Anthropic CEO criticizes U.S. approval of some AI chip sales to China, warning of security risks

Anthropic CEO Dario Amodei delivered a pointed critique at the World Economic Forum in Davos after the U.S. administration approved the sale of Nvidia’s H200 chips — and an AMD chip line — to approved Chinese customers. While not the very most advanced chips, the processors are still considered high-performance and relevant to cutting-edge AI workloads. ([techcrunch.com](https://techcrunch.com/2026/01/20/anthropics-ceo-stuns-davos-with-nvidia-criticism/?utm_source=openai))

According to TechCrunch, Amodei argued the policy could come back to hurt the United States and warned that exporting these chips may be strategically unwise given what he described as the geopolitical stakes of advanced AI. The comments were notable because Nvidia is described as a major partner and investor in Anthropic, putting him at odds with an important ally. ([techcrunch.com](https://techcrunch.com/2026/01/20/anthropics-ceo-stuns-davos-with-nvidia-criticism/?utm_source=openai))

Amodei framed the issue in national-security terms, emphasizing that access to compute affects the frontier of AI model development. He suggested the U.S. maintains a lead in chipmaking and that shipping powerful chips could narrow that advantage faster than policymakers anticipate. ([techcrunch.com](https://techcrunch.com/2026/01/20/anthropics-ceo-stuns-davos-with-nvidia-criticism/?utm_source=openai))

In Davos remarks summarized by TechCrunch, he also used vivid analogies to convey the potential magnitude of advanced AI — comparing future systems to vast concentrations of intelligence — to argue that chip policy should be treated as strategic infrastructure rather than routine commerce. ([techcrunch.com](https://techcrunch.com/2026/01/20/anthropics-ceo-stuns-davos-with-nvidia-criticism/?utm_source=openai))

The controversy sits at the intersection of economic and security goals. U.S. companies want to sell hardware into enormous global markets, while policymakers attempt to limit the transfer of capabilities that could enhance rival states’ military or intelligence capacity. Amodei’s argument, as presented, is that these incentives are in tension — and that the wrong balance could be irreversible if competitors catch up. ([techcrunch.com](https://techcrunch.com/2026/01/20/anthropics-ceo-stuns-davos-with-nvidia-criticism/?utm_source=openai))

For the AI industry, the episode is another sign that “compute diplomacy” is now core policy: chip export rules, cloud access, and model deployment constraints are becoming as consequential as software innovation. How governments define “approved customers,” monitor end-use, and update restrictions will shape not just supply chains but also the pace and geography of frontier model training.