Compute governance

In plain English

Training a frontier model takes enormous amounts of specialized hardware, energy and money, concentrated in a handful of chip makers and cloud providers. That makes compute one of the few parts of SI development that governments can see and count. Compute governance uses this: export controls on advanced chips, reporting duties for large training runs, know your customer rules for cloud providers and proposals for chips that can verify how they are used.

Compute thresholds are the most visible tool. The EU AI Act presumes systemic risk for general purpose models trained with more than 10^25 FLOP, and several US state laws use 10^26 operations to define frontier developers. Compute governance cannot see everything. Algorithms keep getting more efficient, so the same capability needs less compute each year, and once weights are released, compute controls no longer limit who can run the model.

Why it matters for rogue SI

Compute is the physical choke point of SI development. If a dangerous training run ever needs to be prevented, control over chips and data centers is one of the few levers that works without the model's cooperation.

Laws and rules that use it

• In force EU AI Act (GPAI) European Union
• In force SB 53 California
• Passed, not yet in force RAISE Act New York
• Passed, not yet in force SB 315 Illinois
• In force AI Basic Act South Korea

Sources

1. Guidelines for providers of general-purpose AI models European Commission · Jul 18, 2025
2. SB 53: Artificial intelligence models: large developers CalMatters Digital Democracy