Chinese military seeks restricted Nvidia GPUs for AI and robotics amid U.S. policy debate

The Chinese military has been actively attempting to procure high-end Nvidia graphics processing units (GPUs) for use in AI servers and robotics, according to procurement records from the People’s Liberation Army (PLA) bidding portal [1]. Over the past year, multiple requests were made for both restricted and permitted models, including the U.S.-banned H100, RTX 6000, and H20 chips [1]. Notably, three bids published in April explicitly requested the banned chips to support servers capable of running advanced AI models such as DeepSeek-R1 671B [1].

One of the procurement requests in April called for the H20 GPU, while another sought four RTX 6000 GPUs as part of an intelligent decision-making support system [1]. A provisional supplier for the RTX 6000 was identified in June, though the contract had not yet been finalized [1]. In the same month, the PLA also requested H100 chips, which have been restricted since 2022. The military emphasized the need for original packaging and on-site installation, highlighting the strategic importance of these chips [1]. A similar H100 request was made again in July 2024 [1].

In another project, the PLA sought to equip a 33-pound robotic dog with an Nvidia Jetson computing module for training purposes [1]. Although this request was later canceled, the Jetson module is not typically restricted for most applications. It remains unclear whether the PLA successfully obtained any of the requested chips [1].

The continued interest in restricted technology has drawn attention from U.S. security experts. Ryan Fedasiuk, a former U.S. State Department advisor, explained that the PLA often uses indirect means—including subsidiaries, shell companies, and third-party intermediaries—to acquire U.S. technology [1]. Fedasiuk emphasized that these methods allow the PLA to bypass direct procurement restrictions [1]. A Nvidia spokesperson has dismissed concerns, stating that China has sufficient domestic alternatives and that the small-scale purchase of older chips poses no national security risk [1]. The spokesperson also noted that restricted products would lack ongoing support and maintenance [1].

The U.S. government recently lifted the export ban on the H20 GPU, effectively allowing Nvidia to resume sales to China [1]. However, the decision has sparked pushback from security officials. A group of 20 U.S. policymakers has urged the Commerce Department to reinstate the ban, citing concerns that the chips could be used to bolster China’s military capabilities [1]. Craig Singleton of the Foundation for Defense Democracies warned that the policy shift could open the floodgates to increased technology transfers [1]. The Commerce Department has maintained that it has already imposed tighter controls and will continue to review license applications with care [1].

Nvidia has also faced scrutiny in China, where it was summoned by the country’s internet regulator to address allegations of backdoor security risks in its H20 AI chips. The company denied the claims [1]. In a recent interview, Nvidia CEO Jensen Huang stated that the Chinese military would likely avoid relying on U.S. technology due to its unpredictability [1]. The Chinese Ministry of National Defense has not yet commented on the issue [1].

Ryan Fedasiuk, who reviewed some of the procurement documents, noted that the PLA and Chinese AI firms favor Nvidia hardware due to its superior processing power [1]. He argued that no domestic or foreign competitor has yet matched Nvidia’s technological edge [1].

Source:

[1] https://coinmarketcap.com/community/articles/688ce21736d7940fc7598d1e/