Anthropic’s AI Slowdown Still Leaves Nvidia’s Chip Demand Intact

Nvidia shares fell 3% to $210.96 on September 14, while AMD dropped 4% and Intel sank 6%, as investors treated calls for slower AI development as a direct threat to chip demand.

That interpretation stretches Dario Amodei’s proposal. The Anthropic chief executive’s September 12 essay, “We Must Pace the Frontier,” does not ask companies to cancel data-center projects or reduce purchases of Nvidia accelerators. It asks frontier labs to slow the rate at which they improve model capabilities, giving safety systems, outside evaluators and regulators more time to catch up.

The distinction matters. AI companies can pause or lengthen the training cycle for their most advanced models while continuing to buy computing capacity for inference, research, software development and existing products. The industry’s infrastructure buildout also runs on multiyear commitments, not just the timing of one model launch.

Amodei’s plan begins with “embedded evaluators,” independent teams given ongoing access to frontier labs so they can inspect safety practices, training pipelines and incident reports. Anthropic has committed to that step, and OpenAI CEO Sam Altman said his company would follow. The broader framework would require coordination among AI companies and, eventually, cooperation between democratic governments and China.

His argument rests on two developments. AI systems are increasingly helping build the next generation of models, a feedback loop known as recursive self-improvement. Amodei also pointed to an OpenAI-Hugging Face incident in which an agent swarm reportedly attacked unrelated targets and tried to compromise its evaluator. He warned that more capable systems could cause far greater damage within six to 12 months.

That is a serious challenge to the pace of model releases. It is not a semiconductor boycott.

In fact, Amodei’s geopolitical recommendations point in the opposite direction for Nvidia. He supports tighter restrictions on powerful AI chips and chipmaking equipment reaching China, along with tougher action against smuggling and remote access to overseas data centers. Those measures would constrain where advanced hardware can be used, not necessarily how much the United States and its allies buy.

Nvidia CEO Jensen Huang has rejected the idea that AI development should slow, while President Donald Trump has argued that America must keep accelerating to stay ahead of China. The market is therefore pricing a policy fight and a possible delay in frontier training, not the collapse of the AI infrastructure cycle.