The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race
← Specific Events

Brazil

Scorecard

Brazil Creates Shared Public AI Compute Infrastructure

What happened

On August 20, Brazil committed roughly R$2.3 billion to expand publicly anchored AI infrastructure, including approximately R$1 billion for a high-capacity AI supercomputer in Rio Grande do Norte and R$1.276 billion over five years for a separate RNP supercomputing project developed with Huawei and iFlytek. Capacity from the Rio Grande do Norte system is intended to be available to universities, scientific institutions, government bodies and companies for model training, research and AI development. The broader strategy also includes a Brazilian cloud initiative, open-architecture RISC-V semiconductor development, technology-transfer and workforce requirements, nationally developed language models, and a national algorithmic-transparency center. Rather than requiring smaller institutions to own frontier-scale infrastructure themselves, Brazil is using public investment to socialize access to infrastructure that few could independently finance while deliberately reducing dependence on any single foreign technology provider.

Potential direction of power

Dispersion · Dispersion +2

This is a strong example of government acting not merely as AI regulator or purchaser but as capacity distributor. Shared public compute offers a second route to contestable scale alongside infrastructure miniaturization: consequential capability can become more widely available either because the infrastructure gets small enough for institutions to own or because public institutions make otherwise unaffordable infrastructure broadly accessible. The remaining test is whether allocation, pricing and priority rules make that access genuinely broad and contestable once the systems become operational.

Window of contestability

Opens

Supporting sources

  1. agenciagov.ebc.com.br · Supporting source 1 ↗