobs. BAT is also responding to declining cigarette volumes, regulatory pressures, changing product markets, and an ordinary corporate desire to reduce costs. AI is one component of a much larger restructuring. But the combination of workforce reduction, outsourcing, and AI investment shows why looking only for direct job replacement misses an important concentration mechanism. A job can disappear from a company without disappearing from the economy. Work may move to a consultancy, managed-service provider, lower-cost labor market, or technology platform. The incumbent reduces its permanent workforce while continuing to receive the service through a contract. That changes the boundary of the firm: what the company knows how to do itself, what it employs people to do, and what it purchases from organizations beyond its walls. It may also change where power resides. Workers inside a large company may have career paths, institutional knowledge, legal protections, and sometimes collective representation. Workers at an outside provider may face different wages, bargaining arrangements, and performance systems. Responsibility also becomes easier to divide between client and vendor. The knowledge needed to operate the function may also move outward. A consultancy or service provider that implements AI across many clients accumulates reusable expertise about workflows, integration, and organizational redesign. Each client receives a service. The provider builds a playbook. The firm may become leaner while becoming more dependent. This is why outsourcing should be treated as a distinct AI workforce pathway rather than an accounting detail appended to layoffs. AI may not simply substitute machines for employees. It may help large organizations substitute contractual relationships for internal capacity. ## A smaller workforce may be paired with a more specialized core The emerging pattern is not only fewer workers. It is a different composition of workers. [Thomson Reuters](https://www.reuters.com/legal/litigation/thomson-reuters-cut-small-number-engineering-jobs-2026-07-13/?ref=theraceai.org) is eliminating part of its existing engineering workforce while planning to add more than 250 net-new engineering positions over the next two years. The company says most of the new jobs will be senior and AI-native. That is not a simple story of engineering jobs disappearing. Nor is it a neutral exchange of one set of skills for another. It suggests that AI may change the internal ladder through which expertise is built. A company may need fewer general engineers while competing intensely for senior workers who can design AI systems, evaluate their output, understand a customer’s domain, and place the technology inside consequential workflows. The surviving organization may have a smaller base and a more selective technical core. The same possibility appears from the other direction at [Tata Consultancy Services](https://www.reuters.com/world/india/indias-tata-consultancy-services-plans-up-8900-ai-deployment-engineers-seeks-ai-2026-07-12/?ref=theraceai.org). TCS plans to create a corps of approximately 5,900 to 8,900 forward-deployed engineers—roughly 1 to 1.5 percent of its workforce—who will work closely with customers to adapt and integrate AI into real operations. The company has not yet said how many of those positions will be filled through internal retraining and how many through external hiring. The forward-deployed engineer represents a different kind of labor model altogether. These workers are not merely building a general product. They combine technical knowledge with detailed understanding of a customer’s processes, data, constraints, and institutional habits. They operate at the point where a model becomes an organizational system. That implementation knowledge may be among the most valuable forms of expertise in the AI economy. It may also become highly concentrated. A company can give thousands of employees access to AI while depending on a much smaller group to connect the systems, redesign work, set safeguards, and define performance. If that expertise sits inside a vendor, the customer may become more productive without becoming more capable of operating independently. This is the connection between workforce restructuring and implementation capacity. AI-driven restructuring does more than sort workers according to whether their tasks can be automated. It may create a new organizational hierarchy in the economy between people who simply use AI and the smaller group that designs, integrates, evaluates, and governs the systems everyone else must use. The pattern at monday.com combines these elements unusually clearly: a substantial workforce reduction, fewer layers, smaller teams, selective hiring, an AI-centered platform strategy, and greater emphasis on customer implementation. The most important change may not be that agents took 600 jobs. It may be that the company has formed a new view of the minimum human organization required to build and sell its products. If that model succeeds, other enterprise-software vendors will not merely offer customers AI tools. They will offer their own reorganized companies as evidence of what successful AI adoption is supposed to look like. The vendor becomes both supplier and demonstration project. ## Redeployment is real, but it needs evidence Not every company is presenting AI investment as a reason to reduce its workforce. [SAP has restricted hiring and travel spending](https://www.cio.com/article/4192663/sap-cuts-hiring-and-travel-to-fund-ai.html?ref=theraceai.org) to redirect resources toward AI. The company says it wants to avoid another large layoff program by retraining and redeploying existing employees into AI-enhanced work. Hiring will concentrate on selected technical and AI roles while current workers are encouraged to build new capabilities. This is an important alternative pathway. If firms can preserve institutional knowledge, teach employees to use new systems, and move people into more valuable work, AI could expand capability without requiring the first response to every productivity gain to be a workforce reduction. The ONS evidence suggests that this is presently more common than direct replacement. Training or retraining existing employees was the most frequently reported way UK businesses were integrating AI-related skills. But the effort remained limited: only 11 percent said that more than half of their workforce had received AI-related training. That encouraging direction should remain provisional. A credible redeployment claim requires more than a company announcement. We should eventually be able to see how many employees completed training, how many moved into durable new roles, whether their wages and job quality were maintained, and whether the transition preserved a meaningful career path. Hiring freezes and attrition can produce labor compression more slowly than layoffs while leaving the same final organization. Redeployment is a real pathway. It should not become a euphemism that remains permanently ahead of the evidence. ## Workers experience the same transition from very different positions The workforce debate often focuses on what AI can do. The democratic question is who can see and contest what an employer decides to do with it. A lawsuit involving 26 Meta employees shows the weakest position: an after-the-fact challenge in which the employer controls most of the evidence. The workers allege that Meta used productivity scores, AI-usage measures, and AI-assisted systems in ways that disadvantaged employees who had taken medical or family leave. Meta denies that AI selected the employees and says human managers made the decisions. A federal judge [declined to block the layoffs](https://www.reuters.com/business/world-at-work/us-judge-wont-block-meta-laying-off-workers-who-filed-ai-discrimination-lawsuit-2026-07-17/?ref=theraceai.org), while recognizing that the plaintiffs had raised serious questions about the alleged use of AI. Whatever the merits, the procedural problem is important. Workers challenging an AI-assisted employment decision are rarely present when the system is designed, the data selected, or the scores interpreted. They may lose their jobs before obtaining the records needed to understand what happened. Human review offers limited protection if workers cannot see how the system influenced the decision. Litigation may provide a remedy eventually. It is a poor substitute for visibility before the harm occurs. Google employees are trying to intervene earlier. More than 4,500 workers signed a [petition seeking guaranteed severance, voluntary buyouts before mandatory layoffs, and an end to forced performance-rating distributions](https://www.theguardian.com/technology/2026/jul/16/google-workers-layoff-protections-ai?ref=theraceai.org). The petition does not create binding rights, and it does not give employees shared authority over Google’s AI strategy. It does represent collective organization aimed at setting procedural terms before the next restructuring is complete. Collective bargaining can go further. The NewsGuild-CWA now reports approximately [85 to 90 contracts containing explicit AI provisions](https://www.axios.com/2026/07/26/union-contracts-ai-workplace-disruption?ref=theraceai.org). Existing agreements can require notice and bargaining before deployment, preserve bargaining-unit work, require human oversight, and create remedies when employers violate the agreed terms. These provisions do not stop technological change. They alter who participates in defining it. Illinois has created a still stronger form of prospective protection in one specific professional setting. A new law prohibits administrators from using AI to write teacher evaluations or perform evaluation tasks requiring professional judgment. AI may still assist with administrative work, but evaluators must disclose the tool and its purpose to the teacher. The law makes operational a clear principle before a disputed evaluation occurs: professional judgment remains a human responsibility. Together, these cases show four increasingly contestable positions from which workers may encounter AI-era restructuring. A Meta employee may seek evidence after the decision. Google employees are organizing for procedural leverage before another decision. A union contract can turn that leverage into an enforceable right to notice and bargaining. Illinois teachers work under a legal boundary defining something AI may not be allowed to decide. That progression—from remedy after harm, to collective pressure before harm, to enforceable bargaining rights, to a prospective legal boundary—is a useful measure of worker contestability. General assurances of “human oversight” are not enough. A person can approve a decision inside a process still defined by management and hidden from the worker affected. Contestability requires information, an opportunity to intervene, and an institution capable of changing the result. ## The question is who controls the redesign The early workforce evidence does not show an economy rapidly shedding the need for human work. It shows something less dramatic and, in the near term, more consequential. AI adoption is spreading faster than deep organizational transformation. Most firms are still experimenting, improving existing operations, and training at least some workers. At the same time, a smaller group of influential companies is beginning to reorganize: compressing staffing, changing firm boundaries, reducing layers, concentrating implementation expertise, and selecting a different mix of skills. Those companies do not have to prove that AI can autonomously perform every eliminated job. They can act on early productivity evidence, investor expectations, competitive pressure, and a belief that the technology will improve quickly. The restructuring may arrive before the causal evidence becomes clean. This is why the AI workforce question should not be reduced to counting layoffs attributed to automation. We should ask what work is changing, which roles and entry paths are disappearing, where organizational knowledge is moving, and who gains authority over the new systems. We should distinguish direct replacement from compression, outsourcing, redeployment, and recomposition. Most importantly, we should examine whether workers can see and influence the redesign before its assumptions become embedded in software, contracts, and a new org chart. Three tests will matter. Can workers and the public identify which tasks, workflows, and decision rights are changing? Do workers participate before the redesign is finalized, or only receive its consequences? Are there enforceable limits on what AI may decide, what employers must disclose, and what remedies are available? Rumors of the human-powered economy’s death may have been exaggerated. The struggle over who controls its reorganization has barely begun. ## References and further reading - Office for National Statistics, [“Artificial intelligence in UK businesses: 2023 to 2026”](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026?ref=theraceai.org), July 20, 2026. - monday.com, [“monday.com Goes All In on AI: From Work Management Platform to AI Work Platform”](https://ir.monday.com/news-and-events/news-releases/news-details/2026/monday-com-Goes-All-In-on-AI-From-Work-Management-Platform-to-AI-Work-Platform/default.aspx?ref=theraceai.org), May 6, 2026. - monday.com, [SEC filings](https://www.sec.gov/edgar/browse/?CIK=1845338&owner=exclude&ref=theraceai.org), July 2026. - CIO, [“monday.com cuts 20% of its workforce to restructure for the AI era”](https://www.cio.com/article/4200330/monday-com-cuts-20-of-its-workforce-to-restructure-for-the-ai-era.html?ref=theraceai.org), July 22, 2026. - Reuters, [“Allianz to cut up to 1,800 jobs due to increasing AI use”](https://www.reuters.com/legal/litigation/allianz-cut-up-1800-jobs-due-increasing-ai-use-2026-07-08/?ref=theraceai.org), July 8, 2026. - Reuters, [“British American Tobacco cost-cutting hits 9,000 roles”](https://www.reuters.com/business/world-at-work/british-american-tobacco-cuts-5500-jobs-globally-2026-06-29/?ref=theraceai.org), June 29, 2026. - Reuters, [“Thomson Reuters to cut ‘small number’ of engineering jobs”](https://www.reuters.com/legal/litigation/thomson-reuters-cut-small-number-engineering-jobs-2026-07-13/?ref=theraceai.org), July 13, 2026. - Reuters, [“India’s Tata Consultancy Services plans up to 8,900 AI deployment engineers”](https://www.reuters.com/world/india/indias-tata-consultancy-services-plans-up-8900-ai-deployment-engineers-seeks-ai-2026-07-12/?ref=theraceai.org), July 12, 2026. - Reuters, [“U.S. judge won’t block Meta from laying off workers who filed AI discrimination lawsuit”](https://www.reuters.com/business/world-at-work/us-judge-wont-block-meta-laying-off-workers-who-filed-ai-discrimination-lawsuit-2026-07-17/?ref=theraceai.org), July 17, 2026. - The Guardian, [“Thousands of Google workers demand layoff protections amid AI push”](https://www.theguardian.com/technology/2026/jul/16/google-workers-layoff-protections-ai?ref=theraceai.org), July 16, 2026. - Axios, [“Inside the AI protections union workers are winning”](https://www.axios.com/2026/07/26/union-contracts-ai-workplace-disruption?ref=theraceai.org), July 26, 2026. - Illinois Senate Democratic Caucus, [“Illinois bans AI use for teacher evaluations”](https://www.illinoissenatedemocrats.com/caucus-news/20-sen-christopher-belt-news/7027-illinois-bans-ai-use-for-teacher-evaluations?ref=theraceai.org), July 10, 2026. - CIO, [“SAP cuts hiring and travel to fund AI”](https://www.cio.com/article/4192663/sap-cuts-hiring-and-travel-to-fund-ai.html?ref=theraceai.org), July 3, 2026. ### Is the Window of Contestability Still Open? URL: https://www.theraceai.org/is-the-window-of-contestability-still-open/ Last updated: 2026-07-29T17:37:37.000Z The first article in this series argued that the institutional order around AI is still forming. Infrastructure is being built. Platforms are becoming operating layers. Governments and companies are signing contracts, reorganizing work, and deciding which systems will become difficult to live without. The period before those choices harden is the **window of contestability**. The second article argued that an open window matters only when institutions and affected groups have the capacity to act through it. That leaves a practical problem. When a new development occurs, how can we tell whether the window is still open? The answer will not come from one signal, score, or indicator. The arrangements forming around AI are too varied, and their effects often point in different directions. A law may create new authority without the staff needed to use it. An open model may expand choice while remaining dependent on concentrated compute. A platform may offer extraordinary new capacity while making departure increasingly expensive. What we need is a small set of questions that can be applied to actual events. In our own analysis, we have begun using six questions: 1. **Can we see how AI is being used?** 2. **Can we see what commitments and dependencies are forming?** 3. **Can anyone intervene before those commitments harden?** 4. **Can affected people obtain review, correction, or remedy?** 5. **Can institutions or users leave without losing their capacity?** 6. **Is another arrangement genuinely available?** These are evaluation questions, not permanent metrics. They will need to be revised as events reveal new forms of power and dependency. Their value lies in helping us see how power is shifting in practice, not in the abstract. Four recent cases show how differently the window can move. ## New York Buys Time In July 2026, the state of New York [temporarily paused incomplete state permit applications for new or expanded hyperscale data centers capable of using at least 50 megawatts](https://www.reuters.com/world/new-york-becomes-first-state-impose-data-center-moratorium-2026-07-14/?ref=theraceai.org). The state did not prohibit data centers. It stopped part of the approval pipeline for a limited set of projects while agencies developed common standards covering grid interconnection, environmental effects, water use, cost allocation, and community benefits. [The executive order requires a formal public process, including public comment and a hearing](https://www.governor.ny.gov/executive-order/no-62-establishing-temporary-moratorium-data-centers-new-york-while-state-develops?ref=theraceai.org). That distinction matters. A pause can be an evasion—government postponing a decision it does not want to make. It can also be a way to prevent a series of private projects from becoming public policy by accumulation. New York’s action is the second kind, at least for now. ### Applying the six questions **1\. Can we see how AI is being used?** Partly. The order is directed at large data centers rather than at the AI systems they will support. But it makes clear that very large computational facilities are no longer ordinary commercial developments. **2\. Can we see what commitments and dependencies are forming?** More clearly than before. The state has identified electricity, transmission, water, environmental effects, public costs, and community benefits as parts of one connected decision rather than separate technical details. **3\. Can anyone intervene before those commitments harden?** Yes. This is the case’s strongest feature. The pause applies before incomplete state approvals are granted. Public institutions are acting while project terms and infrastructure obligations can still be changed. **4\. Can affected people obtain review, correction, or remedy?** Potentially. The promised public process creates a route into the decision, although its value will depend on the standards eventually adopted and whether public participation changes them. **5\. Can developers or communities leave without losing their capacity?** This question is less central here, but the pause preserves flexibility. Projects can be redesigned, conditioned, relocated, or rejected before public and private investments make those choices much harder. **6\. Is another arrangement genuinely available?** Yes. New York can allow data-center development under different cost-allocation, water, grid, reporting, and community-benefit rules. The choice is not simply construction or prohibition. ### The read: the window is being preserved New York has not solved data-center governance. It has done something more basic and more urgent: prevented the current pipeline from settling the rules before the state writes them. The window is open because the state has retained time, authority, and alternatives. Whether it remains open will depend on what New York does with them. ## Europe Tries to Reopen Android A different kind of contestability problem is forming around AI assistants. As assistants become capable of sending messages, accessing files, controlling devices, shopping, searching, and completing work, the owner of the operating system can determine which assistant works best and which one merely appears in an application window. In July 2026, the [European Commission issued binding measures](https://digital-markets-act.ec.europa.eu/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data-under-2026-07-16%5Fen?ref=theraceai.org) requiring Google to give qualifying rival AI assistants free and effectively equivalent access to 11 Android hardware and software features. The Commission also required access to certain anonymized Google Search interaction data for qualifying competing search and AI services. The measures do not prevent Google from developing Gemini or integrating it with Android. They attempt to stop ownership of Android and Search from deciding the assistant market in advance. ### Applying the six questions **1\. Can we see how AI is being used?** Yes. The Commission has identified the operating system itself as a source of AI advantage. The problem is not only model quality; it is privileged access to device functions and search data. **2\. Can we see what commitments and dependencies are forming?** Yes. If one assistant has superior access to cameras, microphones, notifications, applications, and user context, people may become dependent on it before competitors can offer an equivalent service. **3\. Can anyone intervene before those commitments harden?** The Commission has intervened, although not at the beginning. Google already controls the operating system and search infrastructure. The action is an attempt to prevent those positions from determining the next layer of the market. **4\. Can affected people obtain review, correction, or remedy?** More than before. The measures are binding and enforceable under the Digital Markets Act. The practical test will be whether rivals can challenge delays, degraded access, or technical restrictions quickly enough to matter. **5\. Can users leave without losing their capacity?** The measures should make choosing another assistant more practical. But they do not yet solve the deeper problem of accumulated memory, preferences, permissions, files, and workflow history. **6\. Is another arrangement genuinely available?** Potentially. Rival assistants may now be able to compete at the operating-system level rather than as isolated applications. That possibility depends on implementation, not the language of the decision alone. ### The read: the window may be reopening The Commission is not merely slowing concentration. It is trying to reverse an advantage that had already begun to harden. That makes this a reopening case. The result is not guaranteed. Formal access can be undermined by technical design, privacy review, delay, unequal performance, or the inability to move accumulated context. But the intervention changes the structure of the contest. It creates a legal basis for alternatives that the platform owner might otherwise have been able to exclude. ## OpenAI Presence Moves Deeper Into the Institution A new product called [OpenAI Presence](https://openai.com/index/introducing-openai-presence/?ref=theraceai.org) shows the window moving in the other direction. Presence is a managed enterprise platform for building and operating AI agents. It combines OpenAI models with company systems, policies, permissions, testing, monitoring, human escalation, and continuing improvement. The customer decides what an agent may do, which actions require approval, and when a person must take over. Those are useful capabilities. They also move OpenAI beyond supplying a model. OpenAI can become part of the machinery through which an organization defines, evaluates, monitors, and improves its workflows. That is where dependence may become more durable. ### Applying the six questions **1\. Can we see how AI is being used?** Within the customer organization, probably more clearly than in many improvised deployments. Presence includes policies, permissions, monitoring, testing, and escalation. From outside, visibility is limited. The public will generally not know which organizations use it or which decisions have been delegated to agents. **2\. Can we see what commitments and dependencies are forming?** Only partly. The product description explains what Presence manages. It does not yet establish whether customers can export their policies, evaluations, configurations, action histories, and other accumulated operating context in a form another provider can use. **3\. Can anyone intervene before those commitments harden?** Customers can set limits on individual agents. That is real operational control. The harder question is whether they can impose terms on the platform relationship itself before their workflows and governance systems become deeply integrated with it. **4\. Can affected people obtain review, correction, or remedy?** Presence includes human escalation and monitoring. Those are valuable safeguards. They are not the same as independent review. Much will depend on whether workers, customers, regulators, and auditors can obtain the evidence needed to challenge consequential uses. **5\. Can the institution leave without losing its capacity?** We do not yet know. That is the central question for this case. If the customer can move its policies, evaluations, workflows, connectors, and monitoring history, Presence may build transferable capacity. If those remain embedded in the OpenAI relationship, leaving may mean rebuilding the institution’s AI operating system. **6\. Is another arrangement genuinely available?** Other models and enterprise platforms exist. The question is whether they remain genuine substitutes after years of institutional knowledge accumulate inside Presence. ### The read: the window is narrowing Presence does not close the window. It may produce safer deployments, clearer internal rules, and better human control than organizations would build on their own. But convenience can become dependence one reasonable decision at a time. The risk is not simply that OpenAI supplies the model. It is that OpenAI becomes the custodian of the knowledge required to operate and govern the system. The window narrows as the cost of leaving rises. ## Louisiana Builds Around Hyperion The Hyperion data center in Richland Parish, Louisiana, shows what it looks like when infrastructure commitments begin to approach closure. [Louisiana says the expanded project will exceed $50 billion and reach five gigawatts of compute capacity](https://www.opportunitylouisiana.gov/news/meta-commits-more-than-50-billion-for-north-louisiana-project-becoming-one-of-the-largest-data-centers-in-history?ref=theraceai.org). The state describes thousands of short-term construction jobs, 1,000 permanent jobs, major contracts for Louisiana businesses, and more than $1 billion in local infrastructure investment. Those benefits are substantial. So are the dependencies. A [New York Times investigation](https://www.nytimes.com/2026/07/27/technology/meta-data-center-louisiana.html?smid=url-share&unlocked%5Farticle%5Fcode=1.01A.KrHS.NSFvkYAG4XLV&ref=theraceai.org) found that the state negotiated the project through a code-named Meta subsidiary and under a nondisclosure agreement that went so far as to conceal the negotiations themselves. [WWNO later published details of the agreement](https://www.wwno.org/politics/2026-07-24/gov-jeff-landry-personally-signed-an-nda-with-meta-heres-what-it-says?ref=theraceai.org). Electricity, transmission, water, roads, workforce programs, tax treatment, and other public systems are being reorganized by multiple state and local agencies around one integrated corporate project. Meta could evaluate the complete package. The public encountered it through separate decisions made by separate institutions. That asymmetry is the heart of the case. The individual decisions appear to have moved through the legally required processes, but those processes were fragmented across agencies and attracted little public awareness or involvement. By the time the full project and its accumulated commitments became visible, Hyperion had acquired momentum that made meaningful public intervention far more difficult. ### Applying the six questions **1\. Can we see how AI is being used?** At the broadest level, yes. Hyperion is being built as infrastructure for Meta’s AI ambitions. The specific workloads, expansion sequence, operating requirements, and long-term public obligations are much less visible. **2\. Can we see what commitments and dependencies are forming?** Only in pieces. The public can see an enormous investment, power requirement, infrastructure program, and promised economic benefits. It has had much less opportunity to examine the cumulative bargain linking subsidies, energy, transmission, water, public services, employment, and later expansion. **3\. Can anyone intervene before those commitments harden?** Public agencies retain authority over individual components. But the integrated decision has moved far ahead. Once a state has reorganized infrastructure, workforce programs, political expectations, and regional development around a $50 billion project, rejecting or substantially changing later phases becomes much harder. **4\. Can affected people obtain review, correction, or remedy?** There are regulatory proceedings and public decisions around parts of the project. What appears to be missing is one institution or process capable of reviewing the complete arrangement and imposing conditions across it. **5\. Can Louisiana leave without losing its capacity?** Not easily. Infrastructure built for a five-gigawatt AI campus is not readily repurposed. Local revenue, employment expectations, utility planning, and public investment can make the state increasingly dependent on Meta completing and continuing the project. Meta retains greater mobility than Louisiana does. **6\. Is another arrangement genuinely available?** At this scale, alternatives are limited. Few companies can replace a five-gigawatt compute customer or a promised investment of more than $50 billion. As public systems are built around Hyperion, the state’s bargaining position may weaken rather than strengthen. ### The read: the window is closing The Hyperion window is not completely closed. Later phases still require infrastructure, financing, approvals, and execution. Public disclosure, enforceable cost allocation, parent guarantees, employment commitments, and cumulative review could preserve some leverage. But the direction is clear. The project’s scale creates its own political gravity. Each new commitment makes the next one easier to justify and the entire arrangement harder to reconsider. This is how a window closes without one formal decision announcing that it has closed. ## Where Practical Choice Remains The six questions do not produce automatic answers. They reveal where practical choice still resides. New York preserved choice by acting before permits and infrastructure commitments hardened. Europe attempted to restore choice by separating the assistant layer from control of Android and Search. OpenAI Presence may narrow choice by allowing institutional knowledge to accumulate inside one provider relationship. Hyperion approaches closure because an entire network of public and private commitments is being organized around one project before the public can evaluate the whole. The difference is not whether the underlying technology is useful. Each case may produce real value. The difference is whether that value is being built through arrangements that remain visible, challengeable, and replaceable. These six questions are provisional. Events will show which ones need to be revised, combined, or divided. But they give us something more useful than an abstract warning that concentration may someday occur. They let us ask, while the decisions are still being made: Where does the power sit now? Who can still change the arrangement? And how expensive is another path becoming? ## Sources and Further Reading The links embedded above support the principal facts used in each case. The following sources provide additional detail, primary documents, and competing perspectives. ### New York data-center pause - [Reuters: “New York becomes first state to impose data center moratorium”](https://www.reuters.com/world/new-york-becomes-first-state-impose-data-center-moratorium-2026-07-14/?ref=theraceai.org) — independent overview of the one-year pause and the debate surrounding it. - [New York Executive Order No. 62](https://www.governor.ny.gov/executive-order/no-62-establishing-temporary-moratorium-data-centers-new-york-while-state-develops?ref=theraceai.org) — the complete order, including the permitting pause, public process, community-investment framework, and proposed cost protections. ### European action on Android and Search - [European Commission: AI interoperability on Android and sharing of Google Search data](https://digital-markets-act.ec.europa.eu/commission-provides-guidance-google-ai-interoperability-android-and-sharing-google-search-data-under-2026-07-16%5Fen?ref=theraceai.org) — official summary of the two binding specification decisions. - [Associated Press: “EU forces Google to share search data and open Android to rival AI companies”](https://abcnews.com/Technology/wireStory/eu-forces-google-share-search-data-open-android-134811713?ref=theraceai.org) — independent account of the decision and its competitive implications. - [Kluwer Competition Law Blog: analysis of the Android interoperability decision](https://legalblogs.wolterskluwer.com/competition-blog/keeping-the-gate-the-european-commission-issues-its-specification-decision-on-alphabets-operating-system-interoperability-with-ai-assistants/?ref=theraceai.org) — a closer legal examination of what the Commission required and what implementation may involve. ### OpenAI Presence - [OpenAI: “Introducing OpenAI Presence”](https://openai.com/index/introducing-openai-presence/?ref=theraceai.org) — the product announcement and OpenAI’s description of its policies, permissions, evaluations, monitoring, and continuing improvement process. - [The Wall Street Journal: “OpenAI’s High-Stakes, High-Touch Push to Make AI Work for Business”](https://www.wsj.com/cio-journal/openais-high-stakes-high-touch-push-to-make-ai-work-for-business-7f08b7d4?ref=theraceai.org) — reporting on OpenAI’s broader effort to embed its technology and personnel more deeply in enterprise operations. - [VentureBeat: overview of the Presence launch](https://venturebeat.com/orchestration/openai-unveils-presence-a-new-platform-that-lets-enterprises-launch-and-manage-realtime-voice-agents-and-chatbots?ref=theraceai.org) — additional product detail and enterprise context. ### Meta Hyperion in Louisiana - [The New York Times: “How Meta Got Everything It Wanted in a Secret Louisiana Data Center Deal”](https://www.nytimes.com/2026/07/27/technology/meta-data-center-louisiana.html?smid=url-share&unlocked%5Farticle%5Fcode=1.01A.KrHS.NSFvkYAG4XLV&ref=theraceai.org) — the principal investigation into the negotiations, secrecy, and fragmented public decision-making. - [The New York Times: five takeaways from the Louisiana deal](https://www.nytimes.com/2026/07/27/business/meta-data-center-louisiana-takeaways.html?smid=url-share&unlocked%5Farticle%5Fcode=1.01A.WcqH.bENbo7ElClxp&ref=theraceai.org) — a shorter summary of the investigation’s major findings. - [WWNO: “Gov. Jeff Landry personally signed an NDA with Meta. Here’s what it says.”](https://www.wwno.org/politics/2026-07-24/gov-jeff-landry-personally-signed-an-nda-with-meta-heres-what-it-says?ref=theraceai.org) — local reporting on the agreement and a link to the underlying document. - [Louisiana Economic Development: Meta’s $50 billion expansion announcement](https://www.opportunitylouisiana.gov/news/meta-commits-more-than-50-billion-for-north-louisiana-project-becoming-one-of-the-largest-data-centers-in-history?ref=theraceai.org) — the state’s account of the project’s scale, jobs, contracts, and infrastructure benefits. - [Entergy Louisiana: the expanded power agreement with Meta](https://www.entergy.com/news/entergy-louisiana-announces-a-new-agreement-with-meta-that-will-deliver-an-additional-2b-in-customer-savings?ref=theraceai.org) — the utility’s account of generation plans, cost allocation, and projected customer savings. ### From Land Use to Capacity Governance: The AI Infrastructure Problem Coming Into View URL: https://www.theraceai.org/from-land-use-to-capacity-governance-the-ai-infrastructure-problem-coming-into-view/ Last updated: 2026-07-28T18:10:58.000Z AI data centers are still commonly governed as buildings on parcels of land, large utility customers, or private capital projects. At the scale now coming into view, they are becoming something larger: claims on regional capacity. That claim crosses systems that government usually manages separately. Electricity demand can shape generation and transmission. Those choices can affect water use, emissions, and rates. Local siting decisions can create costs and constraints well beyond the jurisdiction that approved the project. No single institution is necessarily ignoring the problem. The harder issue is that each may ask a reasonable question about one part of it while no public process asks whether the region can absorb the whole. That is the emerging shift from land-use governance to capacity governance. ## The mismatch The United States is beginning to build AI infrastructure at a scale usually associated with national projects. Much of it is still governed as ordinary private development. Some of that infrastructure will be necessary if AI is going to support medicine, science, and public services. The issue is not whether data centers should be built. It is whether the institutions around them are asking a large enough question. A local government asks whether a project belongs on a particular site. A utility asks whether it can serve the load and recover its costs. A water provider asks whether supply is available. All three questions are legitimate. None is sufficient by itself. The larger question is: > Can this region absorb the linked demand for power, water, transmission, land, and public cost—in this place, at this time, and under these conditions? By **capacity governance**, I mean the public rules and institutions used to decide how much regional capacity can be committed to AI infrastructure, on what terms, and at whose risk. AI data centers are not just buildings. They are claims on systems that other people also depend upon. ## The scale is forcing the issue For years, data centers sat in the background of the digital economy. AI is moving them toward the center of infrastructure planning. The [International Energy Agency](https://www.iea.org/reports/energy-and-ai/energy-demand-from-ai?ref=theraceai.org) projects that global data-center electricity use will more than double by 2030 in its base case, growing more than four times as fast as electricity demand from other sectors. In the United States, [Lawrence Berkeley National Laboratory](https://newscenter.lbl.gov/2025/01/15/berkeley-lab-report-evaluates-increase-in-electricity-demand-from-data-centers/?ref=theraceai.org) estimates that data centers could consume between 6.7 and 12 percent of the nation’s electricity by 2028, up from about 4.4 percent in 2023. The forecasts will not be exactly right. Some announced projects will be delayed, reduced, or abandoned. But uncertainty at this scale is itself a governance problem. Utilities and communities must make long-lived decisions today for loads that may—or may not—reshape regional systems tomorrow. Private spending points in the same direction. A Bridgewater analysis [reported by Reuters](https://www.reuters.com/business/big-tech-invest-about-650-billion-ai-2026-bridgewater-says-2026-02-23/?ref=theraceai.org) estimated that Alphabet, Amazon, Meta, and Microsoft could invest about $650 billion in AI-related infrastructure in 2026 alone. [A Wall Street Journal analysis found that, over the previous three years, leading technology companies had committed more to AI data centers, chips, and energy than the inflation-adjusted cost of building the Interstate Highway System over four decades.](https://www.wsj.com/tech/ai/ai-bubble-building-spree-55ee6128?ref=theraceai.org) This is private investment, not a public works program. Yet once investment reaches this scale, its consequences do not remain inside corporate balance sheets. They arrive in electricity plans, water commitments, and utility rates. “Tech capex” becomes a public-capacity question. ## The project creates a whole; government reviews pieces Public authority is divided for understandable reasons. Land-use departments know land use. Utilities know their systems. Regulatory commissions know rates and service obligations. The problem is that a large data-center project does not divide its effects so neatly. Start with electricity and water. A data center requires enormous amounts of power, but the water implications depend on how that power is produced and how the facility is cooled. Water-based cooling can reduce electricity use while increasing direct water consumption. Air cooling can conserve water while requiring more power. A choice made to protect one system can therefore place additional pressure on another. That tradeoff becomes more consequential when a local development decision commits regional resources. In 2019, Mesa, Arizona, agreed to make 1,120 acre-feet of water available annually for a proposed Google data center. The amount could rise to 4,480 acre-feet—roughly four million gallons a day—if Google met specified development milestones. ([TIME](https://time.com/5814276/google-data-centers-water/?ref=theraceai.org)) Mesa was negotiating a project within its boundaries, but the agreement did more than decide what could be built on a parcel. It conditionally reserved a substantial amount of capacity within a municipal water system that draws upon shared regional supplies in an already water-stressed state. Repeated across several cities, such agreements can determine how much regional water remains available for other development or for responding to future shortages—without any single decision appearing to make that larger allocation. Google later shifted the Mesa facility to air cooling, so the authorized maximum did not become its actual operating demand. That was a meaningful response to local water conditions. It also moved part of the capacity claim from water to electricity. The example is therefore not evidence that Mesa made the wrong decision. It shows how a nominally local agreement can implicate several regional systems, and how solving one part of the problem can change another. ([City of Mesa](https://apps.mesaaz.gov/meetingarchive/ArchiveDocuments/Documents/%7BE84289C7-5E1F-4E0E-8688-DC4A8D66412B%7D%5F0.pdf?ref=theraceai.org)) The same pattern appears in utility finance. A utility may conclude that it can serve a large customer by building new generation, transmission, or substations. But if the projected load arrives late—or never arrives—the infrastructure remains. Whether the data-center operator or existing customers bear that cost depends on collateral requirements, minimum-payment commitments, and exit fees. What begins as a service agreement with one customer can become a long-term ratepayer obligation. The existing governance system of distributed decisions is hardly idle. Local governments review sites and development agreements. Utilities and state commissions decide how service will be provided and costs allocated. Grid operators study reliability and transmission. Water providers assess supply. Each institution can be active and competent while still seeing only one part of the claim. The Federal Energy Regulatory Commission’s recent large-load proceedings illustrate both the progress and the limitation. FERC has pressed regional grid operators to clarify how data centers connect to the transmission system while protecting reliability and ratepayers. ([FERC, 2025](https://www.ferc.gov/news-events/news/ferc-orders-action-co-location-issues-related-data-centers-running-ai?ref=theraceai.org); [FERC, 2026](https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration?ref=theraceai.org)) That work matters. It can make the electricity portion of the decision more transparent and fair. But FERC does not decide whether a region should commit scarce water to the same project, whether a local tax agreement justifies the public costs, or whether several projects together exceed the capacity a metropolitan area can comfortably support. The project creates a whole. The existing system distributes responsibility for its pieces. What remains uncertain is who, if anyone, is responsible for putting those pieces back together before the commitments become difficult to reverse. ## Bad outcomes do not require bad actors It is tempting to tell this story as a conflict between technology companies and the public. Sometimes that framing fits. Secrecy, rushed approvals, and weak disclosure can make public oversight little more than a procedural cameo. But the deeper problem does not require a villain. A city can reasonably approve a project that fits its plan. A utility can reasonably agree to serve a new customer. A regulator can reasonably approve a tariff that appears to protect existing ratepayers. The overall result can still be poorly governed. Ratepayers may inherit costs built around an optimistic demand forecast. Water commitments may accumulate before their regional effect is visible. Local approvals may lock in assumptions before wider grid consequences are understood. None of this requires institutional incompetence. It requires only a mismatch between the scale of the capacity claim and the structure of the decision-making system. ## Private capital will not wait for coherence Private capital is one reason the United States can build quickly. It can finance facilities, secure equipment, and take risks that government would struggle to take directly. It also seeks speed and certainty. When public systems are fragmented or slow, companies will route around the bottlenecks. They may co-locate with power plants, build generation behind the meter, or offer to reduce demand during grid emergencies in exchange for faster service. Those arrangements may be sensible. A company that adds power, pays its own costs, and can curtail demand may be easier to serve than one that simply asks the grid for immediate full service. But every workaround raises the same question: does it solve the regional capacity problem, or move the pressure somewhere less visible? A private power agreement may shorten the interconnection wait while increasing emissions. Air cooling may conserve water while increasing electricity demand. A confidential development agreement may speed land assembly while weakening public trust. Private routing decisions will shape the public system whether public institutions are ready or not. If government cannot ask the full capacity question early enough, the important commitments will be made before the public can see how they fit together. ## From managed fragmentation to capacity governance The practical near-term path is to strengthen the institutions we already have. Three improvements matter most. Large projects should disclose their expected demands clearly enough for the public and regulators to evaluate them. Customers should bear the costs and risks of infrastructure built primarily for their benefit. Projects with regional effects should trigger notice and coordination beyond the jurisdiction issuing the land-use approval. That would be managed fragmentation: a divided system made more transparent, connected, and fair. It may be enough in some regions. In others, cumulative pressure may require a new coordinating layer—not a national agency that replaces local governments and utility regulators, but a process capable of putting the full regional claim into view before the pieces are committed one at a time. The form should follow the problem. A water-stressed basin may need one model. A grid-constrained data-center corridor may need another. A fast-growing metropolitan region facing simultaneous land, power, and water pressures may need a third. Uniformity is not the point. Matching the scale of governance to the scale of the capacity claim is. ## Naming the problem while the choices are still open The United States is still early enough in the AI infrastructure buildout to shape its rules. The window will not remain open indefinitely. If each project is treated as an isolated land-use case or utility request, larger capacity decisions will be made by default through private deals, one-off approvals, and emergency responses. By the time the cumulative effects are obvious, many choices may already be expensive to reverse. This is not an argument against building AI infrastructure. It is an argument for governing the buildout at the scale at which its consequences arrive. The first step is to name the problem correctly. AI data centers are claims on regional capacity. The next step is to decide whether our distributed institutions can manage those claims—or whether some regions will need a new public layer capable of seeing the whole. That is where this series will go next. ### Updates: Changes in The Race and the website URL: https://www.theraceai.org/updates-changes-in-the-race-and-the-website/ Last updated: 2026-07-26T21:36:55.000Z I have spent the last several weeks reorganizing and restructuring the website of The Race. The major change is rather than try to reduce coverage to a single stream, we're now tracking activity across eight 'lanes'. Activity in AI adoption is happening in so many places in so many ways it proved ungainly and somewhat useless to try to reduce to a single summary stream. The eight lanes admittedly overlap, but they help give more angles from which to view events. Identifying meaningful trends and providing thoughtful analysis then becomes easier. The lanes overlap, but I've found the focus to be helpful. [You can read more about the eight lanes here.](https://www.theraceai.org/the-eight-lanes/) I'm also trying to figure out the best way to give weekly updates without getting bogged down in detail, and without overselling specific events. I will start with a brief weekly summary and then just provide links to the latest Signal and/or Essay articles we've written. Enjoy! ## Summary of Events: Week Ending July 26 The Race tilted toward concentration this week, although meaningful counterweights kept the window open. [Hut 8 placed 704 megawatts of its Beacon Point campus under 15-year leases to one unnamed AI tenant](https://www.hut8.com/news-insights/press-releases/hut-8-fully-commercializes-1-gw-beacon-point-ai-data-center-campus-with-second-352-mw-it-lease?ref=theraceai.org), locking in control of scarce infrastructure before the tenant and its public obligations are clear. A [Northern Virginia transmission fault caused data centers to switch abruptly to backup power](https://www.investing.com/news/stock-market-news/massive-disconnect-of-power-roiled-largest-us-electric-grid-4807202?ref=theraceai.org), removing more than three gigawatts of demand from the PJM grid and demonstrating that concentrated computational loads can now have system-wide operational effects. OpenAI also disclosed that [models undergoing a cyber evaluation escaped their constrained environment and compromised Hugging Face infrastructure](https://openai.com/index/hugging-face-model-evaluation-security-incident/?ref=theraceai.org), while the launch of [OpenAI Presence](https://openai.com/index/introducing-openai-presence/?ref=theraceai.org) extends the company’s role from supplying models into managing enterprise agents, workflows, policies, evaluations, and continuing deployment. The countervailing developments were significant but less binding. The administration’s voluntary [Ratepayer Protection Pledge expanded to governors, utilities, and data-center developers](https://apnews.com/article/trump-ai-data-centers-pledge-490ea7e4c7227d5e550b00a0056c33c9?ref=theraceai.org), strengthening the principle that large AI users should pay the infrastructure costs they create. A broad industry coalition also defended [open-weight models as infrastructure for competition, independent research, and institutional sovereignty](https://images.nvidia.com/pdf/Open-Weights-and-American-AI-Leadership.pdf?ref=theraceai.org). Neither development yet guarantees enforceable cost allocation or genuinely independent AI capacity, but both preserve avenues for contesting the more concentrated future taking shape. Overall, the week strengthened the concentration side of the Race while leaving important opportunities for public rules, open infrastructure, and institutional intervention. ## Latest Signal ### Utility Governance Is Democracy’s Entry Point Into the AI Race AI’s physical expansion depends on electricity, water, land, and grid access. That gives utilities, regulators, and local governments a crucial opening to shape who pays, who benefits, and whether the public retains a meaningful say in the AI race. [Read the article →](https://www.theraceai.org/utility-governance-is-democracys-entry-point-into-the-ai-race/) ## Latest Essay ### States Are Starting to Build the Public Layer for AI States are beginning to build a public layer for AI through inventories, procurement rules, training, shared tools, and limits on consequential uses. The framework is emerging, but dependence on private platforms remains its weakest point. [Read the article →](https://www.theraceai.org/states-are-starting-to-build-the-public-layer-for-ai/) ## Learn More From The Race [Home](https://www.theraceai.org/home/) [Scoreboard](https://www.theraceai.org/scoreboard/) [Articles](https://www.theraceai.org/tag/articles/) [How The Race Works](https://www.theraceai.org/framework/) [The Eight Lanes](https://www.theraceai.org/the-eight-lanes/) [About Mark Howard and The Race](https://www.theraceai.org/about-author/) ### Institutional Capacity to Contest: Why an Open Window Means Little Without the Power to Act URL: https://www.theraceai.org/institutional-capacity-to-contest-why-an-open-window-means-little-without-the-power-to-act/ Last updated: 2026-07-25T21:00:54.000Z [The AI Concentration Window](https://www.theraceai.org/the-ai-concentration-window-why-the-productivity-paradox-may-shape-who-controls-the-next-economy/) argued that the institutional order around AI is still unsettled. The models are improving. The infrastructure is being built. Companies and governments are choosing platforms, writing contracts, and redesigning work. Many of those arrangements have not yet hardened beyond challenge. That makes the present period a window of contestability. But a window can be open in theory and unavailable in practice. A local government may have legal authority to review a data-center proposal but lack the expertise to evaluate it. A worker may have a right to question an automated decision without access to a meaningful appeal. A public agency may be authorized to regulate AI while relying on the companies it oversees for technical evidence. Nothing in those examples is formally closed. Very little is meaningfully contestable. By **institutional capacity to contest**, I mean the practical ability of public institutions and affected groups to understand how AI is being used, intervene before decisions harden, and obtain change or remedy. Contestability is not the same as opposition. A system is not contestable only when someone can stop it. It is contestable when affected people and institutions can shape its purposes, conditions, and distribution of risk—and when their participation can alter the outcome. That requires more than a principle. It requires a chain capable of turning principle into power. ## From Principle to Contestability Democratic governance often describes openness through procedure. Was notice provided? Was public comment allowed? Is there a right to appeal? Those questions matter. Procedural openings are better than decisions made entirely in secret. But they can also create a reassuring picture of contestability that does not survive contact with institutional reality. A thirty-day comment period means one thing to a company with lawyers, technical staff, and an established relationship with the regulator. It means something rather different to a school district trying to understand a new AI procurement or a small city facing an infrastructure proposal that may reshape its utility system. Everyone may be invited into the same process. They do not arrive with the same capacity. The [2026 Global Index on Responsible AI](https://www.global-index.ai/?ref=theraceai.org) illustrates the gap. Across 135 countries and jurisdictions, only 27 percent had operational mechanisms for civil-society participation. Just 26 percent had frameworks for fair and accountable public procurement of AI, while 18 percent required public disclosure of government algorithmic systems. ([GIRAI key findings](https://www.global-index.ai/takeaways?ref=theraceai.org)) Governance is arriving in layers, and the layers do not always connect. A useful way to see the problem is as a five-step ladder: > **Principle → Obligation → Institution → Operation → Contestability** A **principle** establishes direction: fairness, transparency, or human accountability. An **obligation** turns that principle into a duty. Someone must disclose, assess, or correct. An **institution** must interpret and enforce the duty. **Operation** is where that institution acquires staff, evidence, routines, and technical competence. Only then does **contestability** become real. Affected people can intervene, bargain, appeal, or obtain remedy. The European Union’s AI Act shows how the first three steps can be connected. For certain deployers of high-risk systems, it requires a [fundamental-rights impact assessment](https://digital-strategy.ec.europa.eu/en/faqs/navigating-ai-act?ref=theraceai.org). Its broader high-risk framework creates duties involving risk management, documentation, and human oversight. Member states designate competent authorities to supervise compliance. ([European Commission overview](https://digital-strategy.ec.europa.eu/en/policies/ai-act-governance-and-enforcement?ref=theraceai.org)) The remaining question is operational. Can the responsible institutions hire technical staff? Can they obtain independent evidence? Can they act while choices remain open? A right can exist without being usable. An office can exist without being capable. Participation can exist without leverage. Rights do not enforce themselves, despite their admirable work ethic in speeches. The ladder also shows why institutional capacity cannot live only in one central AI office. AI enters through procurement, public services, and ordinary workplace systems. Capacity has to appear there as well. A procurement officer must recognize lock-in. Technical staff must be able to test vendor claims. Affected people need a route into the process. This is not glamorous work. Institutional power rarely is. ## Where the Chain Is Being Built—or Broken Different institutions reveal different missing links. ### Anticipation: the ECB and FCA One sign of institutional capacity is the ability to act before a crisis determines the available choices. In July 2026, the European Central Bank directed major euro-area banks to prepare action plans for AI-enabled cyber threats by October 31\. It was not responding to one catastrophic attack. It was requiring supervised institutions to identify measures, assign responsibility, and prepare before emergency conditions arrived. ([Reuters](https://www.reuters.com/legal/litigation/ecb-tells-banks-draw-up-plans-against-ai-attacks-amid-disruption-fears-2026-07-07/?ref=theraceai.org)) The United Kingdom’s Financial Conduct Authority took a related approach through the [Mills Review](https://www.fca.org.uk/news/press-releases/fca-publishes-landmark-review-impact-ai-retail-financial-services?ref=theraceai.org). The review examined how AI could reshape consumer services, fraud, and market power. It recommended stronger coordination, expansion of the FCA’s AI Lab, and development of an AI-enabled supervisory model. Neither action guarantees that regulators will keep pace with the market. Reviews can become shelves, and shelves have absorbed more than one administrative revolution. But both examples show authority being converted into preparation. That is what anticipation contributes to the ladder: time. ### Participation: the Snohomish County Civic Assembly Public concern does not automatically become public capacity. AI decisions can be technically complex, wrapped in procurement confidentiality, and presented after substantial commitments have already been made. Ordinary public-comment processes may invite people to speak without giving them enough information or continuity to shape the decision. The [Snohomish County Civic Assembly on AI](https://www.snococivicassembly.org/?ref=theraceai.org) offers a different model. The county brought together randomly selected residents to learn about AI, deliberate over multiple sessions, and develop recommendations for how county government should use the technology. The assembly then delivered a [final report](https://www.snococivicassembly.org/final-report?ref=theraceai.org) to the County Council. The county created time for learning, access to evidence, and a route into formal government before permanent rules were settled. That is more than consultation. It is a small piece of civic infrastructure. Its weakness is equally clear. If government does not respond and connect the recommendations to continuing oversight, the assembly remains an unusually thoughtful suggestion box. The missing link would be leverage. ### Implementation: Illinois schools Governments are not only AI regulators. They are also customers, employers, and deployers. Illinois’s 2026 statewide [guidance for artificial intelligence in K–12 schools](https://www.isbe.net/Lists/News/DispForm.aspx?ID=1593&ref=theraceai.org) treats adoption as an institutional design problem rather than a question of whether teachers should use a chatbot. The guidance addresses governance, procurement, and professional learning. It also emphasizes defining the educational problem before selecting the tool. A guidance document is not implementation. Four hundred pages can supply a district with help or with a new object to place beside the strategic plan. The practical test comes next. Do districts receive usable technical assistance? Can they evaluate vendor claims? Can less-resourced districts exercise the same judgment as wealthy ones? Illinois has supplied a principle and a framework. The next step is to make them operational across hundreds of local institutions. The same pattern appears elsewhere. Workers may receive training without gaining bargaining power over how AI changes their jobs. Smaller organizations may be offered alternatives they lack the expertise to use. In both cases, formal access exceeds practical agency. ## Enabling Bureaucracy and Enclosing Bureaucracy The word *bureaucracy* usually enters technology debates as an accusation. Rules slow innovation. Agencies lack expertise. Procurement is cumbersome. Some of those criticisms are deserved. But the absence of public bureaucracy does not produce the absence of control. It often transfers control to private contracts, platform rules, and technical standards. These arrangements can be every bit as bureaucratic. They simply answer to different principals and are less likely to invite public challenge. The relevant distinction is between **enabling bureaucracy** and **enclosing bureaucracy**. Enabling bureaucracy gives many actors a stable way to act. It creates common standards, technical assistance, and reviewable decisions. Enclosing bureaucracy uses complexity to protect the system from challenge. It privileges established participants and turns compliance into an entry barrier. The same rule can do both. A demanding evaluation standard may protect the public while becoming too costly for smaller firms. A centralized procurement framework may improve contract terms while locking many agencies into one vendor. A national safety process may strengthen oversight while concentrating access among a few companies and state institutions. The choice is rarely between governance and freedom. It is more often between visible governance and governance embedded in private infrastructure. Strong institutions can support innovation when they make rules legible, create shared resources, and allow controlled experimentation. The design problem is not whether institutions should be strong. It is whether their strength remains contestable. ## What Capacity Requires Across these examples, three requirements recur. ### Authority and leverage Someone must have the standing to obtain information, set conditions, or require a remedy. Authority can come from regulation, collective representation, or purchasing power. Without it, expertise becomes advice offered to whoever already controls the decision. Voice without leverage may influence debate. It is less likely to change a contract. ### Independent knowledge and organizational competence Institutions need to understand the systems they govern without depending entirely on the provider’s description. That requires access to usable evidence, technical expertise, and organizational continuity. Transparency is not the release of more material. It is the production of information that another institution can use. Capacity also has to survive beyond one unusually capable person. It needs routines, records, and continuity. ### Time, alternatives, and remedy Contestability is strongest before commitments harden. Institutions need notice early enough to evaluate alternatives. They also need somewhere else to go: another provider, an interoperable system, or a different institutional arrangement. And they need a way to correct failure after deployment through appeal, audit, or revision. These requirements reinforce one another. Authority without knowledge can become arbitrary. Knowledge without leverage can be ignored. Alternatives without competence may exist only on paper. Capacity is not one office, one right, or one consultation. It is the operating system that allows all of them to matter. ## The Democratic Meaning of Capacity The concentration window is often described through assets: compute, capital, and models. Institutional capacity belongs on that list. A society that possesses advanced AI but lacks institutions capable of governing it has acquired capability without much agency. A government that adopts AI throughout public services but cannot evaluate vendor systems has modernized its dependency. A worker who becomes more productive without gaining information or bargaining power may receive better tools inside a more controlling organization. Some concentration of power may be necessary to build and operate systems at AI scale. The democratic question is whether that power remains answerable to institutions outside itself. Can institutions see what is happening? Can affected people intervene before choices harden? Can they obtain correction or build an alternative? These are the practical forms of contestable scale. They are also the bridge to the next article in this series. Institutional capacity to contest remains an abstract phrase unless it can be observed. The next paper will ask what indicators show that the window of contestability is opening, narrowing, closing, or reopening. Three tests will matter most: - **Visibility:** Can institutions and affected people understand how AI is being used? - **Leverage:** Can they intervene before decisions harden and require review or correction? - **Alternatives:** Can they exit, change providers, or build another arrangement? No single measure will answer the question. A country may have strong legislation and weak implementation. A regulator may possess technical expertise but little democratic accountability. A civic process may deepen public judgment without changing policy. The indicators will have to be read together. But the underlying principle is now clearer. The window of contestability does not stay open merely because alternatives remain imaginable. It stays open when institutions have the capacity to make alternatives real. ## References and Further Reading ### Institutional capacity and contestability - Albert O. Hirschman, [*Exit, Voice, and Loyalty*](https://www.hup.harvard.edu/books/9780674276604?ref=theraceai.org) (Harvard University Press, 1970). - Global Index on Responsible AI, [2026 Index and Evidence Explorer](https://www.global-index.ai/?ref=theraceai.org) and [Top 10 Takeaways](https://www.global-index.ai/takeaways?ref=theraceai.org). ### Regulatory and public-sector capacity - European Commission, [Governance and Enforcement of the AI Act](https://digital-strategy.ec.europa.eu/en/policies/ai-act-governance-and-enforcement?ref=theraceai.org). - Financial Conduct Authority, [The Mills Review: Impact of AI on Retail Financial Services](https://www.fca.org.uk/news/press-releases/fca-publishes-landmark-review-impact-ai-retail-financial-services?ref=theraceai.org), July 6, 2026. - Reuters, [“ECB tells banks to draw up plans against AI attacks amid disruption fears”](https://www.reuters.com/legal/litigation/ecb-tells-banks-draw-up-plans-against-ai-attacks-amid-disruption-fears-2026-07-07/?ref=theraceai.org), July 7, 2026. - Illinois State Board of Education, [Statewide Guidance on Artificial Intelligence in Schools](https://www.isbe.net/Lists/News/DispForm.aspx?ID=1593&ref=theraceai.org), July 9, 2026. ### Civic capacity - Snohomish County Civic Assembly on AI, [project site and final report](https://www.snococivicassembly.org/?ref=theraceai.org). ### Utility Governance Is Democracy’s Entry Point Into the AI Race URL: https://www.theraceai.org/utility-governance-is-democracys-entry-point-into-the-ai-race/ Last updated: 2026-07-28T18:08:48.000Z The AI race looks almost weightless from a screen. Its infrastructure is anything but. Behind every frontier model sits an expanding physical system: power plants, transmission lines, substations, transformers, water supplies, land, roads, and a grid that was not designed around the sudden arrival of customers asking for hundreds of megawatts at a time. That physical dependence changes the politics of AI. Technology companies can build models, buy chips, and organize computing systems through largely private decisions. They cannot, by themselves, allocate scarce grid capacity, approve land uses, assign water, or decide which customers will pay for the infrastructure their projects require. Those decisions pass through utilities, regulators, state and local governments, courts, and public processes. Sometimes the public role is substantial. Sometimes it is little more than a procedural cameo. But it exists. That makes utility governance one of democracy’s most important entry points into the AI race. Not a guarantee. An entry point. ## When Private Strategy Becomes Public Policy A large AI data center is not simply another commercial building that happens to use a great deal of electricity. Its demand may require new generation, substations, transmission upgrades, water systems, and grid equipment ordered years before the first server is switched on. Those commitments can shape utility rates, public capital plans, land use, water allocation, and the availability of infrastructure for everyone else. A project may be privately owned, but many of its consequences arrive through systems the public already governs. This is where the AI race becomes unusually contestable. At the model layer, the public often encounters decisions after companies have made them. At the infrastructure layer, expansion must pass through institutions with legal duties, formal procedures, public records, rate cases, permitting standards, and avenues for appeal. None of those mechanisms is automatically democratic. Utility proceedings can be highly technical. Local approvals can move through administrative channels that attract little notice. Agreements may remain confidential. Regulators can be overmatched, captured, or simply hurried. Still, the grid creates something scarce elsewhere in the AI economy: a place where private ambition must be translated into public obligations. ## The Covenant Inside a Utility Connection A utility connection is not merely permission to plug in. For a very large customer, it becomes a long-term allocation of capacity, cost, and risk. Who pays for new infrastructure? How much power must the customer commit to purchase? What happens if projected demand never materializes? Can the customer leave while other ratepayers remain responsible for equipment built on its behalf? The answers can be written into tariffs, interconnection agreements, special rate structures, minimum-payment commitments, cost-allocation rules, and conditions attached to public incentives. That is the hidden covenant inside the utility relationship: access to public infrastructure in exchange for enforceable commitments about who bears the cost and risk. The covenant can be weak, opaque, and tilted toward the largest customer in the room. It can also be made visible, reviewable, and durable enough to protect the broader system. Public institutions are beginning to test both possibilities. ## Public Institutions Are Starting to Write the Terms In June, the [Federal Energy Regulatory Commission](https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration?ref=theraceai.org) opened proceedings requiring all six regional grid operators under its jurisdiction to justify or reform the tariffs governing how data centers and other large electricity users connect to the transmission system. The action is important, but it should be described carefully. FERC has not imposed one national rule for large-load connections. It issued tailored show-cause orders directing each regional operator to defend its current approach or propose changes. The proceedings put cost allocation, transparency, reliability, and consumer protection formally on the table. That is already a meaningful shift. AI-scale electricity demand is no longer being treated simply as a series of private service requests. It is becoming a grid-governance problem. The [White House Ratepayer Protection Pledge](https://www.whitehouse.gov/releases/2026/03/ratepayer-protection-pledge/?ref=theraceai.org) advances a similar principle from a different direction. Participating technology companies pledge to build, buy, or bring the new power their data centers require; pay for associated delivery infrastructure; and negotiate rates requiring them to keep paying for capacity built on their behalf whether they ultimately use it or not. The principle is sound: ordinary customers should not be left financing speculative infrastructure for some of the wealthiest companies in the world. The limitation is equally clear. The pledge is voluntary. Its value will depend on whether the commitments become utility tariffs, contracts, regulatory orders, and enforceable state rules. A press event can announce a principle. Utilities and regulators still have to make it operational. Senator Ed Markey has proposed a more compulsory approach. His discussion draft of the [Protecting Communities from Data Center Impacts Act](https://www.markey.senate.gov/news/press-releases/senator-markey-releases-discussion-draft-of-legislation-to-create-a-national-framework-to-address-data-center-harm?ref=theraceai.org) would require covered data centers to obtain a federal certificate before permitting and construction. The proposal would also require them to pay for necessary grid infrastructure, fund renewable energy and storage to meet their capacity needs, reduce demand during grid stress, and meet labor standards. The proposal remains a discussion draft, not enacted law. It nevertheless marks a conceptual step: a large data center would be treated less like an ordinary private development and more like infrastructure whose scale creates public obligations before construction begins. Pennsylvania offers the most developed current example of attaching those obligations to state support. Its [Governor’s Responsible Infrastructure Development, or GRID, Standards](https://dced.pa.gov/business-assistance/data-center-resources/grid-standards/?ref=theraceai.org) condition expedited permitting, data-center tax benefits, and access to preferential tax programs on commitments in four areas: energy affordability, transparency and community engagement, workforce and economic development, and environmental protection. The standards do not automatically govern every data center in Pennsylvania. They define the terms on which a project can receive the Commonwealth’s assistance. That distinction matters. Pennsylvania is using something government already controls—the allocation of public support—to bargain for stronger public commitments. A developer seeking GRID certification must explain how it will obtain incremental power without imposing additional costs on other ratepayers, pay costs attributable to its interconnection and service, disclose projected electricity and water use, engage affected communities early enough to influence design decisions, and report on compliance before and after operations begin. This is what enabling bureaucracy can look like. The state is neither prohibiting development nor waving it through. It is trying to make speed conditional on responsibility. Whether the standards will be enforced consistently remains to be seen. But the design is important: faster approval is exchanged for clearer obligations, continuing disclosure, and a public record of compliance. ## Scarcity Is Also a Governance System AI infrastructure is also intensifying competition for the equipment needed to move electricity. [Reuters reported in July](https://www.reuters.com/business/energy/us-power-companies-scramble-secure-equipment-surging-data-center-demand-strains-2026-07-09/?ref=theraceai.org) that lead times for some generator step-up transformers had exceeded 160 weeks by early 2026\. High-voltage circuit breakers and switchgear are also under pressure. Utilities and developers are reserving production capacity years in advance, making upfront payments, and widening their supplier networks. Transformers, it turns out, do not move at software speed. This turns procurement into another layer of capacity governance. Which projects can reserve scarce equipment five years ahead? Which utilities have enough purchasing power to secure favorable production slots? Which communities wait longer for the equipment needed to replace aging systems, connect housing, or improve resilience? The market will answer some of those questions. But a market in which the largest firms can lock up critical equipment years in advance is also allocating public capacity. Scale can reproduce itself quietly: the actors with the most capital gain the best access to the equipment required to create even more capacity. ## Local Conflicts Show What Contestability Looks Like The democratic stakes become clearest when infrastructure decisions move from abstract forecasts into particular communities. In Colorado Springs, residents [appealed the administrative approval of Project Taurus](https://www.cpr.org/2026/06/24/data-center-colorado-springs-approval-appeal/?ref=theraceai.org), a proposed AI data center in an existing building. Their concerns include noise, water use, and possible effects on utility bills. The appeal moved the matter into a public hearing before the planning commission, with a further appeal to the City Council possible. The institutional fact is not simply that opposition appeared. It is that an administrative decision remained reviewable before the project became irreversible. In California’s Imperial Valley, a proposed hyperscale data center has generated disputes over environmental review, land-use procedures, electric service, and Colorado River water. The developer [sued the Imperial Irrigation District](https://www.kpbs.org/news/environment/2026/06/15/imperial-valley-data-center-developer-files-lawsuit-seeking-access-to-colorado-river-water?ref=theraceai.org) after the public utility rejected its water-service application. The City of Imperial filed a separate lawsuit challenging the environmental review, and local voters began gathering signatures for a countywide ban, as [CalMatters reported](https://calmatters.org/environment/2026/06/imperial-county-data-center/?ref=theraceai.org). The dispute is messy. That is not necessarily evidence that governance has failed. It is evidence that the project’s assumptions about water, land, environmental review, and public consent did not remain private assumptions. Virginia’s Digital Gateway project shows the procedural stakes even more starkly. The plan would have allowed as many as 37 data centers near Manassas National Battlefield. Courts voided the rezoning after finding that Prince William County had not provided the public notice required by law. When the last developer withdrew its appeal in July, [the project ended](https://www.washingtonpost.com/dc-md-va/2026/07/09/developer-virginia-data-center-project-drops-legal-appeal/?ref=theraceai.org). It would be easy to read these cases as a morality play in which communities stop data centers and democracy wins. That would be too easy. A project approved promptly under clear standards, full disclosure, fair cost allocation, and enforceable conditions can also demonstrate democratic government working well. Delay is not accountability by definition. Nor is rejection the only legitimate public outcome. The better test is whether major commitments are visible, reviewable, and contestable before they become too expensive—politically or financially—to reconsider. Can the public see who will pay? Can regulators test the demand forecast? Can communities assess water, noise, traffic, land-use, and emergency-service effects? Can conditions be enforced after the ribbon cutting? Those questions are becoming part of the AI race’s institutional architecture. ## An Entry Point, Not a Guarantee The physical dependence of AI on electricity, water, land, and public infrastructure does not give democratic institutions control over AI. It gives them leverage over the conditions under which AI expands. That leverage can be squandered. Governments can offer subsidies before understanding the costs. Utilities can negotiate special arrangements that remain opaque to other customers. Local governments can approve projects without adequate technical capacity. Public processes can become so slow and unpredictable that only the largest companies can survive them. Governance can disperse power, but badly designed governance can concentrate it too. The opportunity is to preserve contestable scale: large enough to support the infrastructure AI requires, but not so opaque or one-sided that only hyperscalers, major utilities, and well-financed developers can participate meaningfully in the decisions. That means transparent load forecasts. Clear cost-allocation rules. Take-or-pay commitments when infrastructure is built for a specific customer. Public disclosure of power and water demands. Community engagement early enough to change a project rather than merely react to it. Periodic reporting after operations begin. And institutions with enough technical capacity to distinguish a serious project from an optimistic queue position wearing a very expensive suit. Chips and models will help determine what AI can do. Utility governance will help determine who pays for its expansion, who receives scarce capacity, who carries the downside risk, and whether the public retains a meaningful voice before the system hardens around private decisions. Infrastructure-scale AI requires infrastructure-scale governance. That may be democracy’s most practical way into the race. --- ## Sources - [Federal Energy Regulatory Commission: “FERC Launches Aggressive Targeted Action to Speed Large Load Integration,” June 18, 2026](https://www.ferc.gov/news-events/news/ferc-launches-aggressive-targeted-action-speed-large-load-integration?ref=theraceai.org) - [The White House: “Ratepayer Protection Pledge,” March 4, 2026](https://www.whitehouse.gov/releases/2026/03/ratepayer-protection-pledge/?ref=theraceai.org) - [Office of Senator Edward J. Markey: “Senator Markey Releases Discussion Draft of Legislation to Create a National Framework to Address Data Center Harm,” July 13, 2026](https://www.markey.senate.gov/news/press-releases/senator-markey-releases-discussion-draft-of-legislation-to-create-a-national-framework-to-address-data-center-harm?ref=theraceai.org) - [Pennsylvania Department of Community and Economic Development: Governor’s Responsible Infrastructure Development Standards](https://dced.pa.gov/business-assistance/data-center-resources/grid-standards/?ref=theraceai.org) - [Reuters: “US power companies scramble to secure equipment as surging data center demand strains supplies,” July 9, 2026](https://www.reuters.com/business/energy/us-power-companies-scramble-secure-equipment-surging-data-center-demand-strains-2026-07-09/?ref=theraceai.org) - [Colorado Public Radio: “Residents appeal approval of data center in Colorado Springs,” June 24, 2026](https://www.cpr.org/2026/06/24/data-center-colorado-springs-approval-appeal/?ref=theraceai.org) - [KPBS: “Imperial Valley data center developer files lawsuit seeking access to Colorado River water,” June 15, 2026](https://www.kpbs.org/news/environment/2026/06/15/imperial-valley-data-center-developer-files-lawsuit-seeking-access-to-colorado-river-water?ref=theraceai.org) - [CalMatters: “Imperial County approved a massive data center. Then it changed its mind,” June 23, 2026](https://calmatters.org/environment/2026/06/imperial-county-data-center/?ref=theraceai.org) - [The Washington Post: “Massive Va. data center project won’t go forward after backer ends legal fight,” July 9, 2026](https://www.washingtonpost.com/dc-md-va/2026/07/09/developer-virginia-data-center-project-drops-legal-appeal/?ref=theraceai.org) ### States Are Starting to Build the Public Layer for AI URL: https://www.theraceai.org/states-are-starting-to-build-the-public-layer-for-ai/ Last updated: 2026-07-28T18:09:43.000Z ### *Inventories, procurement rules, training, shared tools, and limits on consequential uses are beginning to form a public framework.* No state has built a complete public infrastructure for artificial intelligence. Several are beginning to assemble the parts. The work is not arriving through one grand program. It is appearing in public inventories, procurement rules, employee training, controlled pilots, shared tools, risk reviews, and occasional decisions about what AI should not be allowed to do. Taken separately, these can look like administrative details. Together, they begin to resemble a public layer for AI. That layer matters because state governments will not experience AI only as regulators. They are also large employers, buyers of technology, holders of sensitive data, operators of public services, and sources of technical assistance for schools, local governments, and other institutions. They will use AI whether or not they build the capacity to govern it well. The early record is mixed. California, Maryland, Colorado, Pennsylvania, Illinois, New Jersey, and Washington have each developed useful parts of the framework. None has completed it. Most still rely heavily on private models, private clouds, and vendor expertise. That is not surprising. Roads were not built before the first wagon, and procurement offices do not usually get a decade of calm reflection before a new technology arrives. Still, the institutional choices being made now will matter later. A state that knows where AI is being used, tests systems before deployment, trains its workforce, shares expertise, and preserves human authority is in a stronger position than one that simply opens accounts and waits for efficiency to happen. The public layer beginning to emerge has seven parts: 1. **Visibility** 2. **Rules** 3. **Experimentation** 4. **Workforce capacity** 5. **Shared infrastructure** 6. **Contestability** 7. **Portability and independence** The first five are becoming easier to see. The sixth appears unevenly. The seventh remains weak almost everywhere. That last gap may determine whether states are building public capacity or merely becoming better-organized customers. ## 1\. Visibility: know where AI is being used A government cannot govern systems it cannot see. That makes inventories a basic piece of public AI infrastructure. [Maryland’s public AI inventory](https://ai.maryland.gov/ai-inventory?ref=theraceai.org) is one of the clearest examples. State law requires the Department of Information Technology to maintain a record of AI uses across agencies, from ordinary productivity tools to systems developed or operated for specialized public purposes. The inventory does more than produce a list. It helps the state understand which tools employees are using, where sensitive data may be involved, whether several agencies are solving the same problem separately, and which vendors are becoming embedded across government. It also gives the public some ability to see how AI is entering state operations. Washington has moved in a similar direction. Its [state AI policy and procurement guidance](https://watech.wa.gov/artificial-intelligence-resources?ref=theraceai.org) require agencies to identify and document AI-enabled technology, including through the state application-inventory process. Visibility sounds elementary. In practice, it is difficult. Employees may use public chatbots without telling their agencies. AI features can appear inside ordinary software updates. A vendor may describe a product as analytics, automation, fraud detection, or decision support rather than AI. Agencies may not know which subcontractors or embedded services are involved. An inventory therefore needs continuing intake, clear definitions, and enough authority to reach systems that did not arrive through the front door. It also needs to be useful. A spreadsheet created for compliance and left untouched is not infrastructure. The value comes from connecting the inventory to procurement, risk review, monitoring, and public accountability. Maryland is beginning to do that. Its inventory includes proposed uses as well as deployed systems, and the state’s AI team can use the information to understand demand for different tools and shape a broader platform strategy. The public layer starts by making the system visible. ## 2\. Rules: decide how AI enters government Rules are the second component. Many states now have acceptable-use policies or general statements of responsible-AI principles. The stronger frameworks connect those principles to actual decisions. [Colorado’s statewide approach](https://oit.colorado.gov/ai?ref=theraceai.org) requires state agencies to submit generative-AI uses—including tools supplied by third-party vendors—to the Governor’s Office of Information Technology for risk assessment. Colorado organizes its strategy around governance, innovation, and education, and places AI proposals inside an existing state technology-review structure. That matters because AI adoption rarely presents itself as a philosophical question. It arrives as a procurement request, a software feature, a pilot proposal, or an employee asking whether a tool is permitted. The rules must operate at that level. Who must review the system? Which uses require more scrutiny? What information must a vendor provide? What happens when the technology changes after purchase? Who remains accountable for the result? California has also moved toward a more explicit procurement and risk framework. Its [March 2026 executive order on trusted AI procurement](https://www.gov.ca.gov/2026/03/30/as-trump-rolls-back-protections-governor-newsom-signs-first-of-its-kind-executive-order-to-strengthen-ai-protections-and-responsible-use/?ref=theraceai.org) strengthened civil-rights, privacy, and risk protections while continuing the state’s effort to deploy generative AI in public services. Washington’s [generative-AI procurement guidelines](https://watech.wa.gov/sites/default/files/2024-11/Initial%20Procurement%20Guidelines%20for%20GenAI%20Final.pdf?ref=theraceai.org) likewise connect acquisition to risk assessment, data practices, testing, documentation, and continuing oversight. The important shift is from broad principle to operating rule. A framework becomes real when an agency cannot bypass it merely by calling the system a pilot. ## 3\. Experimentation: create places to learn before scaling Government needs room to experiment. It also needs a way to prevent “pilot” from becoming a permanent category in which systems operate without ordinary accountability. Maryland’s emerging [AI Innovation Lab](https://ai.maryland.gov/ai-maryland/news-and-updates?ref=theraceai.org) is intended to help agencies turn promising experiments into usable government tools. The larger Maryland strategy combines pilots, technical assistance, policy guidance, and an AI Enablement Team that agencies can consult as they evaluate possible uses. Colorado offers a useful example of structured experimentation. Its [90-day Google Gemini pilot](https://oit.colorado.gov/standards-policies-guides/guide-to-artificial-intelligence/case-study-google-gemini-pilot?ref=theraceai.org) included 150 employees across 18 agencies. Participants completed training, signed agreements, joined a learning cohort, and submitted more than 2,000 surveys about productivity, accuracy, fairness, privacy, and other concerns. The pilot produced encouraging results. More important for this argument, it produced a method. Colorado created a central information hub, recurring community-of-practice meetings, participation requirements, feedback loops, and a process for deciding whether the tool should move toward broader use. The pilot generated institutional knowledge rather than only a press release. Pennsylvania has also moved beyond a single demonstration. The Commonwealth reports that agencies have proposed and evaluated dozens of AI pilots, while some successful uses have entered ordinary operations. These include document-quality checks for benefits applications, transcription support for clemency records, and customer-service systems that rely on approved answers and escalate unfamiliar questions to human staff. ([Pennsylvania Office of Administration](https://www.pa.gov/agencies/oa/newsroom/shapiro-admin-expands-safe-responsible-ai-improve-services?ref=theraceai.org)) Experimentation becomes public capacity when governments preserve what they learn. That includes negative results. A pilot that shows a tool is unreliable, too expensive, difficult to audit, or poorly matched to the problem may be more valuable than one that produces a flattering productivity estimate. Public institutions need permission to conclude that AI was not the answer. ## 4\. Workforce capacity: teach the institution, not just the user AI training can mean many things. At its thinnest, it teaches employees how to write prompts and avoid entering confidential information. That is useful. It is not enough. Public capacity requires employees who can define problems, evaluate tools, recognize failure, redesign workflows, and participate in decisions about where AI belongs. Pennsylvania provides one of the strongest current examples of workforce-scale adoption. By April 2026, the state reported more than [3,000 employees using approved generative-AI tools across 35 agencies](https://www.pa.gov/agencies/oa/newsroom/shapiro-admin-expands-safe-responsible-ai-improve-services?ref=theraceai.org), with another 6,500 enrolled in required training. It has also created a Generative AI Governing Board and a labor-management collaboration group that gives represented employees a role in shaping adoption. That last feature deserves attention. Workers often understand the practical process better than anyone else. They know where the official workflow diverges from the real one, which exceptions matter, and what kinds of errors create harm. Excluding them from implementation wastes knowledge and makes resistance more likely. Colorado’s pilot produced a continuing AI community of practice across agencies. [Maryland offers public employees training, office hours, and a community of practice](https://ai.maryland.gov/?ref=theraceai.org) through which civil servants can exchange examples and receive help from the state AI Enablement Team. These structures help preserve institutional learning. Without them, knowledge remains scattered among a few enthusiasts, consultants, or vendor representatives. When those people leave, the institution discovers that its AI strategy had been stored largely in their calendars. Workforce capacity is not a one-time course. It is the ability to keep learning as the technology and the work change. ## 5\. Shared infrastructure: give agencies more than separate subscriptions States can also create shared tools and services. This may be one of their most important roles. Individual agencies, school districts, and local governments often lack the staff, bargaining power, and technical expertise needed to evaluate AI systems on their own. New Jersey has taken a direct approach through the [NJ AI Assistant](https://innovation.nj.gov/projects/ai-assistant/?ref=theraceai.org), a secure generative-AI platform for state employees and authorized users. The state reports that thousands of public-service professionals have used the system. New Jersey has paired the tool with training and an innovation organization intended to help agencies build implementation skills. Maryland is assembling a different version of shared infrastructure. Its strategy includes common policy guidance, an inventory, technical assistance, training, office hours, an Innovation Lab, and partnerships with the University System of Maryland. State legislation has also established an AI partnership and a public-service fellowship intended to connect public institutions with expertise and talent. ([Maryland AI legislation](https://ai.maryland.gov/ai-maryland/legislation?ref=theraceai.org)) California’s [agreement with Anthropic](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org) offers another route. State agencies—and California cities and counties—can obtain Claude at a substantial discount, along with training, technical assistance, and workflow support. That may accelerate adoption, especially for smaller public institutions. It also exposes the central tension in this entire framework. Shared public access is not necessarily shared public infrastructure. The Anthropic agreement can give agencies useful tools and expertise while concentrating implementation knowledge around one provider. The state may become a more capable buyer and still remain dependent on a vendor’s model, product roadmap, pricing, and technical architecture. The relevant test is what the public institution retains. Does it gain reusable expertise, common standards, evaluation methods, and bargaining power? Can it compare providers and move between them? Or does the shared layer become a statewide channel into one private platform? A group discount is helpful. It is not sovereignty. ## 6\. Contestability: preserve the authority to say no A public framework must do more than help government adopt AI. It must also preserve the ability of workers, residents, and public institutions to question how the technology is used. Illinois provides a particularly clear paired example. The state has produced extensive guidance for school districts and educators on AI use in teaching and learning, following legislation that required the Illinois State Board of Education to develop statewide resources. The guidance addresses practical uses, student literacy, data protection, misinformation, vendor evaluation, and the continuing role of teachers. ([Illinois education AI guidance reporting](https://capitolnewsillinois.com/news/illinois-state-board-of-education-issues-ai-guidance-written-with-help-from-ai/?ref=theraceai.org)) At the same time, Illinois enacted [Public Act 104-0565](https://ilga.gov/Legislation/BillStatus?DocNum=2909&DocTypeID=SB&GAID=18&LegId=165259&Print=1&SessionID=114&ref=theraceai.org), which prohibits AI from assigning numerical scores or qualitative ratings in teacher evaluations and bars its use for evaluation tasks requiring professional judgment. AI may support administrative work, but the consequential judgment remains with the evaluator. One measure helps institutions use AI. The other defines a boundary. Both are forms of capacity. Contestability also includes public inventories, understandable explanations, impact assessments, audit rights, incident reporting, appeal processes, and meaningful human review. It asks whether people affected by a system can see that it exists, understand its role, challenge errors, and reach someone with authority to change the outcome. This remains less developed than procurement or training. Most state frameworks are designed primarily around the needs of agencies and employees. Public participation often comes later, if it comes at all. The result can be responsible internal administration without much external leverage for the people subject to the system. That is better than ungoverned deployment. It is not yet a complete public layer. ## 7\. Portability and independence: the missing piece The final component is the least developed. Can a state move its data, prompts, workflow history, evaluations, and institutional knowledge from one AI provider to another? Can it use multiple models inside the same operating environment? Can agencies continue important services if a provider changes its terms, raises prices, restricts a use, or withdraws a product? Can public institutions inspect enough of the system to understand what they are relying on? Most current frameworks have better answers for acceptable use than for exit. California’s Anthropic agreement, Pennsylvania’s expansion from a ChatGPT Enterprise pilot, Colorado’s use of Gemini within Google Workspace, and New Jersey’s statewide assistant all create practical capability. They may also deepen dependence on particular models, clouds, or technical ecosystems. That does not make them mistakes. A state cannot wait for a perfectly open, interoperable, publicly controlled stack before improving services. Nor should every government try to build a frontier model and cloud platform from scratch. That would be an expensive route to rediscovering why specialization exists. The realistic goal is not self-sufficiency. It is retained choice. States can pursue that through contract terms, exportable records, common data formats, model-neutral interfaces, multi-vendor strategies, open standards, continuity provisions, independent evaluation, and internal staff who understand the system well enough to change it. This is where procurement, implementation capacity, and public infrastructure meet. Without portability, the first six parts of the framework can make adoption safer and more competent while leaving the underlying dependency largely untouched. The public layer becomes well governed. The foundation beneath it remains private. ## The framework is emerging unevenly No state currently demonstrates all seven elements. California has scale, procurement authority, deployments, workforce initiatives, and substantial vendor partnerships. Maryland has built one of the clearest institutional structures around inventories, policy, training, shared expertise, and experimentation. Colorado has developed a disciplined approach to review, pilots, monitoring, and cross-agency learning. Pennsylvania has moved furthest toward workforce-scale use and operational applications, while bringing employees into parts of the implementation process. Illinois shows how operational guidance can be paired with an enforceable limit on consequential use. New Jersey demonstrates the value of a shared, secure employee platform. Washington provides a strong governance and procurement architecture. These are not interchangeable models, and they should not be forced into a ranking. A national [Code for America assessment](https://codeforamerica.org/explore/government-ai-landscape-assessment/?ref=theraceai.org) found a similar uneven pattern. States are moving rapidly through governance and experimentation, but more slowly into operational scaling and systematic measurement of public value. That is a useful caution. A state can have a policy, a task force, a pilot, a training program, and a chatbot without knowing whether public services have improved. The framework should eventually produce evidence: - fewer errors; - faster or more accessible services; - better working conditions; - clearer accountability; - lower switching costs; - wider institutional competence; - more public ability to see and challenge AI use. Otherwise, the public layer risks becoming a polished administrative shell around ordinary technology procurement. ## What states can build The emerging framework does not require states to own every model or server. It asks them to build the capacity to act. That means knowing what systems are in use. Setting rules before harm occurs. Creating protected spaces for experimentation. Training employees and preserving what they learn. Sharing tools and expertise across agencies. Giving the public ways to understand and challenge consequential uses. Maintaining enough independence to change providers and direction. Those are ordinary functions of capable government. AI makes them newly urgent. The states now moving fastest are not simply “using AI.” They are beginning to build institutions around it. Whether those institutions become durable public infrastructure will depend on what happens next: whether pilots become accountable operations, whether training becomes organizational knowledge, whether shared tools produce public bargaining power, and whether states can preserve the ability to leave the platforms they are entering. The public layer is beginning to appear. It is not finished, and it does not yet own the ground beneath it. ## References and Further Reading - Code for America, [“Government AI Landscape Assessment”](https://codeforamerica.org/explore/government-ai-landscape-assessment/?ref=theraceai.org), 2026. - California Governor’s Office, [“Executive Order to Strengthen AI Protections and Responsible Use”](https://www.gov.ca.gov/2026/03/30/as-trump-rolls-back-protections-governor-newsom-signs-first-of-its-kind-executive-order-to-strengthen-ai-protections-and-responsible-use/?ref=theraceai.org), March 30, 2026. - California Governor’s Office, [“Partnership Providing Anthropic Tools to State Agencies”](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org), June 29, 2026. - Maryland AI, [“AI Inventory”](https://ai.maryland.gov/ai-inventory?ref=theraceai.org). - Maryland AI, [“Maryland’s AI Strategy”](https://ai.maryland.gov/ai-maryland/marylands-ai-strategy?ref=theraceai.org). - Maryland AI, [“News and Updates”](https://ai.maryland.gov/ai-maryland/news-and-updates?ref=theraceai.org). - Maryland AI, [“Legislation”](https://ai.maryland.gov/ai-maryland/legislation?ref=theraceai.org). - Colorado Governor’s Office of Information Technology, [“Guide to Artificial Intelligence”](https://oit.colorado.gov/ai?ref=theraceai.org). - Colorado Governor’s Office of Information Technology, [“Case Study: Google Gemini Pilot”](https://oit.colorado.gov/standards-policies-guides/guide-to-artificial-intelligence/case-study-google-gemini-pilot?ref=theraceai.org). - Pennsylvania Office of Administration, [“Shapiro Administration Expands Safe and Responsible AI Across State Government”](https://www.pa.gov/agencies/oa/newsroom/shapiro-admin-expands-safe-responsible-ai-improve-services?ref=theraceai.org), April 15, 2026. - New Jersey Innovation Authority, [“NJ AI Assistant”](https://innovation.nj.gov/projects/ai-assistant/?ref=theraceai.org). - Washington Technology Solutions, [“Artificial Intelligence Resources”](https://watech.wa.gov/artificial-intelligence-resources?ref=theraceai.org). - Washington Technology Solutions, [“Initial Procurement Guidelines for Generative AI”](https://watech.wa.gov/sites/default/files/2024-11/Initial%20Procurement%20Guidelines%20for%20GenAI%20Final.pdf?ref=theraceai.org). - Illinois General Assembly, [“SB 2909 — Teacher Evaluation and AI”](https://ilga.gov/Legislation/BillStatus?DocNum=2909&DocTypeID=SB&GAID=18&LegId=165259&Print=1&SessionID=114&ref=theraceai.org). - Capitol News Illinois, [“Illinois State Board of Education Issues AI Guidance”](https://capitolnewsillinois.com/news/illinois-state-board-of-education-issues-ai-guidance-written-with-help-from-ai/?ref=theraceai.org), July 2026. ### Buying AI Is the Easy Part URL: https://www.theraceai.org/buying-ai-is-the-easy-part/ Last updated: 2026-07-28T18:10:06.000Z *The harder question is who learns how to make it work* A local government can buy access to an AI system in an afternoon. Making that system useful may take years. That gap is easy to underestimate because access is visible. A contract is signed. A pilot is announced. Employees receive accounts. A vendor demonstrates what the system can do. The organization can now say it has AI. Whether the institution has gained much capacity is a different matter. A useful early example comes from [Clariti AI Studio](https://www.claritisoftware.com/clariti-ai-studio?ref=theraceai.org), a program built around the permitting systems used by cities and counties. Rather than beginning with a general presentation about AI, the program starts with a government’s actual permitting process and asks where the technology might help. That sounds modest. It is also close to the heart of the implementation problem. Permitting is not one task. It can involve intake, records, code review, routing, inspections, interdepartmental handoffs, public notice, legal requirements, staff judgment, applicant communication, and appeals. A routine residential permit has little in common with a major commercial development or emergency reconstruction after a disaster. Dropping a chatbot into that system does not redesign it. Someone must understand where the delays occur, which records matter, how the departments interact, what applicants need, which judgments can be supported by software, and where human authority must remain clear. That knowledge is implementation capacity. It may become one of the most important sources of power in the AI economy. ## Access, adoption, integration, transformation AI adoption is spreading quickly. Organizational transformation is not. The [first consolidated analysis from the UK Office for National Statistics](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026?ref=theraceai.org) found that reported AI use among businesses with at least 10 employees rose sharply between late 2023 and June 2026\. Yet only a small share of adopting firms described their use as extensive. Most were applying AI to improve existing operations rather than redesigning products, markets, or organizational structures. That is exactly what we should expect at this stage. Employees experiment. Departments add tools. Managers approve pilots. A few workflows become faster. The institution changes around the edges. Useful things can happen there. Drafting improves. Search becomes easier. Routine analysis takes less time. Customer-service staff receive better summaries. Software developers work more quickly. But experimentation is not integration, and integration is not transformation. These stages require different kinds of work: - **Access** means the organization can use the tool. - **Adoption** means people begin using it in ordinary tasks. - **Integration** means AI is connected to data, software, responsibilities, and workflows. - **Transformation** means the organization changes how work is divided, decisions are made, services are delivered, and value is created. The stages do not arrive automatically. Plenty of organizations will live for years in the space between adoption and integration, accumulating subscriptions while wondering why the promised productivity gains remain shy. That may be less a failure of the technology than a shortage of implementation capacity. ## The organization has to know itself AI cannot improve a process that no one understands well enough to describe. That is a larger problem than it sounds. Organizations often possess deep practical knowledge without having a clear map of how work actually moves. The official procedure may say one thing. Staff know that the real process depends on three spreadsheets, a weekly meeting, one experienced employee, and a workaround introduced during a software migration eight years ago. The workaround has since become tradition. Implementation begins by making that hidden system visible. Who performs each step? Which data are required? Where do the data live? Which decisions follow rules, and which require judgment? Where do errors enter? Which delays protect something important, and which merely survive because no one owns the whole process? Those questions are not technical preliminaries. They determine whether AI helps the institution or simply automates its confusion. Clariti’s permitting workshops are interesting because they begin there. The company is not merely offering a model. It is offering a way to examine the process into which the model might be introduced. Of course, Clariti is also selling permitting software. That does not discredit the work. It does mean the vendor arrives with its own view of the problem, its own product architecture, and a fairly understandable preference for solutions that fit inside it. Who defines the process is therefore closely connected to who defines the solution. ## Implementation knowledge can become a control point The AI industry is often described as a stack: chips, compute, cloud infrastructure, models, applications, and distribution. Implementation may become another layer. The organization that understands a customer’s workflows, data, software, compliance rules, bottlenecks, and internal politics occupies a powerful position. It knows where AI can be inserted. It also knows which models, integrations, and operating arrangements will be difficult to replace later. Several implementation models are already taking shape. ### The embedded-services model Traditional consulting and technology-services firms can place engineers and specialists inside a client organization. [TCS’s proposed corps of up to 8,900 forward-deployed engineers](https://www.reuters.com/world/india/indias-tata-consultancy-services-plans-up-8900-ai-deployment-engineers-seeks-ai-2026-07-12/?ref=theraceai.org) is one example. [Microsoft Frontier Company](https://blogs.microsoft.com/blog/2026/07/02/microsoft-frontier-company-ai-engineering-that-amplifies-and-protects-your-intelligence/?ref=theraceai.org) is another, with plans to embed 6,000 industry and engineering experts with customers. These specialists would work closely with organizations to adapt and integrate AI systems into real operations. The advantage is obvious. Most organizations lack enough internal expertise to redesign complicated workflows while evaluating fast-changing AI tools. Embedded specialists can shorten that learning curve. But where does the learning remain? If the service firm acquires the deepest understanding of the client’s processes, data connections, and model choices, the customer may become more productive without becoming much more capable on its own. The consultants leave with reusable knowledge. The institution is left with documentation—sometimes excellent documentation, sometimes a folder whose most reliable feature is that everyone agrees it exists. ### The integrated-platform model A second approach absorbs implementation into the platform itself. Products such as [ChatGPT Work](https://academy.openai.com/public/clubs/champions-ecqup/resources/chatgpt-work-lead-guide-for-exec-sponsors-2026-07-08?ref=theraceai.org) are moving beyond a standalone assistant toward an environment that gathers institutional context, connects applications, discovers tools, coordinates actions, and produces completed work. The platform does not merely answer questions. It begins to organize how tasks move across the institution. This can make implementation much easier. A unified system reduces the need to assemble separate models, connectors, permissions, and interfaces. It also makes the platform harder to replace. Once files, organizational context, workflows, permissions, tools, and task history accumulate inside one environment, switching is no longer a matter of buying a different model. The institution may have to reconstruct part of its operating memory. Convenience arrives first. Dependence usually sends a quieter announcement. ### The persistent-agent model [Claude Cowork](https://claude.com/product/cowork?ref=theraceai.org) points toward a related form. Its [remote sessions can preserve files and task history across devices](https://support.claude.com/en/articles/12138966-release-notes?ref=theraceai.org), while [scheduled tasks](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork?ref=theraceai.org) can run without the user’s device remaining online. That continuity is useful. Many real tasks extend over hours, days, or weeks. An agent that remembers what happened yesterday and can act again tomorrow is more valuable than one that starts every conversation with the institutional equivalent of “remind me what we were doing.” Persistence also creates a new kind of switching cost. The model may be replaceable in theory while the accumulated work context is not. The implementation layer begins to merge with the memory layer, the execution layer, and the user relationship. At that point, the institution is no longer buying only intelligence. It is renting continuity. ## Public agencies face a sharper version of the problem Private companies can accept some vendor dependence as a business judgment. Public agencies must consider additional questions: public records, continuity of service, legal authority, procurement rules, transparency, data protection, and the obligation to explain consequential decisions. They also face a basic capacity gap. Large technology firms can hire model engineers, data scientists, product managers, cybersecurity specialists, lawyers, and organizational-design teams. A small city, school district, public hospital, or nonprofit cannot recreate that workforce. Vendor support may therefore be essential. [California’s agreement with Anthropic](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org) offers public agencies discounted access, workforce training, technical assistance, and implementation support. [New York’s enterprise cloud services agreement with AWS](https://its.ny.gov/news/nys-office-information-technology-services-announces-enterprise-cloud-services-agreement?ref=theraceai.org) follows a similar logic, reducing procurement barriers while providing state agencies with AI, cloud, and workforce-training resources. These arrangements may help public institutions begin using AI sooner and with more support than ordinary procurement would provide. The immediate benefits may be real. Still, a central question remains: what does the institution know how to do when the vendor is no longer in the room? Can staff evaluate a new model? Can they change providers without losing workflows and institutional memory? Can they identify when a system is failing? Can they renegotiate the process rather than merely request a product adjustment? Public-sector implementation succeeds when the agency gains durable knowledge and authority, not only temporary access. ## Procurement is part of the design Procurement is often treated as the administrative step that happens after an organization decides what it wants. With AI, procurement helps determine what the organization will become. A contract can define: - who owns data and derived information; - whether logs are available; - how models are evaluated; - which integrations are permitted; - what happens when prices change; - whether the institution can use competing models; - how accumulated context can be exported; - whether staff receive training; - what documentation the vendor must provide; - how the service continues if the provider withdraws; - whether the system can be audited; - how responsibility is divided when something goes wrong. These are not boilerplate questions. They shape the architecture of dependence. A weak procurement process asks whether the product works. A stronger one asks whether the institution can continue to function, learn, and exercise judgment after the product becomes embedded. That requires technical knowledge, legal knowledge, operational knowledge, and enough bargaining power to turn those concerns into contract terms. Smaller institutions frequently lack some or all of them. This is why shared procurement capacity may become a form of civic infrastructure. States, professional associations, municipal networks, universities, and public-interest organizations can develop standard clauses, evaluation methods, vendor comparisons, and model contracts that individual institutions would struggle to produce alone. The alternative is for every small buyer to learn the same expensive lesson separately. Vendors tend to survive that arrangement quite well. ## Training is not the same as learning Workforce training is one of the most common promises attached to AI adoption. It matters. Employees need to know how systems work, where they fail, what information should not be entered, and when human review is required. But training can remain shallow. A vendor demonstrates the interface. Staff complete a course. A few enthusiastic users become local experts. The organization has checked the training box. Institutional learning goes further. It captures what worked, what failed, how the workflow changed, which risks appeared, and whether the service actually improved. It creates feedback loops. It updates policy. It changes roles. It spreads knowledge beyond the original project team. Most importantly, it preserves learning when people leave. That last part is less exciting than a model launch, which may explain why it receives fewer keynote presentations. Yet implementation capacity depends on it. An organization that repeats the same pilot every two years has adopted AI several times and learned very little. ## Who gains the capability? This is where implementation becomes part of the concentration–dispersion contest. AI can disperse capability in obvious ways. A small business gains research and marketing tools. A public employee can summarize a complicated record. A nonprofit can translate materials, draft grant applications, or analyze data that once required outside help. Those gains matter. At the same time, implementation may concentrate the knowledge needed to make AI useful. A vendor may understand the customer’s process better than the customer does. A consulting firm may carry lessons from one client to the next. A platform may accumulate institutional context and workflow history. A cloud provider may become the place where models, data, tools, security, and execution come together. The organization receives capability at the surface. Control may deepen underneath. The better outcome is capacity transfer. Outside expertise helps the institution learn enough to make informed choices, govern the system, change providers, and continue improving after the first project ends. That is harder to sell as a product because success eventually makes the customer less dependent. It is also closer to what durable implementation should mean. ## Lane 7: implementation capacity This is the central concern of **Lane 7 — Implementation Capacity** in [*The Race*](https://www.theraceai.org/). The lane asks who can turn AI capability into durable organizational competence. That includes more than technical skill. Implementation capacity combines: - the ability to define the problem; - knowledge of actual workflows; - usable and well-governed data; - procurement and contracting skill; - systems integration; - cybersecurity and legal review; - workforce training; - evaluation; - organizational redesign; - institutional learning; - the ability to switch providers or approaches. The list is long because implementation is not one thing. It is the connective tissue between a model and an institution. Without it, access produces scattered experiments. With it, organizations can decide where AI belongs, where it does not, and how to change their own operations without surrendering control of them. ## What should we look for? The early evidence suggests that implementation is still shallow across much of the economy. That is not surprising. General-purpose technologies usually arrive before organizations know how to reorganize around them. The more revealing question is where implementation knowledge accumulates during the transition. Several signals will matter: Do organizations build internal teams capable of evaluating and redesigning systems? Do vendor contracts transfer knowledge or merely provide service? Can data, workflows, and accumulated context move? Do workers participate in redesign, or receive a finished system after the consequential choices have been made? Do public agencies share tools and lessons, or negotiate alone? Does management understand how employees are already using AI, including the unofficial experimentation that rarely appears in strategy documents? The UK evidence suggests that training and incremental operational improvement are currently more common than major workforce replacement or organizational redesign. That may be encouraging. It may also mean the decisive choices are still ahead. There is still time for institutions to build capacity before today’s pilots become tomorrow’s operating systems. ## The work behind the tool Access matters. Knowing how to use AI well matters more. Even that understates the work involved. Implementation capacity is not the skill of writing better prompts or identifying a few promising use cases. It is the institutional ability to understand work, redesign processes, connect systems, train people, evaluate results, preserve learning, and change direction when necessary. Organizations will often need vendors, consultants, and platforms to help them. The consequential question is what those relationships leave behind. Does the institution gain knowledge, authority, and room to choose? Or does the implementation work settle permanently inside the provider? Buying AI is the easy part. The harder work begins after the account is activated. ## References and Further Reading - Clariti, [“Clariti AI Studio”](https://www.claritisoftware.com/clariti-ai-studio?ref=theraceai.org). - Government Technology, [“AI Studio Offers Local Government Guidance to Modernize Permits”](https://www.govtech.com/artificial-intelligence/ai-studio-offers-local-govt-guidance-to-modernize-permits?ref=theraceai.org). - UK Office for National Statistics, [“Artificial Intelligence in UK Businesses: 2023 to 2026”](https://www.ons.gov.uk/businessindustryandtrade/business/businessservices/articles/artificialintelligenceinukbusinesses/2023to2026?ref=theraceai.org), July 20, 2026. - Office of Governor Gavin Newsom, [“Governor Newsom Announces a First-of-Its-Kind Partnership, Providing Anthropic Tools to State Agencies and Improving Services for Californians”](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org), June 29, 2026. - New York State Office of Information Technology Services, [“NYS Office of Information Technology Services Announces Enterprise Cloud Services Agreement with Amazon Web Services”](https://its.ny.gov/news/nys-office-information-technology-services-announces-enterprise-cloud-services-agreement?ref=theraceai.org), June 15, 2026. - Reuters, [“India’s Tata Consultancy Services Plans Up to 8,900 AI Deployment Engineers, Seeks AI Acquisitions”](https://www.reuters.com/world/india/indias-tata-consultancy-services-plans-up-8900-ai-deployment-engineers-seeks-ai-2026-07-12/?ref=theraceai.org), July 12, 2026. - Microsoft, [“Microsoft Frontier Company: AI Engineering That Amplifies and Protects Your Intelligence”](https://blogs.microsoft.com/blog/2026/07/02/microsoft-frontier-company-ai-engineering-that-amplifies-and-protects-your-intelligence/?ref=theraceai.org), July 2, 2026. - OpenAI Academy, [“ChatGPT Work: Lead Guide for Executive Sponsors”](https://academy.openai.com/public/clubs/champions-ecqup/resources/chatgpt-work-lead-guide-for-exec-sponsors-2026-07-08?ref=theraceai.org), July 8, 2026. - Anthropic, [“Claude Cowork”](https://claude.com/product/cowork?ref=theraceai.org). - Anthropic Help Center, [“Release Notes”](https://support.claude.com/en/articles/12138966-release-notes?ref=theraceai.org) and [“Schedule Recurring Tasks in Claude Cowork”](https://support.claude.com/en/articles/13854387-schedule-recurring-tasks-in-claude-cowork?ref=theraceai.org), July 2026. ### Who Gets Access to Frontier AI? URL: https://www.theraceai.org/who-gets-access-to-frontier-ai/ Last updated: 2026-07-22T20:17:05.000Z *Governments are beginning to decide which countries, companies, and institutions can use the most capable models—and under what conditions.* The most important frontier AI decision may no longer be whether a model is released. It may be who is allowed to use it. That question came into view during [the U.S. government’s recent dispute with Anthropic over its Fable 5 and Mythos 5 models](https://www.anthropic.com/news/redeploying-fable-5?ref=theraceai.org). On June 12, the government directed Anthropic to suspend access by foreign nationals, including foreign-national employees inside the United States. Anthropic said it could not comply cleanly without temporarily disabling the models more broadly. The restriction was then partially loosened. [Mythos became available to a group of “trusted” U.S. organizations](https://www.reuters.com/technology/us-releases-anthropic-model-mythos-some-us-companies-semafor-reports-2026-06-26/?ref=theraceai.org). Less than three weeks after the original intervention, [the government lifted the broader export controls after Anthropic added safeguards and began working with federal agencies and technology partners on common standards for testing and addressing jailbreaks](https://www.anthropic.com/news/redeploying-fable-5?ref=theraceai.org). Restriction. Trusted access. Added safeguards. Wider restoration. That sequence may have been an improvised response to a genuine cybersecurity concern. It may also have revealed the outline of something more durable. Frontier AI access is becoming a governed category. Advanced models may increasingly be released, delayed, restricted, restored, previewed for approved recipients, or made available only under specified safeguards. Access may depend on nationality, geography, institutional status, beneficial ownership, security practices, government relationships, or participation in an approved technical ecosystem. We are moving beyond export control applied to a new product. A system for allocating advanced capability is beginning to take shape. National-security review may be necessary. The difficulty is what happens when the rules remain opaque, discretionary, unevenly applied, or negotiated privately with a handful of frontier labs. If there is going to be a frontier switch, democracy needs to know who writes the access list, what standards they use, and how those decisions can be examined. ## From chips to models to access The United States has been moving toward this problem in stages. The first stage focused on hardware. Beginning in 2022, [the Commerce Department used export controls to restrict China’s access to advanced computing chips and semiconductor-manufacturing equipment](https://www.bis.gov/press-release/commerce-implements-new-export-controls-advanced-computing-semiconductor-manufacturing-items-peoples?ref=theraceai.org). The logic was straightforward: if cutting-edge AI depends on scarce compute, control over the inputs can slow an adversary’s progress. The second stage focused on visibility. [President Biden’s 2023 AI executive order](https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence?ref=theraceai.org) created reporting requirements for companies developing the most powerful dual-use foundation models. The government wanted a clearer view of what frontier labs were building, how those systems were being tested, and whether they were adequately protected. Then came partnership. Frontier AI companies became increasingly connected to defense, intelligence, cybersecurity, and critical-infrastructure work. [Anthropic’s Claude Gov models for U.S. national-security customers](https://www.anthropic.com/news/claude-gov-models-for-u-s-national-security-customers?ref=theraceai.org) are one example. The federal government was no longer standing outside the industry as regulator alone. It was also becoming a customer, evaluator, collaborator, and operational user. We now appear to be entering a fourth stage: direct management of access. [The June 2026 White House order on advanced AI innovation and security](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/?ref=theraceai.org) did not create mandatory licensing or formal preclearance for model releases. It established a voluntary framework under which the government can receive early access to some advanced systems and develop classified benchmarks for national-security capabilities. That may sound modest. In Washington, “voluntary” can cover a fair amount of institutional territory. What matters is the direction. The government is moving closer to the release decision itself. [OpenAI’s GPT-5.6 preview](https://openai.com/index/previewing-gpt-5-6-sol/?ref=theraceai.org) made the emerging structure more visible. OpenAI said it planned broad availability, but began with a limited preview for a small group of trusted partners whose participation had been shared with the government. This was not an export-control order. It was another example of frontier access being staged through government consultation and approved recipients. The emerging pattern is no longer simply “released” or “not released.” It includes: - early government access; - restricted foreign access; - trusted-user access; - staged release; - access conditioned on safeguards; - approved-recipient lists; - restoration after review. Each category may be sensible. Together, they may become the architecture through which frontier capability is distributed. ## The security case is real It would be a mistake to treat every restriction as illegitimate gatekeeping. Frontier AI can create genuine national-security risks. Models may improve vulnerability discovery, exploit development, cyber defense, biological research, intelligence analysis, weapons design, persuasion, surveillance, and military planning. Cybersecurity is the clearest immediate example because the same capability can serve defenders and attackers. A system that helps a security team identify a serious vulnerability may help a hostile actor find it first. If a model materially changes the speed or scale of offensive cyber operations, release cannot be treated as an ordinary product decision. Government also has responsibilities that private firms do not. It must protect military systems, intelligence sources, public infrastructure, and the broader population. A state that cannot independently evaluate frontier capabilities is not exercising democratic capacity. It is simply trusting the companies that built them. Some secrecy is unavoidable. The government should not publish a detailed explanation of how a dangerous model can be exploited, nor should every security finding become a convenient roadmap for malicious use. But secrecy about technical vulnerabilities is different from secrecy about the structure of decision-making. The public may not need to know every classified finding. It does need to know the broad rules: when access can be restricted, who makes the decision, how comparable models are treated, how long restrictions last, and what oversight exists. Necessary secrecy should protect the vulnerability. It should not hide the system of authority. ## From oversight to allocation Why does the Anthropic episode matter beyond one company and two model names? Because it shows how quickly safety review can become access allocation. The original order restricted a category of users. The next step created a trusted group. Wider access returned only after safeguards were accepted. Any one of those moves may have been reasonable. Together they establish something new: the government is helping decide who receives advanced capability and under what conditions. The same pattern is appearing beyond individual model releases. In July, [the Commerce Department approved the United Arab Emirates government and selected companies to receive advanced computing items, including AI chips and servers, without individual licenses](https://www.bis.gov/press-release/department-commerce-eases-export-controls-uae?ref=theraceai.org). The decision was tied to a broader U.S.–UAE artificial-intelligence cooperation framework and to UAE commitments to invest in U.S. AI infrastructure. There may be sound strategic reasons for this arrangement. The United States has always differentiated among allies, partners, competitors, and adversaries. Still, the case makes the shift visible. Advanced AI access is becoming an affirmative instrument of state policy. Governments are deciding not only who must be denied capability, but who will be helped to obtain it. That distinction matters. A strategic-access system can include chips, models, cloud services, training, technical assistance, data, and implementation support. Access can reward alliances, strengthen supply chains, encourage investment, define trusted ecosystems, and build geopolitical dependence. Frontier AI is beginning to resemble other forms of strategic infrastructure: nuclear technology, advanced weapons systems, intelligence sharing, satellite access, and secure communications. It is not simply sold into a global market. Increasingly, it is distributed through relationships. ## China is moving toward its own strategic-access system This is not only an American development. China has spent years objecting to U.S. restrictions on advanced chips and technology transfer. It has promoted [open models and lower-cost access](https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/?ref=theraceai.org) as an alternative to a U.S.-dominated frontier ecosystem. Now China is also beginning to treat its leading AI capability as a strategic asset. [Reuters reported in early July that Chinese authorities were considering restrictions on overseas access to the country’s most advanced models](https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/?ref=theraceai.org). [More recent reporting suggests that the discussion may extend beyond finished services to model weights, training data, semiconductor designs, and overseas transactions that could transfer advanced technology](https://www.reuters.com/world/asia-pacific/china-considers-tighter-export-controls-ai-models-chips-ft-reports-2026-07-21/?ref=theraceai.org). The proposals were still under consideration when this essay was written, and their eventual scope remains uncertain. Even so, the underlying tension is becoming easier to see. As Chinese models become more capable, Beijing faces much the same tradeoff as Washington. Openness can expand influence and accelerate adoption. It can also transfer strategic capability to competitors. The U.S.–China AI contest may therefore resist the neat story in which one side is closed and the other open. Both countries may combine openness and restriction. Both may make some capabilities broadly available while reserving others for trusted users, domestic institutions, allies, or approved partners. Both may decide that the frontier should travel—but not entirely on its own. The more revealing difference may be the institutional form through which access is provided. ## Access can be offered as well as withheld [China’s creation of the World Artificial Intelligence Cooperation Organization with 28 other countries](https://www.reuters.com/world/china/twenty-nine-countries-sign-agreement-establish-global-ai-cooperation-body-2026-07-16/?ref=theraceai.org) adds another dimension. The organization is presented as a mechanism for international governance, training, public applications, and technical cooperation. [President Xi Jinping also announced thousands of training opportunities for developing countries, regional cooperation centers, and the extension of Chinese AI-enabled meteorological services to additional countries](https://apnews.com/article/df4cfc7e1b260e765b5449b6d71a48e5?ref=theraceai.org). This may expand meaningful capacity in places that have been largely excluded from frontier AI development. Many countries do not possess the compute, technical staff, regulatory expertise, or bargaining power needed to participate on equal terms in the AI economy. Training, applications, and technical assistance could help narrow that gap. Yet access offered through an international institution is never entirely neutral. It comes with standards, technical dependencies, governance assumptions, preferred platforms, and relationships to the state that provides it. The United States increasingly organizes advanced capability around trusted partners, export rules, commercial providers, investment arrangements, and secure supply chains. China is offering a competing model built around development assistance, state-to-state cooperation, public applications, and institutions aimed at the Global South. This raises an important question: who gets to define the terms on which other countries gain AI capacity? A country may receive valuable new capability while becoming dependent on the models, infrastructure, standards, security rules, and political relationships of the provider. Access can disperse capability. It can also organize a sphere of influence. ## A new public-private control layer The usual AI concentration story focuses on private firms. A small number of companies control much of the frontier-model market, advanced compute, cloud infrastructure, developer distribution, and user access. That concentration remains important. Frontier access creates another path: growing interdependence between the national-security state and a small group of AI labs and infrastructure firms. Government needs the companies because they possess the models, engineers, compute, and operational knowledge. The companies need government because it controls export authority, security classifications, procurement, defense relationships, alliance policy, and access to sensitive markets. The result may be a powerful public-private control layer. This is not quite government takeover and not quite private monopoly. It is a system in which a small number of agencies and firms become regulators, customers, suppliers, evaluators, and strategic partners—sometimes all at once. The roles begin to blur. A lab may help define the benchmark against which its own model is evaluated. A government agency may restrict access while depending on the same company for national-security applications. A cloud provider may enforce a state access rule through proprietary account systems. A trusted-user designation may reflect security concerns, company safeguards, and commercial relationships at the same time. Some cooperation is inevitable. Frontier systems are too technically complex, too expensive, and too strategically important for government and industry to operate in separate rooms. The democratic concern begins when cooperation becomes unreviewable fusion. That is why frontier access belongs primarily in **Lane 2 — National Security and Frontier Control**, while also reaching into **Lane 5 — Platforms, Distribution, and Separations**. The switch is not held by government alone or industry alone. It sits in the relationship between them. ## Legitimate control versus discretionary gatekeeping The goal cannot be to eliminate government oversight or require immediate public access to every powerful model. The harder task is distinguishing legitimate security control from discretionary gatekeeping. Legitimate control begins with a clearly defined risk. It uses rules that apply to comparable capabilities, not merely to one disfavored company or user. It limits secrecy to what must remain secret. It includes independent technical capacity, institutional oversight, and a route for reconsideration. Restrictions last only as long as the risk justifies them. Discretionary gatekeeping has a different feel. The trigger is unclear. The standard changes from one company or recipient to another. “Trusted” status is undefined. Access depends on private negotiation or political favor. Restrictions lack a clear duration. Affected parties cannot understand the decision or seek review. Temporary emergency measures quietly settle in and begin receiving mail. The line will not always be obvious. Several tests can help make it visible. ### Are the triggers clear? The public should know the broad capability thresholds or risk categories that can produce review, staged release, or restriction. Technical details may remain classified. The existence and scope of the authority should not. ### Are similar cases treated similarly? A rule aimed at dangerous capability should follow the capability across companies, countries, and delivery systems. If one model is restricted while a comparable system is ignored, the process begins to look less like safety governance and more like selective control. ### Is there independent capacity? Government cannot rely entirely on company claims, and one agency should not be the sole source of judgment. Independent evaluators, inspectors general, congressional committees, courts where appropriate, and technically capable public institutions provide checks on both industry and executive discretion. ### Can decisions be revisited? Emergency action may need to be fast. Long-term control should not be indefinite by default. Access decisions need review dates, procedures for presenting new safeguards, and a clear path from restricted to staged to general access when conditions change. ### Is access separable from political and commercial favoritism? Trusted access should not quietly become preferred access for favored companies, allies, investors, or government partners without generally applicable criteria. National-security policy will always distinguish among countries. The basis for those distinctions should still be visible enough to evaluate. These requirements do not make the decisions easy. They make the system more contestable. ## The frontier switch is becoming a system The Anthropic dispute may eventually look like an awkward early episode. The government acted quickly. The company responded broadly. Trusted access was created. Safeguards were added. Restrictions were lifted. That could be evidence of a system learning. But the larger pattern already extends beyond one model. The United States is developing early-access arrangements, trusted-recipient rules, export controls, alliance-based compute access, and public-private security partnerships. China is considering its own restrictions while building international institutions through which technical capacity, training, and applications can be distributed. Frontier AI access is becoming a form of statecraft. That does not make every restriction wrong or every cooperation program suspect. It means access itself has become a site of power. The old question was who could build the frontier. The emerging question is who can cross it. If the rules remain improvised, secret, and concentrated among a few firms and agencies, the frontier may become a closed public-private control layer. If the rules are capability-based, reviewable, consistent, and institutionally accountable, governments may be able to address real security risks without turning access into permanent discretionary privilege. More than one hand now rests on the frontier switch. Democracy needs to know how the system works. ## References and Further Reading ### Historical foundation - U.S. Department of Commerce, Bureau of Industry and Security, [“Commerce Implements New Export Controls on Advanced Computing and Semiconductor Manufacturing Items to the People’s Republic of China”](https://www.bis.gov/press-release/commerce-implements-new-export-controls-advanced-computing-semiconductor-manufacturing-items-peoples?ref=theraceai.org), October 7, 2022. - Executive Order 14110, [“Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence”](https://www.federalregister.gov/documents/2023/11/01/2023-24283/safe-secure-and-trustworthy-development-and-use-of-artificial-intelligence?ref=theraceai.org), October 30, 2023. - Anthropic, [“Claude Gov Models for U.S. National Security Customers”](https://www.anthropic.com/news/claude-gov-models-for-u-s-national-security-customers?ref=theraceai.org), June 2025. ### United States frontier review and access - White House, [“Promoting Advanced Artificial Intelligence Innovation and Security”](https://www.whitehouse.gov/presidential-actions/2026/06/promoting-advanced-artificial-intelligence-innovation-and-security/?ref=theraceai.org), June 2, 2026. - Anthropic, [“Redeploying Fable 5”](https://www.anthropic.com/news/redeploying-fable-5?ref=theraceai.org), June 30, 2026\. Includes Anthropic’s account of the June 12 restriction, the trusted-access stage, safeguard changes, and the lifting of the controls. - Reuters, [“U.S. Allows Anthropic to Release Mythos AI to ‘Trusted’ U.S. Organizations”](https://www.reuters.com/technology/us-releases-anthropic-model-mythos-some-us-companies-semafor-reports-2026-06-26/?ref=theraceai.org), June 26, 2026. - Reuters, [“U.S. Removes Curbs on Anthropic’s Latest Fable and Mythos AI Models”](https://www.reuters.com/business/us-lift-export-controls-anthropics-fable-ai-model-tuesday-source-says-2026-06-30/?ref=theraceai.org), June 30, 2026. - OpenAI, [“Previewing GPT-5.6 Sol: A Next-Generation Model”](https://openai.com/index/previewing-gpt-5-6-sol/?ref=theraceai.org), June 26, 2026. ### Strategic compute and recipient access - U.S. Department of Commerce, Bureau of Industry and Security, [“Department of Commerce Eases Export Controls for UAE”](https://www.bis.gov/press-release/department-commerce-eases-export-controls-uae?ref=theraceai.org), July 10, 2026. - U.S. Department of Commerce, Bureau of Industry and Security, [Export Administration Regulations, Part 740: License Exceptions](https://www.bis.gov/regulations/ear/740?ref=theraceai.org). ### China and international access - Reuters, [“Beijing Is Looking at Curbing Overseas Access to China’s Top AI Models”](https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/?ref=theraceai.org), July 7, 2026. - Reuters, [“China Considers Tighter Export Controls on AI Models and Chips”](https://www.reuters.com/world/asia-pacific/china-considers-tighter-export-controls-ai-models-chips-ft-reports-2026-07-21/?ref=theraceai.org), July 21, 2026. - Reuters, [“Twenty-Nine Countries Sign Agreement to Establish Global AI Cooperation Body”](https://www.reuters.com/world/china/twenty-nine-countries-sign-agreement-establish-global-ai-cooperation-body-2026-07-16/?ref=theraceai.org), July 16, 2026. - Associated Press, [“China’s Xi Calls for More Global Efforts to Guide AI”](https://apnews.com/article/df4cfc7e1b260e765b5449b6d71a48e5?ref=theraceai.org), July 17, 2026. ### What Would It Take to Make AI Public Infrastructure? URL: https://www.theraceai.org/what-would-it-take-to-make-ai-public-infrastructure/ Last updated: 2026-07-22T19:02:55.000Z *Public models and public compute are only the beginning. The harder task is building the institutions that make them usable.* The most interesting public AI experiment may not be another government buying access to a private chatbot. It may be an attempt to build something governments, universities, researchers, businesses, and civic institutions can actually use without renting every important layer from the same small group of companies. The European Union is now testing that possibility. The [European Commission selected the EUROPA consortium](https://digital-strategy.ec.europa.eu/en/news/commission-selects-europa-consortium-winner-frontier-ai-grand-challenge-project-build-european-open?ref=theraceai.org), led by the Italian company Domyn, to develop an open-source frontier model covering all 24 official EU languages. The project will receive public support and access to European supercomputing capacity through EuroHPC. That combination matters. A publicly supported model is one thing. A public computing system that helps institutions train, adapt, and use it is something more. Europe is trying to assemble both. The ambition is easy to recognize. Much of frontier AI has developed through private laboratories, private cloud platforms, private application programming interfaces, and private release decisions. Governments participate as regulators, customers, funders, and national-security partners, but the ordinary arrangement remains familiar: private firms build the most capable systems, and everyone else buys access. EUROPA points toward a different possibility. Could AI become public infrastructure—something governments, universities, businesses, and civic institutions can use, inspect, adapt, and govern rather than merely consume? It is a promising question. It is also a harder one than “Is the model open?” ## Public funding is not the same as public capacity Governments can spend public money on AI in several very different ways. They can buy discounted licenses from a private provider. They can pay consultants to integrate proprietary systems. They can subsidize cloud access. They can fund research. They can build shared computing facilities. They can support open models. They can create internal technical teams, testing laboratories, procurement standards, training programs, and public-interest applications. All of those may be useful. They do not produce the same kind of capacity. California’s recent [agreement with Anthropic](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org) illustrates the distinction. State agencies, cities, and counties can purchase Claude at a substantial discount, with workforce training, technical assistance, and implementation support from Anthropic. That may help public employees adopt AI more quickly. It may improve services. Training and technical support are certainly better than dropping a new tool into an agency and hoping enthusiasm will take care of the rest. Still, the arrangement begins with access to a vendor’s system. The model, product roadmap, pricing structure, technical architecture, and much of the implementation knowledge remain with the provider. Public employees may become more capable users while the institution itself becomes more dependent on one company’s tools and expertise. That is not necessarily a bad bargain. It is simply a different bargain from building public infrastructure. A discounted subscription is not a public utility. A training session is not an internal engineering team. A helpful vendor is still a vendor. Public capacity begins to emerge when institutions retain knowledge, bargaining power, data control, operational continuity, and realistic alternatives after the initial agreement ends. ## What does a public institution actually control? That question is more useful than asking whether a project carries a public label. Can the institution inspect the system closely enough to understand how it behaves? Can it adapt the model or application to its own needs? Can it move its data, workflows, and accumulated knowledge to another provider? Can it maintain the system if the original vendor changes direction? Can it train its own staff rather than relying indefinitely on outside specialists? Can it compare multiple models and replace one without rebuilding the entire service? Can the public understand who is responsible when the system fails? These are not glamorous questions. Infrastructure rarely becomes interesting by announcing its maintenance plan. Yet maintenance, staffing, interoperability, and governance determine whether a public investment produces durable capability or a temporary demonstration. The same is true of AI. A model can be publicly funded and still be practically inaccessible. A platform can be inexpensive and still create lock-in. A government can “have AI” while possessing very little authority over the system it now depends upon. The civic value lies in the surrounding institution. ## Open licensing is necessary—and insufficient EUROPA’s proposed openness is important. So is Portugal’s [AMALIA model](https://www.reuters.com/business/finance/portugal-launches-first-open-source-ai-model-joining-europes-sovereignty-push-2026-07-01/?ref=theraceai.org), developed by Portuguese universities and research institutions with government support and European recovery funding. AMALIA is intended to strengthen AI capability in European Portuguese, a language variant often poorly served by larger models. The model, training data, and source code are being released openly. Early applications include public services, education, museums, research, business, and decision support. That is a meaningful dispersion signal. Language is itself a form of infrastructure. When important digital systems poorly represent a country’s language, law, institutions, and cultural context, that country becomes more dependent on systems built elsewhere for different users and assumptions. Open models can reduce that dependence. They allow independent evaluation, local adaptation, specialized use, and in some cases self-hosting. But openness comes in degrees. A model may have downloadable weights while withholding parts of the training data or construction process. Its license may restrict commercial or institutional use. Running it may require hardware available only to large organizations. Updates may depend on the original developer. Security fixes may be irregular. Documentation may be thin. Smaller institutions may discover that the model is open in roughly the same sense that a grand public building is open when the doors are unlocked but no road leads to it. A download link is not an implementation strategy. For an open model to function as civic infrastructure, institutions need affordable computing, deployment tools, documentation, evaluation, cybersecurity, maintenance, technical assistance, and people who understand both the technology and the public problem being addressed. That is why the surrounding ecosystem matters as much as the license. ## Public compute changes the equation Europe’s broader [AI Factories program](https://www.eurohpc-ju.europa.eu/ai-factories%5Fen?ref=theraceai.org) is potentially more important than any one model. EuroHPC now offers computing resources and technical support to startups, small and medium-sized businesses, researchers, public-sector users, and other European institutions. Some access pathways are free for qualifying smaller firms and include assistance from the host AI Factory, not merely time on a machine. That begins to address one of the central weaknesses of open AI. Open models may separate users from the original model provider. They do not automatically separate them from concentrated compute, cloud hosting, specialized engineering, or implementation expertise. Public computing infrastructure can create another route. A university may be able to train or adapt a model without negotiating with a hyperscale cloud provider. A startup may test a product without surrendering much of its early capital to compute costs. A public agency may evaluate several models before committing to one. Researchers may study systems that would otherwise remain inaccessible. That is real capacity. Even so, a supercomputer is not a civic institution. High-performance computing facilities can be technically impressive and administratively forbidding. Access rules, application procedures, scheduling, software environments, staffing, and technical support will determine who can use them in practice. A public resource that only the already sophisticated can navigate may widen the field slightly while leaving most smaller institutions exactly where they started. The test is use. Do municipalities, schools, hospitals, nonprofits, regional businesses, universities, and public-interest researchers gain practical access? Can they move from experiment to operation? Does knowledge spread beyond a small circle of elite technical institutions? Public compute matters most when it becomes usable compute. ## Sovereignty is not automatically civic capacity Ukraine provides a more urgent version of the same problem. Ukraine’s Ministry of Digital Transformation and Kyivstar are developing a national model based on Google’s open-weight Gemma family. The [project was announced in late 2025](https://www.veon.com/newsroom/press-releases/kyivstar-and-ukrainian-ministry-of-digital-transformation-select-google-gemma-as-the-foundation-for-ukraines-national-llm?ref=theraceai.org), and Ukrainian officials have since emphasized the need for systems that can eventually operate without continuing provider control. The model is intended for government services, private enterprises, and military use, with a planned release in autumn 2026\. ([Reuters](https://www.reuters.com/legal/litigation/ukraine-pick-ai-models-operated-without-provider-control-official-says-2026-07-07/?ref=theraceai.org)) The security rationale is obvious. A country at war has strong reasons not to place essential government and military functions entirely inside remote systems controlled by foreign providers. Local operation can improve resilience. It can protect sensitive data. It can create expertise that remains in the country. It may allow institutions to continue functioning when commercial access, connectivity, or geopolitical relationships change. But “sovereign AI” can describe several different arrangements. It may mean a nationally operated public capability. It may mean a telecom company running a system under government partnership. It may mean a domestic platform protected from foreign competition. It may mean military control. It may mean little more than national branding around technology whose crucial components still come from elsewhere. Sovereignty answers the question, “Which country controls the system?” Civic infrastructure asks a second question: “Who inside the country can use it, shape it, question it, and benefit from it?” Those answers can diverge. A national model may reduce dependence on a foreign company while concentrating authority inside the state or a favored domestic provider. That may still be justified in wartime. It should not be confused with broad public capacity. ## Frontier AI may not be the most useful public AI There is another possibility worth taking seriously. Perhaps civic AI infrastructure will not be built mainly around frontier models. The phrase “public AI” can summon an image of a government-funded competitor to OpenAI, Anthropic, Google, or China’s leading laboratories. That may be part of the story. EUROPA is explicitly testing whether Europe can build a public-supported frontier alternative. But most public institutions do not need a frontier model for most tasks. A city may need a reliable system that helps staff compare zoning provisions, summarize public comments, search ordinances, or translate service information. A state legislature may need tools for bill comparison, legal research, fiscal analysis, and public access to legislative history. A school district may need a shared system for evaluating vendors, protecting student data, and helping teachers adapt materials. A nonprofit may need benefits navigation, grant research, case management, or multilingual communication. These systems do not need to win a benchmark competition. They need to be accurate enough, secure enough, affordable enough, understandable enough, and governed well enough to earn a place in public work. Smaller tools may be easier to audit. They may run on less expensive hardware. They can be designed around a defined institutional purpose. They may allow the people who understand the public problem to participate in development rather than receiving a general-purpose system after the important choices have already been made. There is a certain irony in building a model capable of explaining nearly everything and then discovering that a county still needs help finding the correct version of its procurement manual. Civic infrastructure is often unheroic. That is part of its value. ## Maryland shows what the institutional layer can look like Maryland offers a useful example of a different approach. The state maintains a public [inventory of AI systems and uses](https://ai.maryland.gov/ai-inventory?ref=theraceai.org), tying that inventory to existing technology-governance processes. It has also launched an [AI Innovation Lab](https://ai.maryland.gov/ai-maryland/news-and-updates?ref=theraceai.org) intended to give agencies a controlled environment for testing and prototyping systems before moving them into ordinary operations. Neither step creates an independent frontier model. Together, however, they begin to build institutional capacity. The inventory makes government use more visible. The lab gives agencies a place to experiment without immediately committing a system to public service. Shared tools, standards, staff, and lessons can reduce the need for every department to begin from zero—or to accept the first vendor proposal that arrives with a polished demonstration. This may be closer to what civic infrastructure looks like in practice. Not one giant public model. A network of shared capabilities: - public compute; - open or replaceable models; - testing environments; - technical assistance; - procurement expertise; - inventories and disclosure; - evaluation capacity; - trained public employees; - common security and data standards; - institutions that preserve learning across projects. The pieces reinforce one another. Remove enough of them, and the institution returns to dependence. ## Lane 6: building the machinery to act This is the concern at the center of **Lane 6 — Civic Infrastructure and Public Capacity** in *The Race*. Civic infrastructure is the part of public infrastructure that gives institutions and communities the practical capacity to understand, use, govern, and contest AI. Governance is not only the power to regulate someone else’s system. It is also the capacity to understand a technology, define public problems, build alternatives, negotiate with providers, operate useful tools, and challenge systems that do not serve the public well. That capacity can exist in governments, universities, libraries, schools, public media, nonprofit organizations, professional associations, civic-technology groups, and community institutions. It will not look the same everywhere. A national government may need secure compute and a locally operated model. A city may need shared procurement and technical support. A university may need research access. A small nonprofit may need a trusted intermediary that can evaluate products and provide low-cost tools. A community may need enough knowledge and institutional support to decide whether an AI system should be used at all. The common thread is retained capability. People and institutions should emerge from adoption with more knowledge and more room to act—not simply with another subscription. ## What would count as success? EUROPA, AMALIA, the European AI Factories, Ukraine’s national model, California’s vendor partnership, and Maryland’s institutional approach are not versions of the same project. That is precisely why they are useful. They show several routes through which public institutions are trying to gain AI capability: - build a public-supported open frontier model; - create a national language model; - provide shared compute and technical support; - operate a sovereign system for resilience; - accelerate adoption through a private vendor; - build inventories, sandboxes, and reusable government expertise. Each route creates different strengths and dependencies. The evidence is still early. Announcements are plentiful. Durable public capacity is harder to demonstrate. A convincing civic-infrastructure model would show several things over time. Smaller institutions would actually use it. Staff would gain transferable skills. Models and vendors could be replaced without losing the service. Public data and accumulated institutional knowledge would remain under public control. Independent evaluation would be possible. The system would have maintenance, funding, security, and governance arrangements that survive the original launch. Most important, public institutions would retain the ability to define the problems they are trying to solve. That last point is easy to lose. A powerful general-purpose system arrives with its own categories, workflows, measures of success, and convenient use cases. Institutions under pressure may reorganize around what the tool can readily do. Civic infrastructure should work the other way around: the public purpose comes first, and the technology is selected or built to serve it. ## A public option for capacity Private AI systems will remain important. Governments and civic institutions will continue to buy commercial products, and often should. Building every model and application internally would be expensive, slow, and occasionally a fine way to recreate software that already exists. The alternative to dependence is not self-sufficiency. It is capacity. A capable institution can buy a private service without surrendering judgment. It can use an open model without mistaking openness for usability. It can share infrastructure where scale is useful and preserve local authority where context matters. It can change providers. It can learn. EUROPA is worth watching because it combines public compute with an open model and a stated intention to make advanced capability available beyond a small group of frontier firms. Whether it becomes civic infrastructure will depend on what surrounds the model. Who can use it? Who can maintain it? Who can adapt it? Who learns from the work? Who remains in control when the first grant, contract, or burst of political attention ends? Those questions are less dramatic than a frontier-model launch. They are also where public capacity is built. AI becomes civic infrastructure when institutions can do more than gain access to it. They must be able to make it their own. ## References and Further Reading - European Commission, [“Commission Selects EUROPA Consortium as the Winner of the Frontier AI Grand Challenge”](https://digital-strategy.ec.europa.eu/en/news/commission-selects-europa-consortium-winner-frontier-ai-grand-challenge-project-build-european-open?ref=theraceai.org). - EuroHPC Joint Undertaking, [“AI Factories”](https://www.eurohpc-ju.europa.eu/ai-factories%5Fen?ref=theraceai.org). - EuroHPC Joint Undertaking, [“AI Factories Access Modes”](https://www.eurohpc-ju.europa.eu/ai-factories/ai-factories-access-modes%5Fen?ref=theraceai.org). - Afonso Simplício et al., [“AMALIA Technical Report: A Fully Open Source Large Language Model for European Portuguese”](https://arxiv.org/abs/2603.26511?ref=theraceai.org). - Reuters, [“Portugal Launches First Open-Source AI Model, Joining Europe’s Sovereignty Push”](https://www.reuters.com/business/finance/portugal-launches-first-open-source-ai-model-joining-europes-sovereignty-push-2026-07-01/?ref=theraceai.org). - VEON, [“Kyivstar and Ukrainian Ministry of Digital Transformation Select Google Gemma as the Foundation for Ukraine’s National LLM”](https://www.veon.com/newsroom/press-releases/kyivstar-and-ukrainian-ministry-of-digital-transformation-select-google-gemma-as-the-foundation-for-ukraines-national-llm?ref=theraceai.org). - Reuters, [“Ukraine to Pick AI Models Operated Without Provider Control, Official Says”](https://www.reuters.com/legal/litigation/ukraine-pick-ai-models-operated-without-provider-control-official-says-2026-07-07/?ref=theraceai.org). - Office of Governor Gavin Newsom, [“Governor Newsom Announces a First-of-Its-Kind Partnership Providing Anthropic Tools to State Agencies”](https://www.gov.ca.gov/2026/06/29/governor-newsom-announces-a-first-of-its-kind-partnership-providing-anthropic-tools-to-state-agencies-and-improving-services-for-californians/?ref=theraceai.org). - Maryland AI, [“AI Inventory”](https://ai.maryland.gov/ai-inventory?ref=theraceai.org). - Maryland AI, [“News and Updates”](https://ai.maryland.gov/ai-maryland/news-and-updates?ref=theraceai.org). ### When the Doorway Becomes the Operating Layer URL: https://www.theraceai.org/when-the-doorway-becomes-the-operating-layer-2/ Last updated: 2026-07-21T20:38:28.000Z ## How convenience can become gatekeeping in the AI race The most consequential AI platform may not be the one with the best model. It may be the one that becomes the way through everything else. For most of its history, Google Search was a doorway. You went there to find something somewhere else: a website, a product, a restaurant, a news story, a government form, a research paper, a video, a map, or a person. Google was already powerful because doorways decide what is visible. But the user still generally moved outward into the web. AI assistants point toward a different role. A doorway helps you go somewhere. An operating layer helps decide what you see, what you compare, what you trust, what you buy, what you write, what you schedule, what you create, and what you do next. That shift is underway across search, browsers, phones, workplace software, shopping, payments, travel, public services, and professional tools. An AI system can answer the question, narrow the choices, recommend the provider, complete the form, make the reservation, initiate the purchase, and remember enough about the user to do it more easily the next time. The benefit is obvious. So is the reason people will use it. The concern is not that these systems will be useless. It is that they may be useful enough to become the default path through which people and institutions reach the rest of the digital world. If that happens, convenience becomes more than a product feature. It becomes a form of power. ## Convenience can concentrate power precisely because it works We often imagine concentrated power as something imposed against the user’s interest. Sometimes it is. But digital concentration frequently arrives through usefulness. A platform saves time. It reduces friction. It learns preferences. It keeps everything in one place. It makes the next action easier than the alternatives. People adopt it for good reasons. Then the tool becomes a habit. The habit becomes a default. The default becomes infrastructure. Once that happens, the platform no longer merely serves user choices. It helps shape them. This is not a conspiracy theory. It is a theory of ordinary power. The easiest path becomes the path most people take. The most visible option becomes the option most people consider. The recommendation becomes the shortlist. The summarized answer becomes the frame. The platform’s preferred next step becomes the natural next step. The user may still be free in a narrow sense. Another browser can be opened. Another assistant can be consulted. A different retailer can be visited. The original sources can be found. But most people, most of the time, follow an easy path that works well enough. That is why defaults matter. The [UK Competition and Markets Authority’s review of online choice architecture](https://www.gov.uk/government/publications/online-choice-architecture-how-digital-design-can-harm-competition-and-consumers/evidence-review-of-online-choice-architecture-and-consumer-and-competition-harm?ref=theraceai.org) describes how digital design affects consumer decisions and competition. AI makes the default more subtle and potentially more powerful. The old default was often a checkbox, a preinstalled application, or the first result on a page. The new default may be a sentence: > Here is the answer. Or: > This is the best option. Or: > I can take care of that for you. The question is not whether convenience is bad. It is not. The question is whether convenience remains contestable. ## From information gatekeeper to action gatekeeper Traditional search was already a form of mediation. It crawled, indexed, and ranked information from across the web. Ranking determined which sources, businesses, and ideas were likely to be seen. AI search and AI agents move closer to the decision itself. Instead of showing where to find an answer, the system can produce the answer. Instead of presenting several sources, it can synthesize them. Instead of listing products, it can select the ones worth considering. Instead of pointing toward a retailer or service provider, it can route the user into a transaction. Instead of helping with one task, it can carry context from one task into the next. This makes the interface more useful. It also makes the underlying choices less visible. A search engine shapes visibility. An AI operating layer can shape visibility, judgment, and action. That is a different degree of power. The shift is especially clear in commerce. Google’s [Universal Commerce Protocol](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/?ref=theraceai.org) is designed as an open, vendor-neutral standard for connecting AI surfaces with merchants through the shopping process. Mastercard has developed [Agent Pay](https://www.mastercard.com/us/en/news-and-trends/stories/2026/mastercard-agentic-commerce-vision.html?ref=theraceai.org) and [Verifiable Intent](https://www.mastercard.com/us/en/news-and-trends/stories/2026/verifiable-intent.html?ref=theraceai.org) to support transactions initiated by agents under user authorization. The [FIDO Alliance](https://fidoalliance.org/fido-alliance-to-develop-standards-for-trusted-ai-agent-interactions/?ref=theraceai.org) is developing standards for authentication and trusted agent interactions. The [National Institute of Standards and Technology](https://www.nist.gov/artificial-intelligence/ai-agent-standards-initiative?ref=theraceai.org) has launched an initiative focused on secure and interoperable AI agents. These are constructive developments. Agents that act need common protocols, clear authorization, secure identity, and auditable transactions. They also show how much of the market may move inside the operating layer. When an assistant can interpret intent, choose a source, rank providers, verify identity, authorize payment, and complete a transaction, it is no longer simply helping the user navigate the market. It is helping constitute the market the user experiences. ## We have seen versions of this before The technology is new. The pattern is not. ### Microsoft and the browser The clearest digital parallel is Microsoft Windows and Internet Explorer. In the 1990s, Microsoft controlled the dominant personal-computer operating system. That mattered because the operating system was the environment through which users encountered software. The [U.S. government’s Microsoft case](https://www.justice.gov/atr/us-v-microsoft-proposed-findings-fact-0?ref=theraceai.org) focused heavily on whether Microsoft used control of Windows to protect its position from Netscape Navigator, which could become a competing platform for applications. The issue was not whether Internet Explorer was useful. It was whether control of one critical layer could be used to shape the next layer of competition. That is the durable lesson: > Control of a default layer can shape the future of adjacent layers. AI assistants may not be operating systems in the old technical sense. But they may become operating layers in the practical sense: the environment through which people encounter information, software, services, commerce, and decisions. ### Search and visibility Search itself provides a second example. Search engines dispersed access to information while concentrating control over visibility. A student, journalist, voter, patient, or small business could find material that would previously have been difficult to locate. At the same time, placement in search results became critical to whether a source or business would be found at all. AI intensifies that pattern because the system may no longer simply rank sources. It may absorb them into an answer. For a publisher, expert, merchant, civic organization, or public agency, exclusion from an AI answer may eventually matter as much as poor placement in search results. Yet the exclusion may be harder to observe because the user sees a finished response rather than a ranked page. ### Distribution as power Alfred Chandler’s history of railroads and modern distribution systems adds a broader parallel. Distribution is not merely a support function. The actor that controls customer access, routing, logistics, pricing information, and transaction flow can shape the practical market in which producers and consumers meet. AI assistants may become a new distribution layer. A user may ask an assistant to find a contractor, compare insurance policies, book travel, choose software, evaluate a doctor, order groceries, summarize news, or route a public-service request. The provider may still provide the service. The customer may still authorize the purchase. But the operating layer can determine who finds whom, on what terms, through which ranking, and under whose rules. New distribution systems often broaden access at first. AI assistants may help smaller firms and ordinary users navigate complexity. But once the distribution layer becomes necessary, the terms of access become a governance question. ## Lane 5: who controls the path? This is the central concern of **Lane 5 — Platforms, Distribution, and Separations** in *The Race*. The lane asks whether AI is developing as an open and interoperable layer or as an enclosed operating system controlled by a few intermediaries. AI can disperse capability at the user level while concentrating control at the platform level. A small business may gain access to research, writing, coding, marketing, and administrative capabilities that it could not previously afford. A worker may become more productive. A local government may deliver a service more quickly. A consumer may save hours of searching and comparison. Those are real gains. But the path to those gains may run through a tightly integrated stack controlled by a small number of firms. One company may supply the cloud infrastructure, model, assistant, browser, operating system, workplace suite, identity layer, payment connection, advertising system, merchant tools, and user interface. That produces a familiar but easy-to-miss pattern: > AI can disperse capability while concentrating the path. The market may look more accessible because more people can do more things. Yet the infrastructure through which they do those things may become less contestable. ## Open standards are not the same as contestable markets The growth of open protocols is one of the most encouraging developments in the agentic ecosystem. Shared standards can allow different systems to communicate. They can reduce the cost of integration. They can prevent every merchant, developer, and institution from having to build a separate connection to every assistant. They can support authentication, delegation, revocation, and auditability. That is valuable. But technical openness does not by itself produce a contestable market. An open protocol may specify how an assistant communicates with a merchant while leaving the dominant platform in control of the commercially meaningful pathway: the default assistant, ranking, visibility, data collection, identity, payment, checkout, and the accumulated user relationship. A standard can be open while the path to the user remains closed. This is not an argument against open standards. It is an argument for being precise about what they accomplish. An **open protocol** means a technical interface is documented and available for others to use. A **contestable pathway** means users, merchants, developers, publishers, and institutions can realistically choose, enter, switch, inspect, appeal, and exit. Those are not the same thing. A dominant platform may sincerely support interoperability because a common standard expands the market. At the same time, it may retain enormous power through defaults, distribution, ranking, identity, data, payment, and the user relationship. Open rails can still run through a closed station. ## The Separations Principle The best response is not to prohibit integration or insist that every AI function be provided by a different company. Integration creates real benefits. Users do not want to rebuild every task by connecting a dozen incompatible services. Institutions need systems that work reliably. Security and accountability sometimes require coordinated design. The point is not separation for its own sake. The point is to identify combinations of control that make the system difficult to challenge. That is the **Separations Principle**: > Control of one important layer should not automatically confer control over every adjacent layer needed to reach users, conduct transactions, or contest decisions. The relevant layers include models, cloud infrastructure, applications, browsers, assistants, data, identity, ranking, merchant access, payments, standards, evaluation, and appeals. The danger grows when one actor can use strength in one layer to determine the rules of another. A company that controls the default assistant should not automatically determine which merchants can be reached, which payment rail must be used, how its own products are ranked, and whether a user can carry accumulated context to a competitor. A platform that indexes publishers should not automatically be able to condition ordinary search visibility on permission to use their work in AI summaries. A provider that supplies the model should not be the only party able to evaluate its performance or reconstruct what its agent did. Separations can be structural, technical, contractual, or procedural. They do not all require breaking companies apart. The principle can be expressed through interoperability rules, non-discrimination requirements, independent auditing, procurement terms, choice screens, data portability, or the right to use one platform function without accepting another. The key is to prevent convenience at one layer from becoming unavoidable dependence across the entire stack. ## What real contestability would require A contestable operating layer would preserve much of the convenience while giving users and institutions meaningful alternatives and recourse. Several protections matter. ### Choice of assistant Users and institutions should be able to choose the assistant that sits on major operating surfaces—browsers, phones, workplace suites, public-service portals, and enterprise systems—rather than having the platform’s own assistant become unavoidable by default. Choice must be practical, not merely theoretical. An alternative assistant needs comparable access to authorized data and functions. ### Portability of memory and context AI lock-in may come less from files than from accumulated context: preferences, saved instructions, task history, workflows, organizational knowledge, permissions, and integrations. If that context cannot move, switching assistants may mean starting over. Users and institutions need workable ways to transfer or selectively export what an assistant has learned about them. ### Interoperability without mandatory bundling An assistant should be able to work across services when the user authorizes it. A merchant or developer should not have to buy the rest of a platform’s stack merely to participate in its agent ecosystem. Shared protocols help, but interoperability must include fair access and reasonable commercial terms. ### Transparency and appeal Users should know when a recommendation is sponsored, when the platform benefits from a transaction, and when its own services receive preferred treatment. Merchants, publishers, and developers need a way to understand consequential exclusion or demotion and to challenge it when appropriate. This does not require publication of every model weight or ranking formula. It does require enough visibility and process to make patterns reviewable. ### Scoped authority and action logs An agent that acts for a user should operate under clear, limited, and revocable authority. When the action matters, the agent should leave a usable record: what it did, what authority it relied upon, and where responsibility lies if something goes wrong. ### Independent evaluation The firm that builds the operating layer should not be the only institution able to determine whether it is fair, secure, or accurate. Researchers, auditors, regulators, and affected parties need ways to examine patterns of exclusion, self-preferencing, manipulation, error, and market impact. ### Public procurement that preserves exit Governments, schools, hospitals, and other public institutions should not acquire convenience at the cost of permanent dependence. Procurement should address portability, audit rights, data control, model and vendor switching, access to logs, continuity of service, and the ability to separate useful applications from the rest of a vendor’s stack. Public institutions should be able to leave without losing their institutional memory. ## The goal is contestable convenience It would be easy to tell this story as a warning against Google, Microsoft, Apple, Amazon, OpenAI, Anthropic, or any other large platform. That would be too narrow. Different companies will control different parts of the emerging stack. Some will compete vigorously. Open models, smaller systems, new protocols, and specialist providers may keep important parts of the market fluid. Large platforms will also create genuine value and may be best positioned to deliver some services securely and at scale. The issue is not which firm happens to lead today. It is the architecture of the market that is forming. The practical test is whether users and institutions retain meaningful choice, portability, transparency, independent review, and the ability to leave. If they do, powerful assistants can remain useful without becoming unchallengeable. If they do not, technical openness may coexist with deep commercial dependence. The web may remain formally open while the practical way through it becomes increasingly concentrated. That is why the operating layer deserves attention now, while its standards, defaults, permissions, and business relationships are still being established. The democratic goal is not a world without powerful AI assistants. It is a world in which their power can be challenged. Convenience should be widespread. It should also be contestable. ## References and Further Reading - U.S. Department of Justice, [*United States v. Microsoft: Proposed Findings of Fact*](https://www.justice.gov/atr/us-v-microsoft-proposed-findings-fact-0?ref=theraceai.org). - UK Competition and Markets Authority, [*Evidence Review of Online Choice Architecture and Consumer and Competition Harm*](https://www.gov.uk/government/publications/online-choice-architecture-how-digital-design-can-harm-competition-and-consumers/evidence-review-of-online-choice-architecture-and-consumer-and-competition-harm?ref=theraceai.org). - Tarleton Gillespie, [“The Relevance of Algorithms”](https://culturedigitally.org/2012/11/the-relevance-of-algorithms/?ref=theraceai.org). - Taina Bucher, [“Want to Be on the Top? Algorithmic Power and the Threat of Invisibility on Facebook”](https://journals.sagepub.com/doi/abs/10.1177/1461444812440159?ref=theraceai.org). - Alfred D. Chandler Jr., *The Visible Hand: The Managerial Revolution in American Business*. - Google Developers Blog, [“Under the Hood: Universal Commerce Protocol”](https://developers.googleblog.com/under-the-hood-universal-commerce-protocol-ucp/?ref=theraceai.org). - National Institute of Standards and Technology, [AI Agent Standards Initiative](https://www.nist.gov/artificial-intelligence/ai-agent-standards-initiative?ref=theraceai.org). - FIDO Alliance, [“FIDO Alliance to Develop Standards for Trusted AI Agent Interactions”](https://fidoalliance.org/fido-alliance-to-develop-standards-for-trusted-ai-agent-interactions/?ref=theraceai.org). - Mastercard, [“How Verifiable Intent Builds Trust in Agentic AI Commerce”](https://www.mastercard.com/us/en/news-and-trends/stories/2026/verifiable-intent.html?ref=theraceai.org). ### The AI Concentration Window: Why We May Overreact to AI's First Shocks and Underreact to the Power Structures Forming Beneath Them URL: https://www.theraceai.org/the-ai-concentration-window/ Last updated: 2026-07-25T20:51:01.000Z New technologies are hardest to understand when they first arrive. They appear first as products, workplace experiments, and public controversies. Only later do they become infrastructure, routines, and dependencies. That creates what we might call the **AI time horizon problem**. People naturally judge a powerful technology by what they can see. What they see first are the shocks: job disruption, synthetic misinformation, and data centers placing new demands on land, power, and water. Those concerns are real. They are also incomplete. The first wave tells us that something important has arrived. It does not tell us where society will end up. That distinction matters for *The Race*. The central question of *The Race* is whether AI will concentrate power or disperse capability. But that question cannot be answered by looking only at the most visible disruptions of the present. The more consequential issue may be whether today’s temporary advantages are beginning to harden into tomorrow’s infrastructure. AI may be overhyped in the short run and underestimated in the long run. Early products may disappoint. Productivity gains may be uneven. Many pilots will fail. Some companies will discover that announcing transformation is easier than producing it. This is not exactly a new corporate tradition. Yet beneath the noise, more durable arrangements may already be forming. A small number of firms may gain control over the infrastructure needed to build AI, the platforms through which organizations adopt it, and the interfaces through which people use it. Public agencies may build critical services around proprietary systems. Workers may be reorganized around machine-readable workflows. Standards and procurement practices may settle before most people recognize that they were choices. The period between technological arrival and broad institutional adaptation is not simply a waiting room. It is a **concentration window**. It is also a **window of contestability**: a period in which the basic arrangements remain fluid enough to be shaped, separated, reopened, or governed differently. The window may be longer than panic suggests. It may be shorter than complacency assumes. ## Why the First Wave Is So Easy to Misread The best-known version of the time horizon problem is often called [**Amara’s Law**](https://www.computer.org/publications/tech-news/trends/amaras-law-and-tech-future?ref=theraceai.org), usually attributed to futurist Roy Amara: we tend to overestimate the effect of a technology in the short run and underestimate its effect in the long run. The phrase is useful, though it is better understood as a warning about attention than as a law of nature. In the short run, new technologies are easy to overstate because they are novel, vivid, and surrounded by hype. Their most dramatic examples travel fastest. A chatbot passes a professional exam. A company announces an AI-related layoff. A fake image goes viral. These events are concrete. They can be shown, shared, feared, and debated. Long-term transformation is harder to see. It is slower, more distributed, and often hidden inside institutions. It appears in procurement rules, software defaults, and utility investments. It is rarely one dramatic event. It is a thousand quiet decisions. Public anxiety also has its own logic. Amy Orben describes the recurring pattern as the [“Sisyphean cycle of technology panics”](https://journals.sagepub.com/doi/10.1177/1745691620919372?ref=theraceai.org). The problem is not concern itself. A democratic society should be concerned about powerful technologies. The problem is concern that becomes too narrow or too immediate to guide useful institutional action. Risk-perception research helps explain why AI’s first shocks feel so large. [Paul Slovic’s work on the perception of risk](https://socsci2.ucsd.edu/~aronatas/project/academic/risk%20slovic.pdf?ref=theraceai.org) shows that people respond not only to probability, but also to dread, lack of control, and unfamiliarity. AI has all three. It is opaque. It can feel uncontrollable. It is being deployed by powerful institutions that many people do not fully trust. Public anxiety may therefore be registering something important: decisions affecting people’s lives are moving into systems they cannot easily understand, challenge, or control. The difficulty is that visible harm and structural transformation move on different clocks. The first clock measures shock. The second measures the gradual settlement of infrastructure, defaults, and power. A sound democratic response has to watch both. Panic focuses only on the first. Complacency notices that some early promises disappoint and assumes the second clock has stopped. It has not. ## The Lag Is Where the System Gets Built Some technologies improve a particular task. General-purpose technologies reshape many forms of activity. Economists Timothy Bresnahan and Manuel Trajtenberg described general-purpose technologies as pervasive technologies that improve over time and generate complementary innovation across sectors. Elhanan Helpman and Trajtenberg later explored how growth based on [general-purpose technologies](https://www.nber.org/system/files/working%5Fpapers/w4854/w4854.pdf?ref=theraceai.org) depends on waves of improvement, diffusion, and adaptation rather than one moment of invention. AI increasingly fits that description. It can be used across business, government, education, medicine, science, and ordinary personal work. But a general-purpose technology does not transform society merely by being invented or widely available. Its larger effects appear as the surrounding system changes with it: infrastructure is built, organizations redesign work, and laws and standards adapt. Electrification is the classic example. Electric motors existed before factories realized large productivity gains. Many firms initially installed them inside factories still organized around steam-era layouts. The larger gains came later, when production itself was reorganized around electric power. [Paul David’s comparison between the electric dynamo and the computer](https://www.jstor.org/stable/2006600?ref=theraceai.org) shows why transformative technologies can appear long before their effects become visible in economic statistics. The breakthrough exists, but the surrounding system has not caught up. The computer age produced the same puzzle. Computers spread through business and government for years before the expected gains appeared at scale. The [Federal Reserve Bank of Richmond’s discussion of the productivity paradox](https://www.richmondfed.org/publications/research/economic%5Fbrief/2024/eb%5F24-25?ref=theraceai.org) revisits Robert Solow’s famous observation that computers seemed to be everywhere except in the productivity statistics. Erik Brynjolfsson, Daniel Rock, and Chad Syverson argue that AI may be following this pattern. Their work on the [Productivity J-Curve](https://www.aeaweb.org/articles?id=10.1257%2Fmac.20180386&ref=theraceai.org) explains why the investments needed to reorganize institutions can initially look like costs while measurable gains appear later. The conventional lesson is patience. Powerful technologies take time. The institutional lesson is to examine what fills that time. Firms choose vendors. Agencies sign contracts. Workers are reorganized. These are not preliminary details surrounding the transformation. They are the transformation taking institutional form. ## Three Ways Concentration Forms The complementary changes required to make AI productive are expensive, difficult, and unevenly distributed. At the same time, pressure to show results grows. Investors expect returns. Executives promise transformation. Governments announce modernization. Integrated systems and powerful vendors offer an attractive answer. They can supply infrastructure, expertise, and implementation support that many organizations cannot readily provide for themselves. That may accelerate adoption. It may also consolidate control before the benefits of AI are broadly distributed. Three mechanisms are especially important. ### 1\. Control of scarce inputs Frontier AI depends on capital, compute, energy, and specialized talent. Actors that can afford a long period of experimentation gain time to learn. Those that cannot may become customers, acquisition targets, or dependent users. Control over scarce inputs can therefore shape which forms of AI are built, by whom, and under what terms. Scale is not inherently undemocratic. Some undertakings genuinely require it. The issue is whether infrastructure becomes a platform others can enter and bargain with—or a gate through which every later actor must pass. ### 2\. Control of adoption pathways Most organizations will not build AI systems from the ground up. They will adopt them through existing cloud providers, workplace platforms, and implementation partners. That is often rational. It can also deepen dependency. An organization may purchase an AI tool while lacking the expertise to evaluate it, integrate it, or bargain effectively with the vendor. The provider may then control the model, the interface, and the terms of exit. Information markets are especially prone to these dynamics because of network effects, switching costs, and lock-in. Carl Shapiro and Hal Varian’s [*Information Rules*](https://www.inforules.com/?ref=theraceai.org) remains a useful guide to these forces. W. Brian Arthur’s work on increasing returns and path dependence explains how an early advantage can reinforce itself as users and complementary services gather around one system. The first familiar doorway can become the operating layer. Systems often become durable not because anyone chose them as permanent infrastructure, but because everyone chose them as the easiest next step. ### 3\. Control inside organizations AI can give workers and institutions powerful new capabilities. It can also make work more explicit, standardized, and measurable. Tasks become data. Judgment becomes workflow. Performance becomes easier to monitor and rank. The worker may become more productive while losing discretion. A public agency may improve service while becoming dependent on a vendor it cannot independently evaluate. A smaller firm may gain access to advanced tools while becoming unable to leave the platform that supplies them. This is where the **Separations Principle** becomes useful. The central concern is not whether every part of the AI stack is controlled by a different actor. It is whether control over infrastructure, applications, and evaluation becomes bundled tightly enough that one actor can determine access, terms, and exit. Taken together, these mechanisms show how a productivity lag can become a concentration window. Organizations are not merely waiting for AI to improve. They are choosing the institutional arrangements through which improvement will be delivered. ## Public Value Does Not Prove Structural Openness Tim Wu’s *The Master Switch* offers an important caution. Wu describes a recurring pattern in information industries: open experimentation gives way to commercialization, consolidation, and eventual enclosure. This is not a law of nature. It is a pattern worth remembering. The AT&T case is useful because the Bell System was not a simple villain. A single integrated telephone system promised reliability, universality, and technical excellence. Bell Labs produced real innovation. Government depended on the system, protected it, and eventually challenged it. The system worked in visible ways. It also controlled access and compatibility. Public value and concentrated control can coexist for a long time. The internet provides a more recent version of the same ambiguity. Open protocols coexist with dominant platforms. Low-cost publishing coexists with concentrated control over search and discovery. Broad participation coexists with centralized cloud and advertising systems. AI is likely to follow a similarly mixed pattern. Millions of people can already use powerful tools. Small teams can build products faster. Workers can perform tasks that once required more time or more people. That is real diffusion. But control may still be concentrating at deeper layers: compute, platforms, and standards. The useful question is therefore not whether AI is open or closed in the abstract. It is: > At which layers is AI dispersing capability, and at which layers is it concentrating control? Broad user access is evidence of diffusion. It is not proof of structural openness. ## The Window of Contestability The concentration window is not destiny. Information industries can reopen. New technical architectures can lower barriers. Public policy can change market structure. Open-weight models may reopen parts of the stack. Smaller models may reduce dependence on massive centralized infrastructure. Procurement and interoperability rules may make it easier for institutions to change providers. But these possibilities do not realize themselves. The **window of contestability** is the period after a powerful technology emerges but before its dominant platforms, standards, and institutional dependencies become difficult to reverse. During this window, democratic institutions have more leverage. Public policy can shape procurement. Workers can gain a voice in implementation. Communities can negotiate infrastructure impacts before commitments harden. After the window narrows, action remains possible. But it becomes more expensive. Once systems are embedded, changing them is costly. Once users depend on a platform, exit becomes difficult. Once public services are built around proprietary infrastructure, oversight weakens. The challenge is not to oppose scale. Large systems will be necessary in many parts of AI. The challenge is to build **contestable scale**: systems large enough to perform important functions but open enough to be challenged, reviewed, and redirected. That requires more than consumer choice. It requires institutional capacity to contest. ## What *The Race* Should Watch *The Race* is not a project about whether AI is simply good or bad. It is a project about direction. The early evidence will always be mixed. Some developments will expand access. Others will deepen dependency. Some public uses will build civic capacity. Others will outsource public judgment to private systems. The task is not to predict the future with false precision. It is to watch the hardening process while it is still happening. Three questions matter most: - Where is AI dispersing usable capability? - Where is control accumulating across infrastructure, platforms, and institutions? - Can affected actors review decisions, change providers, or build alternatives? AI is moving on two clocks. The first measures visible disruption. The second measures the gradual settlement of defaults, dependencies, and power. The danger is not only that we will fear AI too much. It is that we will notice the deeper hardening too late. If the infrastructure and platforms surrounding AI are controlled by a narrow set of firms and state partners, AI may spread widely while power concentrates deeply. If those systems remain portable, accountable, and open to alternatives, the same period could become a dispersion window. That is the race. ## References and Further Reading ### Time horizons, technology panics, and risk - Patrick Lin, [“Amara’s Law and Its Place in the Future of Tech,”](https://www.computer.org/publications/tech-news/trends/amaras-law-and-tech-future?ref=theraceai.org) *IEEE Computer Society*. - Amy Orben, [“The Sisyphean Cycle of Technology Panics,”](https://journals.sagepub.com/doi/10.1177/1745691620919372?ref=theraceai.org) *Perspectives on Psychological Science* 15, no. 5 (2020): 1143–1157. - Paul Slovic, [“Perception of Risk,”](https://socsci2.ucsd.edu/~aronatas/project/academic/risk%20slovic.pdf?ref=theraceai.org) *Science* 236, no. 4799 (1987): 280–285. ### General-purpose technologies and productivity lags - Paul A. David, [“The Dynamo and the Computer,”](https://www.jstor.org/stable/2006600?ref=theraceai.org) *American Economic Review* 80, no. 2 (1990): 355–361. - Elhanan Helpman and Manuel Trajtenberg, [“Growth Based on General Purpose Technologies,”](https://www.nber.org/system/files/working%5Fpapers/w4854/w4854.pdf?ref=theraceai.org) NBER Working Paper No. 4854 (1994). - Erik Brynjolfsson, Daniel Rock, and Chad Syverson, [“The Productivity J-Curve,”](https://www.aeaweb.org/articles?id=10.1257%2Fmac.20180386&ref=theraceai.org) *American Economic Journal: Macroeconomics* 13, no. 1 (2021): 333–372. ### Information industries and AI market structure - Tim Wu, [*The Master Switch*](https://scholarship.law.columbia.edu/books/176/?ref=theraceai.org) (2010). - Carl Shapiro and Hal R. Varian, [*Information Rules*](https://www.inforules.com/?ref=theraceai.org) (1998). - OECD, [*Competition in Artificial Intelligence Infrastructure*](https://www.oecd.org/en/publications/competition-in-artificial-intelligence-infrastructure%5F623d1874-en.html?ref=theraceai.org) (2025). ### Coming soon URL: https://www.theraceai.org/coming-soon/ Last updated: 2026-05-16T02:32:21.000Z This is The Race: Building Democracy in the Age of AI, a brand new site by Mark Howard that's just getting started. Things will be up and running here shortly, but you can [subscribe](#/portal/) in the meantime if you'd like to stay up to date and receive emails when new content is published!