Banned Nvidia Blackwell Chips Trained China's Top AI Models

Chinese frontier AI labs trained their most-watched models on Nvidia Blackwell chips that Washington bans from export to China, according to new research from the nonprofit Unemployment Labs. The finding lands days after Nvidia CEO Jensen Huang left the company, and it puts a hard number on a smuggling problem the chip designer can no longer contain.

Will DePue, a former OpenAI researcher who leads Unemployment Labs, analyzed the weights—the unique internal structure—of open large language models released by DeepSeek, Moonshot AI, MiniMax and Ant Group. Model weights still reflect the configuration of the chips used in training. DePue found compelling evidence that Moonshot's Kimi K3 releases and DeepSeek's V4 models carry fingerprints of Blackwell hardware, the GPU generation barred from export to China or Chinese-owned firms.

The claim is not new at the highest levels. White House technology adviser Michael Kratsios said earlier this year that Moonshot had used export-banned Nvidia chips, citing US intelligence sources. Epoch AI, a nonprofit research organization, estimates roughly two million advanced chips entered China in 2024 and 2025 through schemes built to hide their final destination. Spread over two years, that is about a million chips a year moving through channels US law was written to close.

Why did Nvidia part ways with Jensen Huang?

Huang departed this week after admitting to an affair with Alice Huang, an executive at Megaspeed, a company implicated in smuggling Nvidia chips into China. Alice Huang has also been alleged to be an agent of China's intelligence service.

The affair drew the headlines. The corporate exposure likely drove the decision. Nvidia's potential liability for export control violations factored into the split, given Huang's relationships with a customer and two employees.

Megaspeed has sat at the center of the diverted chip trade for years through its subsidiary Speedmatrix, and both the US and Singaporean governments have investigated it. On Friday, Bloomberg reported that a company linked to Alice Huang was involved in smuggling banned Nvidia chips to a data center in Shanghai. Nvidia and Huang have both vouched for Megaspeed's data centers, saying Nvidia officials visited them to confirm they were legitimate—a defense that weakens considerably if investigators confirm smuggling and Huang's awareness of it.

Does training abroad break the law?

Not necessarily, and the research does not claim it does. The chips could have sat in offshore data centers, a popular route for Chinese firms seeking compute—one that has driven what the source calls an industrial revolution in Malaysia. Epoch AI's data estimates 275,000 Nvidia AI accelerators were imported to Malaysia in 2024 and 2025.

Yet offshore loopholes defeat the stated purpose of the bans: keeping the most powerful chips out of Chinese AI training. Sam Bresnick, a researcher at the Georgetown Center for Security and Emerging Technology, told TechCrunch this summer: "If we care about the U.S. tech edge, it doesn't make any sense to me that you would not allow Chinese companies to access chips in mainland China or Hong Kong and Macau, but then say, actually, if you just go over to Malaysia, you can do whatever you want."

China hawks have long demanded controls on who uses chips, not merely who owns them or where they sit. For now, US companies face limited obligations to track what customers do with controlled goods like AI chips. The Unemployment Labs findings are an argument that this gap is the story.

Where does the commercial risk sit?

In May, the short-selling research firm Culper published a report arguing Nvidia had a "China problem": a significant share of its revenue flows from Chinese demand, either through remote data centers or smuggling in violation of US law. Culper detailed Megaspeed's relationship with Nvidia and alleged that Alice Huang's companies are backed by Chinese firms barred from accessing US chips.

Culper's bet, notably, was not a wager on a US crackdown. It targeted Beijing's response: a renewed commitment to a domestic chip industry. If that succeeds, billions of dollars in Chinese demand for Nvidia silicon could evaporate.

The immediate evidence of a frozen market is visible in Hong Kong, where some 600,000 H200 chips are reportedly stockpiled awaiting clearance into China. Observers have no consensus on why they sit. Graham Webster, a Stanford research scholar focused on US-China technology policy, said he does not buy the theory that withholding the chips would accelerate domestic production, since chip demand is simply too large. Other explanations include preparation for a trade deal involving large-volume US imports, or concerns about surveillance technology inside the chips.

The policy machinery itself is also on trial. Secretary of Commerce Howard Lutnick, who oversees the agencies licensing chip exports, could reportedly lose his job after reports that his investigators failed to flag suspicious chip buyers and grew too close to Nvidia officials. Until Washington tightens end-use rules—or proves a smuggling case against Nvidia's former chief—Malaysia will keep looking like the cheapest loophole in the export regime.