Nvidia CEO Jensen Huang used the release of Anthropic Mythos as a springboard to call for direct dialogue between AI researchers in the United States and China. In an April 15 interview on the Dwarkesh Podcast, he warned that tensions between the two countries have created a clear gap in coordination on critical research, and that both sides must agree on boundaries for AI use.
Export Controls Ineffective? Three Reasons from Huang
Asked whether US export curbs have actually stunted China's AI progress, Huang gave a blunt assessment: China is not constrained by a shortage of computing power. He cited three factors — abundant energy, the ability to produce mainstream chips, and techniques to bundle more processors together. The remarks align with Huang's year-long push to persuade the Trump administration to soften restrictions and allow Nvidia to sell its top-tier AI chips to Chinese clients. Last December, the administration permitted the export of the lower-end H200 chip, but no shipments have been completed yet.
Strategic Divergence Between Nvidia and Anthropic
Huang's advocacy for an open China market puts Nvidia at odds with Anthropic, whose CEO has consistently backed stricter export controls. In January, the Anthropic chief called the H200 decision a mistake. Still, the two companies maintain deep commercial ties. Nvidia is investing up to $10 billion in Anthropic; as part of a November deal that also included a $15 billion Microsoft investment, Anthropic committed to purchasing as much as 1 gigawatt of compute from Nvidia.
“We want the US to win,” Huang said, “but I think pushing dialogue — pushing research dialogue — is probably the safest approach.” He called for a cross-border AI safety framework where both sides can agree on acceptable uses. The lack of US-China dialogue, he warned, is becoming the biggest risk in global AI governance.