On July 22, 2026, US Treasury Secretary Scott Bessent warned that the United States will scrutinise Chinese open-source AI models for stolen intellectual property and could sanction their makers within weeks. On the same day, Nvidia chief executive Jensen Huang publicly disagreed, arguing that strong open-weight Chinese models should be allowed into the US market because broader AI adoption ultimately drives demand for Nvidia’s chips.<\/p>\n
Speaking on television, Bessent said the administration supports open-source models but opposes intellectual property theft. The evidence, he claimed, is technical: US officials have found watermarks from American large language models embedded in Chinese models. The practice is known in AI research as distillation, training one model on the outputs of another. Bessent framed it as theft and said the Treasury Department would act within weeks.<\/p>\n
The trigger was Kimi K3, an open-weight model released by Chinese lab Moonshot AI. The model matched or exceeded leading systems from OpenAI and Anthropic on several benchmarks while being offered at a much lower price. Its release contributed to a selloff in chip stocks and renewed concern in Washington about US competitiveness in AI. US labs have accused Chinese competitors of siphoning their models for months.<\/p>\n
Speaking to Axios at a new chip facility in Texas, Nvidia’s Jensen Huang called the concern misplaced. He called the Chinese models excellent and said open-source models that perform well should be used. American firms, he argued, should be allowed to run them. His reasoning is commercial: cheaper and free models spread AI more widely, and wider AI adoption increases demand for the chips, data centres, and power infrastructure that Nvidia sells. Huang also rejected the idea that a downloaded Chinese model is a backdoor for Beijing, saying companies can isolate them in secure sandboxes.<\/p>\n
Not everyone accepts the framing of distillation as theft. Hugging Face chief executive Clem Delangue told TechCrunch that distillation is a small factor in model performance and a practice that everyone is doing, including companies in the United States. There is additional awkward context: Anthropic, one of the most vocal accusers of Chinese distillation, recently had a $1.5 billion settlement approved over pirated books used to train its Claude model. Huang went further, urging the government to stop restricting Anthropic’s own Mythos model and told reporters to let Anthropic run.<\/p>\n
Bessent will lead the US delegation at AI talks with China in September, and the distillation dispute is expected to be on the agenda. For now, the split is clear: the Treasury Department wants to fence Chinese models out of the US market, while the world’s most valuable chipmaker wants them allowed in.<\/p>\n
Bessent said the US will scrutinise Chinese open-source models for stolen intellectual property and could sanction their makers. He said US officials have found watermarks from American large language models embedded in Chinese models and that action could come within weeks.<\/p>\n
Huang said open-weight Chinese models are excellent and argued that US firms should be allowed to run them. His reasoning is commercial: free and cheaper models put AI in more hands, which increases demand for the chips, data centres, and power infrastructure Nvidia sells.<\/p>\n
The trigger was Kimi K3, an open-weight model from Moonshot AI that matched or beat leading US models on several benchmarks at a lower price, contributing to a chip stock selloff and renewed concern in Washington about US competitiveness.<\/p>\n