Export‑banned Nvidia chips found in training of Chinese AI models

Export‑banned Nvidia chips found in training of Chinese AI models

Lead – According to TechCrunch, a nonprofit research effort has identified that several leading open‑weight AI models developed by Chinese companies were trained on Nvidia’s Blackwell GPUs, which are prohibited from export to China under U.S. regulations. The findings come as Nvidia’s chief executive Jensen Huang stepped down amid personal and corporate controversies linked to alleged chip smuggling.

Background

TechCrunch reports that the United States has barred the export of Nvidia’s Blackwell GPUs to China and to Chinese‑owned firms. The ban is intended to preserve the U.S. lead in artificial‑intelligence capabilities. Under the Trump administration, Nvidia was allowed to ship the less powerful H200 accelerator to China only if the shipment followed a circuitous route that enabled the United States to levy a 25 % tariff.

Research findings

Will DePue, a former OpenAI researcher who now leads the nonprofit Unemployment Labs, examined the weight matrices of large‑language models released openly by DeepSeek, Moonshot AI, MiniMax and Ant Group. According to TechCrunch, the structure of these weights “reflects the configurations of the chips used to train them,” and DePue identified “compelling evidence” that the most closely watched models—including Moonshot’s Kimi K3 releases and DeepSeek’s V4 models—bear the distinctive “fingerprints” of training on Blackwell GPUs.

White House technology adviser Michael Kratsios, as cited by TechCrunch, confirmed earlier this year that Moonshot had employed export‑banned Nvidia chips, citing U.S. intelligence sources.

Scale of alleged violations

Epoch AI, a nonprofit research organization, estimates that about two million advanced Nvidia chips entered China in 2024 and 2025 through schemes designed to conceal their final destination. The organization also notes that 275,000 Nvidia AI accelerators were imported to Malaysia during the same period, a route that has become popular for Chinese firms seeking high‑performance compute.

TechCrunch adds that while the presence of these chips does not prove that entire models were trained exclusively on Nvidia hardware, the data suggest that Chinese AI labs have accessed the most powerful accelerators despite the export ban. The report also mentions that 600,000 H200 chips are reportedly stockpiled in Hong Kong, awaiting shipment into China, though the reasons for the hold‑up remain unclear.

Corporate and political fallout

The investigation coincides with the departure of Nvidia CEO Jensen Huang, who, according to TechCrunch, left the company after admitting an affair with Alice Huang, the executive of Megaspeed. Alice Huang has been alleged to be an agent of China’s intelligence services and to have overseen the smuggling of Nvidia chips into China through Megaspeed’s subsidiary Speedmatrix. Both U.S. and Singaporean authorities have reportedly investigated Megaspeed’s activities.

Bloomberg, as referenced by TechCrunch, reported that a company linked to Alice Huang was involved in smuggling banned Nvidia chips to a data center in Shanghai. In May, short‑selling research firm Culper published a report describing a “China problem” for Nvidia, arguing that a sizable share of the company’s revenue derives from Chinese demand, whether through legitimate remote data centers or illicit smuggling. Culper’s analysis suggested that the Chinese government’s response to the bans could accelerate domestic chip development, potentially curtailing future Nvidia sales.

Regulatory context

TechCrunch notes that U.S. companies currently have limited obligations to monitor the end‑use of controlled goods such as AI chips. However, the personal and professional connections between Jensen Huang and Alice Huang could expose Nvidia to heightened scrutiny if investigators confirm that smuggling occurred with the CEO’s knowledge.

The report also references comments from Graham Webster, a research scholar at Stanford who focuses on U.S.–China technology policy, as quoted by TechCrunch. Webster expressed skepticism that withholding chips would spur domestic production, given the magnitude of demand, and suggested alternative explanations for the Hong Kong stockpile, including preparation for a potential trade deal or concerns about embedded surveillance technology.

U.S. Secretary of Commerce Howard Lutnick, who oversees export‑licensing agencies, is reported by TechCrunch to face possible dismissal after allegations that his investigators failed to identify suspicious chip buyers and were overly familiar with Nvidia officials.

Analysis (Opinion)

The convergence of technical evidence, corporate misconduct, and geopolitical tension underscores a growing enforcement challenge for U.S. export controls. If the findings reported by TechCrunch are accurate, they reveal a sophisticated supply‑chain workaround that undermines policy objectives. The situation also highlights the risk that personal relationships at the executive level can exacerbate compliance vulnerabilities, potentially prompting stricter oversight of semiconductor exporters.

Conclusion

According to TechCrunch, the discovery that Chinese AI models were trained on export‑banned Nvidia Blackwell GPUs adds a new layer of complexity to the United States’ effort to limit China’s access to cutting‑edge AI hardware. The episode coincides with leadership changes at Nvidia and ongoing investigations into alleged chip smuggling networks. How U.S. authorities respond may shape the future of semiconductor export policy and the competitive dynamics of the global AI market.

Source: TechCrunch

Frequently Asked Questions

Which Nvidia chips are prohibited from export to China?

According to TechCrunch, Nvidia’s Blackwell GPUs are banned from export to China or Chinese‑owned firms, while the H200 chip may be shipped only via a circuitous route that incurs a 25 % tariff.

What evidence links Chinese AI models to Blackwell GPUs?

TechCrunch reports that researcher Will DePue analyzed the weight structures of several open‑weight models and found “compelling evidence” that they bear the distinctive fingerprints of training on Blackwell GPUs.

Why did Jensen Huang leave Nvidia?

According to TechCrunch, Jensen Huang stepped down after admitting an affair with Alice Huang, whose companies have been implicated in smuggling Nvidia chips into China and alleged ties to Chinese intelligence.

Source: TechCrunch — Read more

💻 Related Articles