Nvidia export ban crisis as smuggled chips power China's AI
A research report reveals China's top AI models were trained on export-restricted Nvidia chips, amid CEO Jensen Huang's resignation over ties to an alleged smuggling operation.

Export-banned Nvidia chips are behind China’s most powerful AI models. The revelation comes as Nvidia co-founder and former CEO Jensen Huang left the company after admitting an affair with Alice Huang, an executive at Megaspeed, a firm implicated in smuggling Nvidia hardware to China. Huang's personal and professional ties could create larger liability for Nvidia if smuggling is confirmed, and his departure may have been driven by these revelations and widespread illicit exports, according to analysis firm SemiAnalysis.
Nvidia's board requested Huang's resignation to distance the company from the controversies. The timing coincides with congressional efforts to advance the Remote Access Security Act (RASA), which aims to extend U.S. export-control authority to remote access of AI hardware hosted outside China. Nvidia and Oracle have actively opposed this legislation. Some analysts speculate that U.S. intelligence community pressure to remove Huang may have stemmed from his opposition to more aggressive restrictions on China's access to American AI hardware, though no direct evidence confirms this.
China's AI model surge using restricted chips
Chinese AI developers have leveraged banned Nvidia chips to train new models despite export controls. Models from leading labs like DeepSeek, Moonshot AI, MiniMax, and Ant Group were trained on restricted Nvidia chips. Will DePue, former OpenAI researcher and head of Unemployment Labs, found that models such as Moonshot’s Kimi K3 and DeepSeek’s V4 reflect training on advanced Blackwell chips. White House technology adviser Michael Kratsios stated that Moonshot used export-banned Nvidia chips, citing U.S. intelligence sources.
SemiAnalysis estimates that China received a massive volume of restricted hardware. Their power calculations suggest the following scale of deployment:
| Model Type | Estimated Quantity | Power per Unit |
|---|---|---|
| H200 SXM GPUs | ~2.142 million | 700W |
| B200 SXM GPUs | ~1.5 million | 1,000W |
This means over 1.5 gigawatts of Nvidia hardware subject to U.S. export restrictions ended up in China in recent years, a figure that excludes processors specifically developed to comply with U.S. rules like the H20 or H800. For comparison, xAI's Colossus One used 220-230,000 Hopper H100/H200 GPUs, while its Colossus 2 project targets 1.2 million Blackwell B200/B300 GPUs. The restricted Nvidia hardware in China is comparable to or exceeds xAI's entire existing operation.
Scale of chip smuggling and transshipment
Massive volumes of restricted Nvidia hardware have entered China through complex smuggling networks and third-country transshipment. Research group Epoch AI estimates about two million advanced chips entered China in 2024-2025 via concealment schemes. They also estimate 275,000 Nvidia AI accelerators were imported to Malaysia in the same period, a common transshipment point. Approximately 600,000 H200 chips are currently stockpiled in Hong Kong awaiting shipment to mainland China. The U.S. Department of Commerce reportedly shares SemiAnalysis' assessment of the scale of these unauthorized shipments.
Specific legal cases highlight the methods. One case from March 2026 charges individuals including Wally Liaw, Steven Chang, and Willy Sun with diverting $510 million worth of equipment to China via Southeast Asia. Another involves nine individuals linked to Nvidia, Albatron, Chief Telecom, Flying Tiger, and Supermicro in Taiwan who arranged unauthorized exports of 130 B300-based servers, 74 of which reached China.
Legal and financial implications for Nvidia
The smuggling allegations and Nvidia's revenue dependence on China expose the company to significant legal and financial risks. Culper Research published a report claiming Nvidia’s revenue is heavily driven by Chinese demand, either through remote data centers or illegal smuggling. An alleged report claims Alice Huang, Jensen Huang's extramarital partner and former Megaspeed executive, facilitated the smuggling of $7.5 billion of Nvidia AI hardware, around 240,000 GPUs mostly H100/H200 models, into China. Alice Huang is also alleged to be an agent of China’s intelligence service.
Megaspeed, led by Alice Huang, has been investigated by U.S. and Singaporean authorities for chip smuggling. Bloomberg reported a company linked to Alice Huang smuggled banned Nvidia chips to a Shanghai data center. Nvidia and Huang have previously defended Megaspeed’s data centers, stating officials visited them to verify legitimacy. U.S. companies currently have limited obligations to track the end use of controlled AI chips, a loophole that may be addressed by new legislation. US Commerce Secretary Howard Lutnick may lose his job after investigators failed to uncover suspicious chip buyers and were perceived as too close to Nvidia officials.
SemiAnalysis warns that complete restrictions on Chinese access to American AI accelerators could encourage domestic development of Chinese accelerators, undermine international AI software collaboration, and weaken American high-tech companies. Chinese developers have created open-source technologies like DeepEP, DeepGEMM, and MSA that improve AI workload efficiency on Nvidia hardware, benefiting global developers. American companies depend on Chinese engineers, component suppliers, and manufacturers. Legal curbs restricting Chinese access could undermine this international cooperation and stall open-source optimizations. The firm advocates for a system where Chinese developers can rent computing capacity from trusted providers outside China under U.S. oversight to preserve American technological influence and limit China's independent acquisition of advanced AI infrastructure.





