Securing America's Compute Advantage: Anthropic’s Position on the Diffusion Rule
In response to the Department of Commerce's "Framework for Artificial Intelligence Diffusion" interim final rule, Anthropic has submitted detailed analysis and recommendations for maintaining and strengthening export controls on advanced semiconductors. At the heart of our recommendations is a clear message: maintaining America's compute advantage through export controls is essential for national security and economic prosperity as powerful new AI systems are developed in the coming years.
The "Diffusion Rule," published in January 2025, establishes export controls on advanced AI chips and model weights worldwide. The Diffusion Framework also creates a three-tier system based on national security risk. Tier 1 includes close allies with few restrictions, Tier 2 includes most other countries with some limits, and Tier 3 includes adversarial nations with strict controls.
The first Trump Administration correctly diagnosed that AI will be central to strategic competition with China, and that the United States can and should use export controls to maintain and strengthen its AI leadership. While the U.S. still maintains a lead in AI development, Chinese AI labs like DeepSeek have made significant progress, using chips obtained before export controls went into effect, and underscore the importance of strong export controls on advanced chips.
We ground our recommendations in several key facts:
Compute Advantage is Critical: Advanced AI systems require massive computing power to train and operate. The U.S. currently leads in advanced semiconductor technology, and export controls capitalize on the trend of computing power doubling every two years, so while U.S. chip technology continues advancing, China’s progress is slowed. This growing efficiency gap means that by 2027, countries using older chips could face AI training costs that are ten times higher than those with cutting-edge American technology.
DeepSeek Shows Controls Work: Chinese AI companies like DeepSeek openly acknowledge that chip restrictions are their primary constraint, requiring them to use 2-4x more power to achieve similar results to U.S. companies. DeepSeek also likely used frontier chips for training their systems, and export controls will force them into less efficient Chinese chips.
Cement America’s Infrastructure Lead: Without strong controls, AI infrastructure development for frontier-scale training could shift overseas (similar to solar panels and batteries), threatening U.S. strategic advantage. The U.S. share of global semiconductor production has fallen from 40% in 1990 to just 12% today, with 90% of the world's leading-edge semiconductors now made outside the U.S. This offshoring represents a strategic vulnerability, and we must not make the mistake again with AI. The Diffusion Framework addresses this risk through domestic compute requirements and foreign compute caps, ensuring the coming AI infrastructure buildout happens in America by American firms. Current supply constraints, combined with the framework's global export regime, create clear incentives for other nations: accessing large-scale cutting-edge compute requires investing in U.S. AI infrastructure and complying with American export controls.
Chip Smuggling is a Major Threat: China has established sophisticated smuggling operations, with documented cases involving hundreds of millions of dollars worth of chips. In some cases, smugglers have employed creative methods to circumvent export controls, including hiding processors in prosthetic baby bumps and packing GPUs alongside live lobsters. Chinese firms continue to establish shell companies in third countries at a rapid pace to evade export controls.
Anthropic’s submission recommends strengthening the rule in three key areas:
1. Adjust the tiering system. We suggest allowing countries in Tier 2 with robust data center security to obtain more chips through government-to-government agreements that prevent smuggling and align technology controls.
2. Reduce the no-license compute threshold for Tier 2 countries. Currently, countries in Tier 2 can buy the equivalent of 1,700 NVIDIA H100 advanced chips without needing government permission—roughly $40 million worth of technology. This creates a potential loophole for smuggling, as people can make multiple purchases just under this limit to avoid scrutiny. We recommend lowering this threshold so that more transactions would require review, making it harder for smugglers to exploit this gap.
3. Increase funding for export enforcement. Export controls are only effective with proper enforcement. Enhanced resources for the Bureau of Industry and Security would significantly improve control effectiveness.
Our submission also highlights the risks of further delaying implementation. Chinese firms have engaged in aggressive stockpiling ahead of the May 15, 2025 implementation date. Any pause would invite further stockpiling and weaken the effectiveness of the rule at this critical moment.
The strategic window for strengthening American export controls is now. By strengthening the Diffusion Framework, America can ensure transformative AI technologies are developed domestically, in alignment with American values and interests. Our continued leadership in AI depends on maintaining our compute advantage through decisive action today.
Anthropic’s leaders have previously outlined our position on competition with China and export controls in the following pieces.
- “Trump Can Keep America’s AI Advantage,” Dario Amodei & Matthew Pottinger, Wall Street Journal, January 6, 2025.
- “On DeepSeek and Export Controls,” Dario Amodei, darioamodei.com, January 2025“Anthropic's Dario Amodei on AI Competition,” China Talk Podcast, February 5, 2025
- “DeepSeek means AI proliferation is guaranteed,” Jack Clark Import AI 397