Last week, after reversing an earlier ban, the U.S. administration officially approved the sale of Nvidia’s H200 chips, along with a chip line by AMD, to approved Chinese customers. Maybe they aren’t these chipmakers’ shiniest, most advanced chips, but they’re high-performance processors used for AI, making the export controversial. And at the World Economic Forum in Davos on Tuesday, Anthropic CEO Dario Amodei unloaded on both the administration and the chip companies over the decision.

The criticism was particularly notable because one of those chipmakers, Nvidia, is a major partner and investor in Anthropic.

“The CEOs of these companies say, ‘It’s the embargo on chips that’s holding us back,’” Amodei said, incredulous, in response to a question about the new rules. The decision is going to come back to bite the U.S., he warned.

“We are many years ahead of China in terms of our ability to make chips,” he told Bloomberg’s editor-in-chief, who was interviewing him. “So I think it would be a big mistake to ship these chips.” Amodei then painted an alarming picture of what’s at stake. He talked about the “incredible national security implications” of AI models that represent “essentially cognition, that are essentially intelligence.” He likened future AI to a “country of geniuses in a data center,” saying to imagine “100 million people smarter than any Nobel Prize winner,” all under the control of one country or another.

The image underscored why he thinks chip exports matter so much. But then came the biggest blow. “I think this is crazy,” Amodei said of the administration’s latest move. “It’s a bit like selling nuclear weapons to North Korea and [bragging that] Boeing made the casings.”

That sound you hear? The team at Nvidia, screaming into their phones.

Nvidia isn’t just another chip company. While Anthropic runs on the servers of Microsoft and Amazon and Google, Nvidia alone supplies the GPUs that power Anthropic’s AI models (every cloud provider needs Nvidia’s GPUs). Not only does Nvidia sit at the center of everything, but it also recently announced it was investing in Anthropic to the tune of up to $10 billion.

Just two months ago, the companies announced that financial relationship, along with a “deep technology partnership” with cheery promises to optimize each other’s technology. Fast-forward to Davos, and Amodei is comparing his partner to an arms dealer.

Maybe it was just an unguarded moment — it’s possible he got swept up in his own rhetoric and blurted out the analogy. But given Anthropic’s strong position in the AI market, it seems more likely he felt comfortable speaking with confidence. The company has raised billions, is valued in the hundreds of billions, and its Claude coding assistant has developed a reputation as a highly beloved and top-tier AI coding tool, particularly among developers working on complex, real-world projects.

It’s also entirely possible that Anthropic genuinely fears Chinese AI labs and wants Washington to act. If you want to get someone’s attention, nuclear proliferation comparisons are probably a pretty effective way to do it.

But what’s perhaps most remarkable is that Amodei could sit onstage at Davos, drop a bomb like that, and walk away to some other gathering without fear that he just adversely impacted his business. News cycles move on, sure. Anthropic is also on solid footing right now. But it does feel that the AI race has grown so existential in the minds of its leaders that the usual constraints — investor relations, strategic partnerships, diplomatic niceties — don’t apply anymore. Amodei isn’t concerned about what he can and can’t say. More than anything else he said on that stage, that fearlessness is worth paying attention to.