Anthropic CEO, Dario Amodei, publicly lamented the U.S. policy change, along with key chipmakers, about the move to authorize the export of high performance AI chips to China. In one of the interviews with Bloomberg, he had said the comments during the World Economic Forum in Davos. Amodei had warned to the foreign governments of the dreadful national security consequences of exporting hi-tech processors.
Amodei said the nation that controls the most powerful models will gain a strategic advantage. He framed the issue as a risk that extends beyond commerce. He said the idea of enabling large scale compute in other countries was reckless. He used stark language to press the point. “I think this is crazy,” he said. “It is a bit like selling nuclear weapons to North Korea.”
Amodei Criticizes Chips
Amodei focused his criticism on the recent decision by the U.S. Commerce Department to permit the sale of Nvidia H200 chips under conditions. The move followed a policy reversal announced by the administration. The rules allow some sales provided exporters meet third party checks and supply constraints are met. Amodei said those conditions are not enough to blunt long term risks.
He also commented on the role of chip companies. He said that business leaders who lobby for broader exports are underestimating the implications. That critique was notable because Nvidia is a partner and major investor in Anthropic. The relationship between the chip maker and the AI firm adds tension to the public exchange. Analysts noted that Amodei’s tone was unusually direct given the commercial ties between the firms.
Amodei argued that the United States still holds a lead in chip making and in building cutting edge models. He said selling advanced chips risks reducing that lead. He described future large scale models as immensely powerful and warned that control of that capability matters geopolitically. His remarks aimed to inject urgency into how policy makers and executives think about exports.
Nvidia and Policy
The Commerce Department set rules that include third party review and limits on quantities tied to domestic supply. The administration said those steps balance national security with commercial interests. Some lawmakers and security experts criticized the decision. They argued that the checks may be difficult to enforce and that chips could still boost foreign AI capability. Bloomberg, reported on the new rules and on the mixed reaction from industry and policy circles.
Nvidia has sought to reopen access to China. Executives say that limiting the market can hurt global business and that denying access may prompt rivals to accelerate their own chip efforts. The company must follow the new export conditions and ensure compliance. Nvidia also recently made a large investment in Anthropic, a step that makes the public rebuke more striking. Bloomberg captured both the interview and the broader context at Davos.
The policy shift matters for two reasons. First, chips like the H200 enable large scale training and inference for modern AI models. Second, the new regulations give precedence on how the United States is going to balance security and trade in a pivotal technology zone. Amodei’s comments brought a human voice to those abstract tradeoffs and forced a public debate at a high level meeting of global leaders.
What Observers Said
Industry watchers gave the remarks wide coverage. Some media described Amodei’s comparison as dramatic. Others said he raised a valid strategic point about the pace of technology transfer. Commentators noted the irony that a company that depends on Nvidia hardware would publicly criticize the chip maker’s role in export policy. That tension will likely influence how other executives speak about the issue in public.
Policy experts said the debate will now move to enforcement details. Questions remain about how third party testing will work and how to ensure chips are not used for military programs. The Commerce Department said it set clear conditions. Critics said those conditions may be porous. The coming weeks will test whether limits can be implemented effectively.
