Major open models developed by Chinese companies were trained on Nvidia chips that the U.S. government restricted shipping to its biggest geopolitical rival, according to a new research effort.
The seemingly widespread access to Blackwell GPUs among China’s border labs shows the scale of the problem facing the leading chip designer. Nvidia CEO Jensen Huang left the company this week after admitting to his affair with Alice Huang, head of Megaspeed, a company involved in smuggling Nvidia chips into China. Alice Huang is also said to be a Chinese intelligence agent.
While Jensen Huang’s dealings with a customer and two employees have generated the most public interest, Nvidia’s potential liability for export control violations likely factored into its decision to part ways with Huang.
U.S. export controls are intended to protect the country’s leadership in AI capabilities against China, but Huang has long sought to open up that market. Blackwell GPUs are banned from export to China or Chinese companies, while the Trump administration is allowing Nvidia to export the H200, a less powerful chip, provided shipments take a circuitous shipping route that allows the United States to demand a 25% tariff.
Will DePue, a former OpenAI researcher who leads the nonprofit research group Unemployment Labs, analyzed the weights — the unique structure — of large language models published openly by DeepSeek, Moonshot AI, MiniMax and Ant Group. The weights always reflect the configurations of the chips used to drive them, and DePue was able to find convincing evidence that the most closely watched models, including Moonshot’s K3 versions of Kimi and DeepSeek’s V4 models, bear the fingerprints of training on Blackwell chips.
White House technology adviser Michael Kratsios said earlier this year that Moonshot used export-banned Nvidia chips, citing U.S. intelligence sources. Epoch AI, a nonprofit research organization, estimates that about two million advanced chips entered China in 2024 and 2025 through schemes designed to hide their final destination.
This does not mean that these models were trained entirely on Nvidia chips, or that export control laws were necessarily violated; the chips could have been in an offshore data center, which proved a popular way for Chinese companies to access computing and led to an industrial revolution in Malaysia. Epoch data estimates that 275,000 Nvidia AI accelerators will be imported into Malaysia in 2024 and 2025.
However, the apparent aim of the export bans is precisely to prevent this: Chinese AI labs train on the most powerful chips. China hawks have long called for tighter controls on who uses the chips, not just who owns them or where they are installed.
“If we care about U.S. technological advantage, it makes no sense to me that you wouldn’t allow Chinese companies access to chips in mainland China and Hong Kong and Macau, but then say, basically, if you just go to Malaysia, you can do whatever you want,” Sam Bresnick, a researcher at the Georgetown Center for Security and Emerging Technology, told TechCrunch this summer.
Megaspeed, the company headed by Alice Huang, has been at the center of these exchanges for several years through its subsidiary Speedmatrix and is the subject of investigations by the American and Singaporean governments. On Friday, Bloomberg also reported that a company with ties to Alice Huang was involved in smuggling banned Nvidia chips to a data center in Shanghai.
In May, short-selling research firm Culper released a report claiming that Nvidia had a “China problem” because a significant portion of its revenue came from Chinese demand, either through remote data centers or through smuggling in violation of U.S. law. It details Megaspeed’s relationship with Nvidia and claims that Alice Huang’s companies are backed by Chinese companies that do not have access to American chips.
Ironically, short seller Culper’s bet against Nvidia was not motivated by the expectation of a crackdown by the US government. Instead, he focused on the Chinese government’s response to the bans, which was to reaffirm its commitment to its domestic chip industry.
If China succeeds, China’s billion-dollar demand for Nvidia chips could dry up. Some 600,000 H200 chips are said to be stored in Hong Kong, awaiting shipment to China. Although Chinese companies might want the chips, observers can’t agree on why they haven’t been approved for import.
Graham Webster, a researcher at Stanford who focuses on U.S.-China technology policy, said he doesn’t buy the idea that holding back chips would boost domestic production, given that demand for chips is so great. Other theories about the hold-up include preparations for a possible trade deal that would involve importing a large volume of U.S. products, or concerns about possible surveillance technology in the chips.
For now, U.S. companies have limited obligations to track what their customers are doing with controlled products like AI chips. However, Jensen Huang’s personal and professional ties to Alice Huang create a much bigger problem for Nvidia if government investigators confirm that chip smuggling occurred and that he knew about it. Nvidia and Huang vouched for Megaspeed’s data centers, saying Nvidia officials visited them to confirm they were legitimate.
The US government’s treatment of Nvidia and Megaspeed is also under scrutiny. Commerce Secretary Howard Lutnick, who oversees agencies that authorize chip exports, could lose his job after reports that his investigators failed to uncover suspicious chip buyers and were too cozy with Nvidia officials.
When you purchase through links in our articles, we may earn a small commission. This does not affect our editorial independence.
Gn bussni
