Export-banned Nvidia chips are behind China’s most powerful AI models—uh oh, Jensen!

A new research effort claims that the leading open-weight AI models from Chinese companies were trained on Nvidia chips that the US government bars from shipping to China. The finding, if it holds up, suggests restricted Blackwell GPUs are in wide use at Chinese frontier labs and underscores the limits of hardware-based export controls.

Will DePue, a former OpenAI researcher who now leads the nonprofit Unemployment Labs, examined the weights — the underlying configuration — of openly released models from DeepSeek, Moonshot AI, MiniMax and Ant Group. Because weights retain traces of the chips used in training, DePue says he found compelling evidence of Blackwell training fingerprints in closely watched releases, including Moonshot's Kimi K3 and DeepSeek's V4. The analysis does not establish that any single model ran entirely on Nvidia hardware, nor that laws were broken; compute could have been rented from offshore data centers.

Independent estimates add scale to the claim. White House technology adviser Michael Kratsios said earlier this year that Moonshot used export-banned Nvidia chips, citing US intelligence sources, while nonprofit Epoch AI reckons roughly two million advanced chips reached China across 2024 and 2025 through routing schemes that obscure final destinations. Epoch's data also shows 275,000 Nvidia AI accelerators imported into Malaysia in the same two-year span, matching accounts that Chinese firms increasingly buy compute abroad rather than at home.

That offshoring pattern sits awkwardly with the stated aim of the bans. Blackwell GPUs are prohibited for export to China or Chinese-owned firms, while the Trump administration permits the weaker H200 to ship along a circuitous route that lets Washington claim a 25% tariff. Critics argue that geography-based rules miss the point: Sam Bresnick of the Georgetown Center for Security and Emerging Technology told TechCrunch this summer that it makes little sense to block chips in mainland China and Hong Kong while permitting unrestricted use just across a border in Malaysia. China hawks have pressed for controls tied to who uses the compute, not merely who owns the hardware or where it is installed.

The chip question overlaps with upheaval at Nvidia itself. 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; she has also been alleged to be an agent of China's intelligence service. TechCrunch reports that potential corporate liability for export violations likely weighed on the decision alongside the personal matters. Megaspeed has moved restricted chips for years through its subsidiary Speedmatrix and faces investigations by US and Singaporean authorities, and Bloomberg reported Friday that a company linked to Alice Huang smuggled banned Nvidia chips to a data center in Shanghai.

What comes next turns on unresolved questions: whether US intelligence-backed estimates trigger new restrictions on chip use rather than ownership, how thoroughly offshore hubs like Malaysia are policed, and how much legal exposure Nvidia faces as investigations deepen. Builders watching the space should expect continued churn in where frontier compute can legally run — and continued ambiguity about which restrictions actually bind.