When AI Infrastructure Can Shop for Its Regulator

AI may become utility-like before we decide whether to govern it that way.

That does not mean AI is the same as electricity, water, sewer, or roads. But the comparison is becoming harder to avoid. AI is moving toward pervasive use across business, education, government, health care, research, shopping, public administration, and daily personal life. If that continues, AI will not feel like a luxury product or a single technology tool. It will feel more like a basic operating layer for modern work and public life.

That creates a governance problem. Utility-like systems do not matter only to the company that builds them or the customer that buys them. They shape shared costs, access, reliability, public capacity, and community bargaining power. They depend on physical networks, scarce resources, and long-term infrastructure choices.

The risk is not only that AI infrastructure will be privately owned. It is that infrastructure with utility-like effects will be governed as if it were ordinary private development.

That is the signal emerging from the data-center fights now spreading across the country.

The federal government is reportedly preparing to let the Federal Data Center Enhancement Act expire. That law was not written for private AI hyperscale facilities. It applied to federal data centers. But the categories it covered — energy efficiency, sustainability, resilience, security, and operational standards — are very close to the questions local governments are now asking about AI data centers.

Meanwhile, some local governments are moving toward stronger oversight. Seattle adopted a one-year moratorium on new large data centers while it studies infrastructure, environmental, economic, and public-health impacts. Charlotte adopted a shorter pause to study water, energy, noise, zoning, and environmental concerns.

Other communities are moving differently. In Colorado Springs, Project Taurus, a proposed AI data-center conversion at the former Intel facility, received administrative approval despite significant local opposition.

These cases are not identical. Federal data centers are not the same as private AI data centers. A city moratorium is not the same as a federal operating standard. An administrative approval is not the same as a subsidy.

But together they show the emerging patchwork.

Some governments are stepping back. Some are pausing. Some are approving. Some are trying to invent new rules. Others are relying on planning systems built for a different kind of development.

Infrastructure-scale projects need infrastructure-scale governance. But right now, AI compute is often being governed as a series of separate real-estate, procurement, utility, and economic-development decisions.

That patchwork creates a concentration risk. Large AI infrastructure players do not need every jurisdiction to say yes. They only need enough jurisdictions to say yes, or to move quickly, or to offer cheaper power, weaker conditions, larger subsidies, or less public resistance.

That is governance arbitrage.

The point is not that data centers are bad. AI needs physical infrastructure. The question is whether utility-like effects will be matched by utility-like public oversight.

The race is not only about models, chips, or chatbots. It is also about who sets the terms for the power, water, land, and public systems underneath them.