Jensen Huang Calls AI Distillation 'Competition,' Not Theft, as Nvidia Weighs China Policy

Nvidia chief executive Jensen Huang said AI distillation amounts to 'competition' rather than something closer to theft, weighing in on a practice that has become a central flashpoint in the U.S.-China technology standoff. The comments, reported by CNBC, land as policymakers in Washington consider whether to restrict Chinese labs from training models on outputs generated by American systems. Huang's framing matters because Nvidia sells the chips that power nearly every major model on both sides of that divide.

Distillation is the technique of training a smaller model on the outputs of a larger, more capable one, letting a developer approximate strong performance at a fraction of the compute cost. American labs including OpenAI and Anthropic have argued that foreign competitors use the method to shortcut years of expensive research, effectively copying capabilities without paying for them. Those companies have pressed for contractual and, in some cases, legal barriers to the practice.

Huang's position puts him at odds with some of Nvidia's largest customers, a tension he has managed before. The company has spent years arguing that export controls on advanced accelerators hurt U.S. competitiveness by pushing buyers toward domestic Chinese alternatives such as Huawei's Ascend line. Huang has repeatedly traveled to Washington to make that case, and he has warned that overly broad restrictions would accelerate China's own chip industry rather than slow it.

The debate carries real money. Nvidia's data-center revenue has scaled into the tens of billions of dollars per quarter, and China once accounted for a meaningful share of that before successive rounds of U.S. licensing requirements narrowed what the company could legally ship there. Every tightening of the rules forces Nvidia to redesign products for the Chinese market or forgo the sales entirely, a cost that shows up in guidance.

What Huang has not done is offer a specific policy proposal, leaving open whether he favors letting distillation continue unfettered or simply wants the argument separated from export controls. The comment threads that surfaced the CNBC piece suggest the technical community is split, with some engineers noting that distillation is standard practice inside American labs too. The next signal will come from Washington: whether regulators treat model outputs as protected property, and whether Nvidia's lobbying shifts that outcome.