Nvidia chief executive Jensen Huang said American companies should be allowed to use Chinese AI models, putting him at odds with Washington efforts to restrict them.
In an interview with Axios co-founder Mike Allen, Huang answered “absolutely” when asked whether US companies should be permitted to use Chinese models. The comment landed as Chinese developer Moonshot AI has drawn attention for Kimi K3, a 2.8 trillion-parameter open-weight model.
Tom’s Hardware reported that Kimi K3 trails frontier systems such as Fable 5, but is comparable to GPT 5.5 and Claude Opus 4.8 while costing about one-third as much to run. That combination, a capable model with weights users can download, is exactly the sort of thing regulators tend to find inconvenient.
Huang disputes the backdoor argument
US officials have raised concerns that Chinese AI models could contain vulnerabilities or hidden access paths that Beijing could use against American interests. Huang rejected that premise in the Axios interview.
“There is a misconception that somehow there are backdoors that are somehow connected to China in some way,” Huang said. He argued that users who download models can modify them, tune them for their own purposes, improve them, and add safeguards.
That is the open-weight argument in its cleanest form: the model is not a remote service controlled only by its maker. A company can run it on its own infrastructure and inspect or alter parts of the system. That does not erase security risk, but Huang’s position is that banning access is the wrong default.
US models have faced controls too
Huang applied the same logic to American AI systems. Tom’s Hardware reported that the US recently imposed export restrictions on Anthropic’s Mythos and Fable 5 models, citing security risks. Access was later restored after Anthropic added a filter intended to stop the tools from identifying software vulnerabilities.
OpenAI’s ChatGPT-5.6 was also hit with similar treatment, according to Tom’s Hardware, and Washington warned OpenAI not to release its newest model without government approval.
Huang’s view is that companies should release powerful models broadly and then improve their safety through fast testing and fixes, rather than bottling them up at launch. He also said fields such as science and cybersecurity need open models, because outside researchers can inspect them for flaws and help repair them.
“If everything just becomes one single model, one single point of attack, one single source of failure, I think the world is much, much more vulnerable,” Huang said.
Cheaper models are not bad news for Nvidia
Huang also pushed back on investor anxiety around lower-cost open-weight models. He said markets reacted negatively when DeepSeek appeared, and are reacting again to Kimi.
His argument is convenient for Nvidia, but not incoherent: cheaper models can increase AI use, and more AI use can increase demand for data centers and GPUs. Huang said lower operating costs encourage broader adoption, which could raise demand for the hardware Nvidia sells rather than reduce it.
This story draws on original reporting from Tom's Hardware.
