Compute Governance
Regulatory approaches that use computational resources as a lever for AI oversight and control.
Definition
Compute Governance refers to regulatory and policy approaches that leverage computational resources (computing power, cloud infrastructure, specialized AI chips) as a mechanism for AI governance, oversight, and control. It recognizes that access to compute is a key bottleneck for developing advanced AI.
US Approach (Executive Order 14110):
- Reporting requirements for models trained above compute thresholds (10^26 FLOP)
- Know-your-customer requirements for cloud providers
- Export controls on advanced AI chips
- Reporting on large-scale compute clusters
EU AI Act Approach:
- GPAI systemic risk threshold based on cumulative compute (10^25 FLOP)
- Compute as proxy for model capability and potential risk
- Commission can adjust thresholds as technology evolves
Governance Mechanisms:
- Thresholds: Using compute metrics to trigger regulatory requirements
- Reporting: Mandatory disclosure of compute usage
- Access controls: Know-your-customer for cloud AI services
- Export controls: Restricting advanced hardware exports
- Allocation: Government influence over compute distribution
Rationale: Compute is measurable, concentrated among few providers, and correlated with AI capability—making it a practical regulatory leverage point.
Related concepts: Systemic Risk, General-Purpose AI Model with Systemic Risk, Frontier AI
Sources
- •Executive Order 14110
- •Frontier AI Governance
Related Terms
AI Export Controls
US restrictions on exporting AI technology, models, and chips to certain countries....
Computing Power Resources
Computational infrastructure and capabilities essential for training, deploying, and operating complex AI models....
Sovereign AI compute
Public-sector controlled compute resources for national missions....
Catastrophic Risk
Risk of severe, large-scale, potentially irreversible negative impacts from AI systems on health, safety, or fundamental rights....
AI Sovereignty
National or regional control over AI development, data, and infrastructure independent of foreign providers....