Nvidia CEO Jensen Huang Urges U.S. to Talk With China on AI

Nvidia CEO Jensen Huang told the Dwarkesh Podcast on Wednesday that the United States should open direct talks with China on artificial intelligence instead of trying to block Chinese progress. He argued engagement is more realistic because China already has the hardware, engineers and compute capacity to train large models.

Huang pointed to the compute used to train systems like Anthropic’s Mythos and said that capacity is widely available in China. He noted China produces about 60% of the world’s mainstream chips and said Washington should accept that AI development will continue there and create channels for researchers to communicate and set shared limits on harmful uses.

The executive framed the issue as both practical and security-related. He said the U.S. should recognize China as an adversary while avoiding a posture that prevents dialogue. Huang urged researchers on both sides to meet and try to agree on uses that should be off-limits. He pushed back on the view that AI finding software flaws represents a fundamentally new danger, calling such behavior part of what AI is designed to do.

Huang warned against fragmenting the global AI stack. He argued the open source ecosystem, open models and open stacks let researchers inspect systems and contribute to safety. He said a split where open source development runs on foreign infrastructure while the U.S. tech stack becomes closed would be “extremely foolish” and “a horrible outcome for the United States.”

On infrastructure, Huang pointed to energy constraints as a limit on expanding domestic compute capacity and said power availability could become a bottleneck if not addressed. He said the U.S. should encourage developers worldwide to build on American tools and contribute improvements back to the U.S. ecosystem.

Huang described a growing market for AI security, where startups build many monitoring agents to oversee powerful AI systems. “That would be insane,” he said of a future in which a single AI system ran unchecked, and argued for multiple layers of oversight and active open-source participation to help keep systems secure.

His comments come as lawmakers consider tighter export controls on advanced chipmaking equipment for China and as a Department of Energy-led national AI effort with 17 national labs works to build an integrated platform for scientific research.