The Core Finding
A new research effort claims that the leading open-weight models developed by Chinese companies were trained on Nvidia chips that the U.S. government has restricted from shipping to its largest geopolitical rival. The apparently widespread access to Blackwell GPUs among Chinese frontier labs underscores the magnitude of the challenge facing Nvidia.
Jensen Huang's Departure
Nvidia CEO Jensen Huang left the company this week after admitting to an affair with Alice Huang, an executive at Megaspeed β a company implicated in smuggling Nvidia chips into China. Alice Huang has also been alleged to be an agent of China's intelligence service. While Jensen Huang's relationships with a customer and two employees have drawn the most public interest, Nvidia's potential corporate liability for export control violations likely factored into its decision to part ways with him.
The Export Control Landscape
U.S. export controls are intended to preserve American leadership in AI capabilities vis-Γ -vis China, but Huang has long sought to open up that market. Blackwell GPUs are banned from export to China or Chinese-owned firms. Meanwhile, the Trump administration allows Nvidia to export the H200 β a less powerful chip β provided shipments take a circuitous route that allows the U.S. to claim a 25% tariff.
Fingerprints in the Weights
Will DePue, a former OpenAI researcher who leads the nonprofit research group Unemployment Labs, analyzed the weights β the unique structure β of large language models released openly by DeepSeek, Moonshot AI, MiniMax, and Ant Group. Model weights still reflect the configurations of the chips used to train them. DePue found compelling evidence that the most closely watched models, including Moonshot's Kimi K3 releases and DeepSeek's V4 models, bear fingerprints from training on Blackwell chips.
White House technology adviser Michael Kratsios said earlier this year that Moonshot had 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.
The Offshore Loophole
That doesn't mean these models were trained entirely on Nvidia chips, or that export control laws were necessarily violated. The chips could have been located in an offshore data center β a popular way for Chinese firms to access compute, and one that has driven an industrial revolution in Malaysia. Epoch's data estimates that 275,000 Nvidia AI accelerators were imported to Malaysia in 2024 and 2025.
However, the ostensible goal of the export bans is to prevent exactly that: Chinese AI labs training on the most powerful chips. China hawks have long called for stronger controls on who uses chips β not just who owns them or where they are installed.
"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 at the Center
Megaspeed, the company led by Alice Huang, has been at the center of that trade for several years through its subsidiary Speedmatrix, and has been the subject of investigations by both the U.S. and Singaporean governments. On Friday, Bloomberg reported that banned Nvidia chips were installed at a China data center linked to Alice Huang.
via TechCrunch AI