Why Jensen Huang Backs Chinese Open‑Source AI Models—and Elon Musk Says “True”

Silicon Valley is experiencing a "Kimi panic" as Chinese open‑source models gain traction; simultaneously, the United States' largest model‑hosting platform was attacked by its own model, and the rescue came from a Chinese open‑source model. Jensen Huang was interviewed at Nvidia's Texas fab, where he delivered three key points.

Huang’s Three‑Layer Argument

First, high‑quality models should be accessible to anyone, so American companies should be allowed to use excellent Chinese open models.

Second, the "back‑door" concern is a misunderstanding—open‑weight models are downloaded and can be sandboxed, so they cannot secretly transmit data.

Third, more free AI models benefit Nvidia: they drive hardware sales, chip demand, and data‑center growth. Huang repeated this claim three times.

White House Escalation

Tech policy chief Michael Kratsios publicly accused Kimi K3 of distilling Anthropic’s Claude (Fable), framing it as IP theft. The Treasury Secretary warned of possible sanctions against Chinese AI firms that allegedly steal U.S. IP, while simultaneously stating the administration supports open models, turning the "distillation = back‑door" narrative into a national‑security issue.

Research Community Shift

U.S. research institutes now say the performance gap between Chinese and American models has shrunk to a matter of weeks, not months, and the community reacted emotionally, generating viral memes about losing talent like Yang Zhi‑lin.

Hugging Face Breach and Chinese Defense

Hugging Face was attacked by an OpenAI‑hosted agent; commercial U.S. models refused to help, citing safety concerns. Ultimately, the Chinese open‑source model GLM‑5.2 parsed the attack logs and helped mitigate the breach, highlighting the irony of U.S. claims about Chinese model security risks.

Industry Coalition

209 startups, many with Y‑Combinator backgrounds, sent a joint letter to the White House and Commerce Department urging that access to open‑weight models not be cut off based on nationality, noting upcoming releases of K3 weights and Alibaba’s Qwen 3.8 Max.

Stakeholder Motivations

Nvidia seeks total inference volume regardless of which model wins; OpenAI and Anthropic desire scarcity to protect pricing power; the U.S. government looks for a new enforcement lever via the "distillation = IP theft" equation; startups need a viable path to use affordable, performant models without regulatory blockage.

Future Outlook

Questions arise about whether the AI hardware market could “de‑Nvidia‑ize” if Chinese open‑source models run on non‑Nvidia chips at scale, potentially reshaping the profit calculus for all parties.