Export-restricted Nvidia chips found in Chinese AI models

Leading open-weight artificial intelligence models created by Chinese firms were trained using Nvidia hardware that the United States government prohibited from being sent to its primary geopolitical rival, according to a new research study. The widespread availability of Blackwell GPUs among Chinese frontier laboratories highlights the significant challenge facing the prominent chip designer.

Nvidia CEO Jensen Huang departed the company this week after acknowledging an affair with Alice Huang, an executive at Megaspeed, a firm implicated in smuggling Nvidia chips into China. Alice Huang has also been accused of acting as an agent for the intelligence service of China. Although Jensen Huang’s personal connections to a customer and two employees gained public attention, Nvidia’s potential corporate liability regarding export control breaches likely influenced the decision to part ways with him.

United States export regulations aim to safeguard the nation’s AI leadership against China, yet Huang consistently worked to expand access to that market. Blackwell GPUs are currently banned from export to China or Chinese-owned businesses, while the Trump administration permits Nvidia to export the less powerful H200 chip, provided the shipments follow a circuitous route allowing the U.S. to claim a 25% tariff.

Will DePue, a former OpenAI researcher who leads the nonprofit group Unemployment Labs, examined the weights—the unique structure—of open-weight models from DeepSeek, Moonshot AI, MiniMax and Ant Group. The weights retain the configurations of the chips used for training, and DePue discovered evidence that models such as Moonshot’s Kimi’s K3 and DeepSeek’s V4 carry fingerprints of training on Blackwell hardware.

White House technology adviser Michael Kratsios stated earlier this year that Moonshot had utilized export-banned Nvidia chips, referencing U.S. intelligence sources. The nonprofit research organization Epoch AI estimates that approximately two million advanced chips entered China during 2024 and 2025 through schemes intended to conceal their final destination.

This does not confirm the models were trained exclusively on Nvidia chips, or that export laws were breached; chips could be located in offshore data centers, a popular method for Chinese firms to access compute power that has driven an industrial revolution in Malaysia. Epoch’s data estimates 275,000 Nvidia AI accelerators were imported to Malaysia in 2024 and 2025.

However, the apparent goal of export bans is to prevent Chinese AI laboratories from training on the most powerful chips. China hawks have long demanded stricter controls on chip usage rather than just ownership or installation locations.

“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,” Sam Bresnick, a researcher at the Georgetown Center for Security and Emerging Technology, told TechCrunch this summer.

Megaspeed, led by Alice Huang, has been central to this trade for years through its subsidiary Speedmatrix and has been investigated by both the U.S. and Singaporean governments. On Friday, Bloomberg reported that a company linked to Alice Huang was involved in smuggling banned Nvidia chips to a Shanghai data center.

In May, the short-selling research firm Culper published a report arguing Nvidia faced a “China problem” because significant revenue relied on Chinese demand, either through remote data centers or illegal smuggling. It detailed Megaspeed’s connection to Nvidia and alleged Alice Huang’s companies are backed by Chinese firms barred from accessing U.S. chips.

Ironically, Culper’s bet against Nvidia was not driven by expectations of a U.S. government crackdown. Instead, it focused on the Chinese government’s response to the bans, which was to recommit to its domestic chip industry.

If China succeeds, billions of dollars in Chinese demand for Nvidia chips could vanish. Approximately 600,000 H200 chips are reportedly stockpiled in Hong Kong, awaiting shipment into China. While Chinese companies may desire the chips, observers have no consensus on why they have not been cleared for import.

Graham Webster, a research scholar at Stanford focusing on U.S.-China technology policy, said he does not believe that withholding the chips will drive domestic production, given the massive demand. Other theories for the delay include preparation for a trade deal involving the import of large volumes of U.S. goods, or concerns regarding potential surveillance technology in the chips. For now, U.S. companies have limited obligations to track what their customers do with components.