Anthropic CEO criticizes Nvidia chip sales to China, warning the U.S. is taking a strategic risk

A Davos moment that surprised the AI industry

Davos is known for carefully calibrated remarks, but one of the week’s most pointed soundbites came from an AI executive whose company depends on the very hardware he criticized. Anthropic CEO Dario Amodei used a Davos appearance to argue that U.S. policy on AI chip exports is moving in the wrong direction, and he aimed his frustration not only at policymakers but also at the semiconductor firms eager to sell abroad.

The dispute centers on the U.S. administration’s decision to approve sales of Nvidia’s H200 chips — along with a chip line by AMD — to approved Chinese customers. Even if these products are not the most advanced variants available, they remain high-performance processors with clear value for AI development.

Amodei’s argument: strategic lead is fragile

Amodei warned that the U.S. should treat its edge in advanced compute as a strategic asset, not a commercial lever to be traded away. He suggested that exporting such chips risks accelerating capabilities in a rival ecosystem, potentially weakening the U.S. position in what many leaders increasingly describe as an AI race with national-security consequences.

In his remarks, he rejected the framing that export restrictions are mainly an obstacle to American business, instead casting them as a key part of preventing dangerous concentration of AI capabilities in adversarial hands. The warning landed harder because it came from the leader of one of the most influential AI labs in the market.

Why the critique is awkward for business

Anthropic is a major customer of Nvidia’s GPUs through the cloud providers that supply the company’s compute, and Nvidia has also been publicly linked to large investments and partnerships across the AI sector. That creates a tension: AI labs need huge volumes of cutting-edge chips to train and run models, but some of those labs also want strict controls to slow down competitors overseas.

Amodei’s willingness to criticize the administration and chip companies in such blunt terms signals how existential some AI leaders view the competition. In previous eras, executives might have softened the message to avoid antagonizing suppliers and investors. At Davos, the tone suggested those constraints are weakening.

The broader fight: exports, regulation, and AI power

The episode reflects a larger debate unfolding across Washington and industry: how to balance economic interests with security concerns, and how to define which chips or systems are “too capable” to export. As AI systems become more powerful, the boundary between commercial technology and strategic infrastructure is becoming harder to draw.

For now, Amodei’s critique adds pressure to an already contentious issue. Companies want clarity and stable rules; security advocates want tighter constraints. Davos made clear that the fight is no longer abstract — and the leaders building frontier models are increasingly willing to argue in public about where the U.S. should draw the line.