Jensen Huang says Nvidia must work with China on AI safety

Nvidia CEO Jensen Huang says his company must cooperate with Chinese AI companies on safety, and he is already doing so. In an interview with CNBC, Huang said he maintains ongoing "dialogs" with AI researchers in China.

That is a short quote doing heavy lifting. The head of the world's dominant AI chipmaker is publicly framing safety work with Chinese researchers as a necessity, not a favor — at a moment when Washington is trying to wall off advanced AI technology from China entirely.

What did Huang actually say?

Huang told CNBC two things: that Nvidia must work with Chinese AI companies on AI safety, and that he personally has ongoing "dialogs" with researchers in China. He did not, according to the CNBC report, name the specific companies or researchers involved, describe the format of those talks, or attach a timeline to them.

The statement is notable precisely because of who is making it. Nvidia designs the GPUs that power most large AI model training worldwide, and Huang's business sits directly on the fault line of US-China technology policy. When he says safety cooperation with China "must" happen, he is speaking as the chief executive of a company whose exports to China have been the subject of repeated US restrictions.

Takeaway: Nvidia's CEO has put his name to the idea that AI safety cannot be a US-only project.

Why does this matter for AI policy?

The stakes here are structural. AI safety — making powerful models reliable, controllable and hard to misuse — is a global technical problem. Models built anywhere can cause harm everywhere, and safety research travels through papers, benchmarks and shared practices. If researchers in the US and China don't compare notes, both sides fly blinder than they need to.

Huang's comments land in a policy environment that pushes the other way. The US has spent years tightening export controls on advanced chips and AI technology bound for China, and pressure to decouple the two AI ecosystems keeps building. Against that backdrop, a chip-industry CEO arguing for mandatory safety dialogs is a small but real counterweight.

Huang did not lay out in the CNBC interview how safety cooperation would survive those restrictions, or whether it should be company-led, academic, or government-brokered. His quote establishes the position, not the mechanism.

Takeaway: Huang is staking out a stance that safety cooperation should continue even as the two AI economies pull apart.

What's the business context?

Nvidia has an obvious commercial interest in keeping channels to China open. The Chinese market has historically been one of Nvidia's largest, and every round of US export restrictions has shrunk what the company can legally sell there. Framing engagement as a safety imperative rather than a revenue question gives Nvidia a policy argument that doesn't rest on its own balance sheet.

At the same time, Huang's "dialogs" with Chinese researchers are consistent with how safety work actually happens in practice: informal researcher-to-researcher contact, conference exchanges and shared technical standards, much of it below the level of formal government agreements. His CNBC comments are an unusually public acknowledgment of that quiet layer of cooperation.

Huang did not say in the interview whether those dialogs have produced anything concrete — joint research, shared safety evaluations or agreed standards. The public record on that point is simply that the conversations exist.

Takeaway: Safety talk with China gives Nvidia an argument for engagement that survives export-control politics.

What happens next?

The open question is whether "dialogs" turn into anything durable. That would mean named participants, published results or a standing channel between US and Chinese AI safety researchers — none of which the CNBC interview specified.

The larger test is political. If US-China technology restrictions tighten further, even informal researcher-to-researcher contact becomes harder to defend in Washington. Huang's position — that safety work with China is a "must," in his words to CNBC — sets up a direct tension between export-control policy and the argument that AI safety is inherently transnational. How that tension resolves will shape whether the world's two biggest AI ecosystems share safety knowledge or build separate ones.