AnthropicAnthropicProblem & solution
Securing America's compute advantage: Anthropic’s position on the diffusion rule
Summary
Anthropic’s analysis of the Diffusion Rule, a January 2025 export control framework for advanced AI chips and model weights, warns that maintaining America’s compute advantage is essential for national security. The rule creates a three-tier system: Tier 1 allies face few restrictions, Tier 2 moderate limits, and Tier 3 adversarial nations strict controls. Anthropic recommends allowing Tier 2 countries with strong data center security to obtain more chips via government agreements, lowering the no-license threshold of 1,700 NVIDIA H100s, and boosting enforcement funding. The company notes that Chinese labs like DeepSeek, using chips obtained before controls, already face 2-4x higher energy costs and that by 2027 older-chip countries could pay ten times more for AI training. Anthropic warns against delaying implementation past May 15, 2025, as Chinese firms are stockpiling and a pause would weaken controls.
Context
The Department of Commerce published the ‘Framework for Artificial Intelligence Diffusion’ interim final rule in January 2025, establishing export controls on advanced AI chips and model weights worldwide.
Approach / What changed
Anthropic recommends adjusting the tiering system to allow Tier 2 countries with robust data center security to obtain more chips through government-to-government agreements; reducing the no-license compute threshold for Tier 2 countries (currently 1,700 NVIDIA H100s equivalent); and increasing funding for export enforcement by the Bureau of Industry and Security.
Takeaways
By 2027, countries using older chips could face AI training costs ten times higher than those with cutting-edge American technology, underscoring the value of the compute advantage.
Chinese AI companies like DeepSeek openly acknowledge that chip restrictions are their primary constraint, requiring 2-4x more power to achieve similar results to U.S. companies.
China has established sophisticated smuggling operations for chips, including hiding processors in prosthetic baby bumps and packing GPUs alongside live lobsters, making enforcement a major threat.
FAQ
Why does Anthropic support export controls on AI chips?
Anthropic argues that maintaining America's compute advantage through export controls is essential for national security and economic prosperity, as advanced AI systems require massive computing power and the U.S. leads in semiconductor technology. The controls capitalize on computing power doubling every two years, slowing China's progress while U.S. technology advances.
How does DeepSeek's progress relate to export controls?
Anthropic notes that Chinese labs like DeepSeek acknowledge chip restrictions as their primary constraint, forcing them to use 2-4x more power. DeepSeek likely used frontier chips obtained before controls, and export controls will force them into less efficient Chinese chips, demonstrating that controls work.
What specific changes does Anthropic recommend to the Diffusion Rule?
Anthropic recommends three changes: adjusting the tiering system to allow Tier 2 countries with robust data center security to obtain more chips via government agreements, reducing the no-license threshold for Tier 2 purchases (currently 1,700 H100s equivalent), and increasing funding for export enforcement by the Bureau of Industry and Security.
Source and freshness
- Source
- anthropic.com
- Published
- Added to Yomu
Summaries are generated from the original by a language model. Check the original before relying on it.
Keep reading
Related reading
The Good Stuff
Paid picks we use ourselves. Links carry a referral code — a subscription through one earns us a commission at no extra cost to you.
The best engineering reads, once a week
One useful email. Unsubscribe anytime.