In a significant policy shift last week, the U.S. government granted official approval for the sale of Nvidia’s H200 chips and a comparable line from AMD to authorized customers in China. While these may not represent the absolute cutting edge from these manufacturers, they are high-performance processors designed for artificial intelligence, rendering the exports a subject of considerable debate.

This decision drew sharp criticism at the World Economic Forum in Davos on Tuesday from Anthropic CEO Dario Amodei, who directed his remarks at both the administration and the chip firms. His critique was especially striking given that one of those companies, Nvidia, is a major partner and a $10 billion investor in Anthropic. Responding to a question about the new regulations, Amodei expressed disbelief, quoting industry leaders who often blame chip embargoes for holding back progress. He warned that this move would ultimately harm U.S. interests.

“We are many years ahead of China in terms of our ability to make chips,” Amodei stated in an interview with Bloomberg’s editor-in-chief. “So I think it would be a big mistake to ship these chips.” He proceeded to outline a grave scenario, emphasizing the “incredible national security implications” of AI systems that embody what he described as “essentially cognition” and “essentially intelligence.”

He compared future advanced AI to a “country of geniuses in a data center,” asking the audience to envision “100 million people smarter than any Nobel Prize winner” all under the control of a single nation. This vivid analogy highlighted his view on the critical importance of controlling chip exports.

He then delivered his most pointed rebuke. “I think this is crazy,” Amodei said of the administration’s action. “It’s a bit like selling nuclear weapons to North Korea and [bragging that] Boeing made the casings.”

One can imagine the reaction at Nvidia headquarters. The company is not merely another chip supplier; it occupies a central role in the AI ecosystem. While Anthropic operates on cloud infrastructure from Microsoft, Amazon, and Google, it is Nvidia’s GPUs that power its AI models, as every major cloud provider relies on them. Furthermore, Nvidia recently announced an investment of up to $10 billion in Anthropic, coupled with a “deep technology partnership” pledged just two months ago to optimize their respective technologies.

Yet at Davos, Amodei likened his financial partner to an arms dealer. This may have been an unguarded moment, perhaps a rhetorical flourish spoken in the heat of the discussion. However, given Anthropic’s formidable standing in the AI sector—having raised billions, achieved a valuation in the hundreds of billions, and developed its Claude coding assistant into a revered, top-tier tool among developers—it is plausible he felt secure in speaking so candidly.

It is also entirely possible that Anthropic holds a genuine concern regarding the capabilities of Chinese AI labs and desires stronger action from Washington. If the goal is to capture attention, comparisons to nuclear proliferation are undoubtedly effective.

What remains particularly notable is that Amodei could deliver such a explosive statement on a global stage like Davos and walk away to his next engagement seemingly unconcerned about potential backlash to his business. News cycles fade, and Anthropic is currently on solid ground. Nevertheless, this episode suggests that for leaders in the AI field, the competitive race has taken on such an existential character that conventional restraints—be they investor relations, strategic partnerships, or diplomatic courtesy—no longer seem to apply. Beyond the specific content of his remarks, it is that very fearlessness which commands attention.