IMPLICATOR
Nvidia CEO Jensen Huang told Axios on Tuesday that American companies should "absolutely" be allowed to use Chinese AI models. The remarks came hours after Treasury Secretary Scott Bessent threatened sanctions if the administration finds that overseas developers stole intellectual property from American labs through model distillation. Moonshot AI's Kimi K3, the release that set off the latest round of the argument, sells at $15 per million output tokens against $50 for Anthropic's Fable 5, Fortune reported.
What Changed
- Nvidia CEO Jensen Huang told Axios on Tuesday that American companies should "absolutely" be allowed to use Chinese AI models, hours after Treasury Secretary Scott Bessent threatened sanctions over model distillation.
- Bessent said the administration is finding watermarks of U.S. large language models on many Chinese models and would look at the matter in the coming days or weeks. He did not identify a Chinese developer as a sanctions target.
- Huang rejected the idea that a downloaded Chinese model creates a backdoor to Beijing, saying companies can control access inside secure sandboxes, and argued that cheaper open models raise demand for the chips and data centers Nvidia supplies.
- Moonshot prices Kimi K3 at $15 per million output tokens against $50 for Anthropic's Fable 5, and says it will release the full model weights by July 27.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Bessent points to model watermarks
Appearing on Fox Business's "Mornings with Maria," Bessent said the administration supports open-source models and could sanction overseas developers if it finds theft. "We are finding watermarks of our US large language models on many Chinese models, and that's unacceptable," he said. "We're going to be looking at that in the coming days or weeks."
The Treasury secretary described allegedly unauthorized distillation, in which developers use a U.S. model's outputs to train another system, as theft. He also floated whether companies using Chinese models should have to tell customers. Meta, Nvidia and Reflection AI were the American companies he named as pursuing open-source models.
Bessent did not identify a Chinese developer as a sanctions target. He will represent the United States at U.S.-China AI talks planned for September, Reuters reported.
Huang argues for open access
Speaking in Fort Worth, Texas, at the opening of a Wistron plant that builds Nvidia AI infrastructure, Huang called the Chinese models "excellent" and said that "open-source models that are excellent should be used." He dismissed the possibility that Chinese developers would displace American labs. "There's no scenario where China runs U.S. companies off the road," Huang told Mike Allen in the interview. "Zero possibility."
Cheaper open models draw more people into AI and raise demand for the chips and data centers Nvidia supplies, Huang argued. Wall Street had "misunderstood the impact of Kimi again this time," he added, and many users would keep paying for the reliability and performance of closed services. "Free AI should be great for hardware," he said.
On security, Huang rejected as a "misconception" the idea that downloaded Chinese models create a "backdoor" to Beijing. Companies can customize the models and control access inside secure "sandboxes," he explained, and outside researchers can inspect open models and build defenses. "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," he said.
Get Implicator.ai in your inbox
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
Against Bessent's theft framing, Huang described distillation as "learning from AI" and from other sources of knowledge, and backed consequences for privacy or contract violations rather than for the models themselves. He also urged Anthropic to offer its restricted Claude Mythos cyber model as a service. "Let Anthropic run," he said.
Jim Cramer, the CNBC host, took the opposite position on X. "We must NOT let our companies use these Chinese models to save a few bucks," he wrote. "OpenAI and Anthropic are correct. This is vital national security."
Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.
Researchers question the theft framing
Hugging Face CEO Clem Delangue told TechCrunch that restricting open models "wouldn't make AI safer" and would "simply hide the risks, concentrate power in the hands of a few." Microsoft CEO Satya Nadella wrote on X last month that he found it "ironic" for model providers to claim fair use rights to train on public data and then "turn around and impose restrictive terms on distillation."
Anthropic, which has pressed the distillation case hardest, agreed to pay $1.5 billion over books downloaded from pirated databases, CNBC noted, and a judge signed off on that settlement this week.
Sam Bresnick, a research fellow at Georgetown's Center for Security and Emerging Technology, questioned why the U.S. government should protect American labs from foreign competitors barred from the U.S. market because of their origin. Bresnick favored halting sales of Nvidia H200 processors to China instead, telling TechCrunch that the step could keep the government out of a dispute over banning open-source technologies that many U.S. companies want to use.
Moonshot says Kimi K3 is available through its apps and API, and that it will release the full model weights by July 27.
Frequently Asked Questions
What did Jensen Huang say about Chinese AI models?
He told Axios on Tuesday that the Chinese models are "excellent," that "open-source models that are excellent should be used," and that American companies should "absolutely" be allowed to use them. He also said there is "no scenario where China runs U.S. companies off the road."
What sanctions did Scott Bessent threaten?
Speaking on Fox Business, Bessent said the administration could sanction overseas developers if it finds they took intellectual property from American labs. He said the U.S. is finding watermarks of its large language models on many Chinese models and would look at that in the coming days or weeks.
What is model distillation?
It is training one system on the outputs of a stronger model. Bessent described allegedly unauthorized distillation as theft. Huang described it as learning from AI and from other sources of knowledge, and said consequences should attach to privacy or contract violations rather than to the models themselves.
Does using a Chinese model create a security backdoor?
Huang called that a misconception. He said companies can customize the models and control access inside secure sandboxes, and that outside researchers can inspect open models and build defenses. Critics including CNBC host Jim Cramer disagree and frame the matter as national security.
When does Moonshot release the Kimi K3 weights?
Moonshot says Kimi K3 is available through its apps and API, and that it will release the full model weights by July 27. The model sells at $15 per million output tokens, against $50 for Anthropic's Fable 5.
AI-generated summary, reviewed by an editor. More on our AI guidelines.
Marcus Schuler
Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai
