Leading open-weight models developed by Chinese companies are running on Nvidia chips that the US government has banned from shipping to its largest geopolitical rival, a new research effort claims.
The seemingly widespread access to Blackwell GPUs among Chinese border laboratories shows the scale of the problem facing the leading chip designer. Nvidia CEO Jensen Huang left the company this week after admitting to an affair with Alice Huang, the CEO of Megaspeed, a company involved in smuggling Nvidia chips into China. Alice Huang is also alleged to be an agent of the Chinese intelligence service.
While Jensen Huang’s relationships with a client and two employees have attracted the most public attention, Nvidia’s potential corporate liability for export control violations likely factored into its decision to part ways with Huang.
US export controls are intended to protect the country’s lead in AI capabilities against China, but Hwang has long sought to open up that market. Exports of Blackwell’s GPUs to China or Chinese-owned companies are prohibited, while the Trump administration is allowing Nvidia to export the H200 chip, a less powerful chip, on the condition that shipments take a circuitous shipping route that allows the US to demand a 25% tariff.
Will DePue, a former OpenAI researcher who leads the nonprofit research group Unemployment Labs, Weight analysis – Unique architecture – for large language models publicly released by DeepSeek, Moonshot AI, MiniMax, and Ant Group. The weights still reflect the chip configurations used to train them, and DePue has been able to find compelling evidence that the most closely monitored models, including Moonshot’s Kimi’s K3 and DeepSeek V4 models, bear fingerprints from training on Blackwell chips.
White House technology adviser Michael Kratsios said earlier this year that Moonshot used Nvidia chipsets banned for export, citing US intelligence sources. Epoch AI, a non-profit research organization, estimates that about 2 million advanced chips will enter China in 2024 and 2025. Through diagrams Designed to hide their final destination.
This does not mean that these models were entirely trained on Nvidia chips, or that export control laws were necessarily violated; The chips would have been located in an offshore data center, which has proven to be a popular way for Chinese companies to access computing and Led the industrial revolution In Malaysia. Epoch data estimates that 275,000 Nvidia AI accelerators will be imported into Malaysia in 2024 and 2025.
However, the ostensible goal of the export ban is to prevent just that: training Chinese AI labs on the most powerful chips. China hawks have long called for stronger controls on who uses chips, not just who owns them or where they are installed.
“If we care about American technological superiority, it doesn’t make sense to me that you wouldn’t allow Chinese companies access to chips in mainland China or Hong Kong and Macau, but then, in fact, if you go to Malaysia, you can do whatever you want,” Sam Bresnik, a researcher at the Georgetown Center for Security and Emerging Technology, told TechCrunch this summer.
Megaspeed, headed 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 United States and the United States. SIcom. ngaporean government. Friday, Bloomberg Also mentioned That a company with links to Alice Huang was involved in smuggling banned Nvidia chips into a data center in Shanghai.
In May, short-selling research firm Culper published a report arguing that Nvidia had “China problem“Because a significant share of its revenue was driven by Chinese demand, either through remote data centers or smuggling in violation of US law.” It detailed Megaspeed’s relationship with Nvidia and alleged that Alice Huang’s companies are backed by Chinese companies blocked from accessing US chips.
Ironically, Culber’s short-selling bet against Nvidia was not driven by expectations of a US government crackdown. Instead, it focused on the Chinese government’s response to the ban, which was to recommit to the domestic chip industry.
If China succeeds, billions of dollars in Chinese demand for Nvidia chips could dry up. There are about 600,000 H200 chips It is said stored In Hong Kong, waiting to ship to China. While Chinese companies may want the chips, observers have no consensus on why they should not be allowed to be imported.
Graham Webster, a researcher at Stanford University who focuses on technology policy between the United States and China, said he doesn’t buy the idea that curbing chips would drive domestic production, because demand for chips is so great. Other theories for the delay include preparation for a potential trade deal that might involve importing a large amount of US goods, or concerns about potential surveillance technology in the chips.
Currently, US companies have limited obligations to track what their customers do with controlled goods such as AI chips. However, the personal and professional entanglements between Jensen Huang and Alice Huang create a much bigger problem for Nvidia if government investigators confirm that a chip smuggling operation occurred and that he was aware of it. Both Nvidia and Huang have confirmed the Megaspeed data centers, saying Nvidia officials have visited them to confirm their legitimacy.
The US government’s treatment of Nvidia and Megaspeed is also under scrutiny. Commerce Secretary Howard Lutnick, who oversees agencies that license chip exports, could reportedly lose his job after reports that his investigators… Failed to detect Suspicious chip buyers were Very comfortable With Nvidia officials.
When you buy through links in our articles, we may earn a small commission. This does not affect our editorial independence.
