The White House says Chinese AI startup Moonshot AI accessed restricted Nvidia GB300 chips in Thailand to train its latest model — a model that already surpassed American competitors on key benchmarks. The stake for ordinary Americans: Washington's national security apparatus is tightening regulatory chokeholds on U.S. innovators while Chinese firms simply route around them, and the only answer from D.C. is more of what already failed.
White House Office of Science and Technology Policy director Michael Kratsios made the accusation Wednesday on X, claiming Moonshot "acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models." He also alleged Moonshot ran a "sophisticated internal platform" for "large scale distillation against US models," specifically Anthropic's Fable — training a cheaper model on a rival's outputs. Treasury Secretary Scott Bessent followed up on Fox Business, warning that "sanctions and Entity List designations will be on the table" for Chinese firms that conduct "covert, industrial-scale distillation attacks that cross the line into IP theft."
The problem: Kratsios published no evidence. TNW noted that directly. The accusations land days after Moonshot released Kimi K3 on July 16 — a 2.8-trillion-parameter open-weight model that climbed to the top of LMArena's Frontend Code Arena leaderboard, overtaking Anthropic's Claude Fable 5, and ranked third overall on Artificial Analysis' Intelligence Index behind only Fable 5 and OpenAI's GPT-5.6 Sol. The model is already out. Anyone can download it. CNBC reported that a Chinese embassy spokesperson called the claims "baseless allegations and malicious smears."
The market panicked anyway. The Philadelphia Semiconductor Index fell 12.5% in a week — its worst stretch in 15 months. TechSpot reported that Taiwan's benchmark dropped more than 6%, Japan closed down 4%, and Chinese AI stocks actually got hit harder, with Zhipu down as much as 30%.
Nvidia CEO Jensen Huang pushed back hard. "These Chinese models are excellent," he told Axios at a Texas plant opening. "Open-source models that are excellent should be used." Asked whether China could displace American labs: "Zero possibility." His argument is straightforward — free, capable models drive more usage, and more usage means more chips running somewhere. He also flipped the security argument: open weights can be inspected and run offline. "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 even called on Anthropic to open its restricted cybersecurity model Claude Mythos, arguing that "holding Anthropic back is not in the benefit of the United States."
Washington's answer is more control. The bipartisan Remote Access Security Act, which would expand export restrictions to cover cloud-based remote access to restricted chips, passed the House in January and awaits Senate action. CNBC reported that the Tony Blair Institute's Keegan McBride said the allegations "demonstrate just how far ahead the US is on compute" — an odd concession buried in the enforcement push. Meanwhile, ByteDance was reportedly assembling computing capacity outside China months ago. The loopholes were never secret.
The pattern is clear: restrict, leak, panic, restrict again. Moonshot is now closing a funding round at up to $50 billion with a possible Hong Kong listing. The model already beat ours. The question isn't whether China can compete — it's whether Washington's reflex to choke American innovation with more rules will accomplish anything except leaving U.S. developers hamstrung while the rest of the world moves on.