Compute Governance in plain English.

Compute governance refers to government policies that control who can access the advanced chips and large-scale data centers needed to train frontier AI models, on the theory that restricting the physical hardware is more enforceable than restricting the resulting software. The clearest example is the U.S. restricting exports of the most advanced AI chips to China, which has shaped where and how Chinese labs like DeepSeek train their models. Compute governance also shows up domestically: reporting requirements for training runs above a certain compute threshold, and proposals to license or monitor very large data centers the way other high-risk infrastructure is regulated. Supporters argue it's the most practical lever governments have to slow down risky capability races; critics argue it mainly slows down legitimate research and pushes capable actors toward workarounds. By 2026, compute governance is one of the most contested and fastest-changing areas of AI policy globally.