Chinese universities bypass US export bans on Nvidia AI chips

Bypassing the blacklist

A Chinese university with deep ties to the nation's military has successfully acquired advanced Nvidia artificial intelligence chips. This acquisition occurred despite the institution being on a United States export blacklist designed to prevent the flow of sensitive technology. The university, Northwestern Polytechnical, purchased two H200 processors for 874,700 yuan ($123,000) in October.

The procurement was handled by a domestic Chinese technology supplier rather than a direct sale from Nvidia. This indicates that the institution used a middleman to navigate restrictions that generally block American companies from supplying controlled hardware to the school. The purchase was specifically made by the university's management school to support its technological goals.

The US Commerce Department first added this university to its Entity List in 2001 due to its alleged involvement in missile proliferation. While Washington tightened these rules in 2019 to deny most licenses for controlled technology, the recent purchase shows a significant gap in enforcement. The university remains a key player in research involving military aircraft, missiles, and satellites.

The role of rented servers

Beyond direct hardware purchases, the investigation found that Chinese universities are using rented computing capacity to access restricted chips. For example, the university's cybersecurity school signed contracts in January to access servers containing eight H200 chips each. These contracts were used specifically for training reinforcement learning and fine-tuning artificial intelligence models.

The university also rented access to servers containing Nvidia A100 and A800 chips, both of which are currently banned from export to Chinese entities. In April, a different university department secured a 375,000-yuan deal for computing power that utilized a minimum of 24 A100 units to perform simulations on fluid dynamics and structural mechanics. These methods allow institutions to use high-end power without needing to physically own the hardware.

These records suggest a broader market where computing services act as a gateway for restricted technology. By renting space from domestic providers, research institutions can bypass the need for direct imports. This creates a layer of separation that makes it harder for regulators to track the ultimate end-user of the hardware.

A pattern of procurement

The investigation reviewed 26 different procurement and corporate records from this year to identify a wider trend. Several other institutions with close ties to China's defense industry were found purchasing restricted Nvidia chips. For instance, Shenzhen University bought three H100 processors in September, while Wuhan University of Technology acquired a server containing an H100 chip in June.

Beihang University, another member of the 'Seven Sons of National Defense,' purchased four A800 processors in April. This group of universities is known for its close relationship with China's aerospace and defense industries. The records also showed government procurement contracts involving H800 and A800 computing services for the University of Chinese Academy of Sciences.

While the records provided limited information on the exact origin of the processors, they confirm a consistent pattern of acquisition. Some transactions were even listed at specific rates, such as 10 yuan per chip-hour for H800 access. This systematic procurement suggests that restricted hardware is becoming a standard component of Chinese academic research.

Supply chain loopholes

The findings build on previous reports documenting restricted Nvidia chips embedded in servers manufactured by companies like Dell, Gigabyte, and Super Micro Computer. These servers are often acquired by Chinese universities through domestic resellers who may not be subject to the same scrutiny as direct exporters. This creates a complex supply chain where the hardware is rebranded or bundled with other components.

There is also evidence of a shadow market for prohibited processors that move through international stocks and various third-party nations. These routes allow the technology to move across borders while avoiding the primary export controls intended to block it. The lack of precise data on these supply routes makes it difficult for the US to shut down the flow of hardware.

This situation has intensified calls in Congress for more rigorous oversight and chip-tracking requirements. Lawmakers from both parties have expressed concern that access to these processors could accelerate the development of AI for weapons design. They are currently seeking more transparency to ensure that blacklisted entities cannot remotely access approved chips.

The lobbying controversy

The ongoing access to these chips comes amid increased scrutiny of Nvidia and its chief executive, Jensen Huang. Huang has lobbied for looser US restrictions, arguing that continued sales would preserve American technological leadership. He suggests that keeping Chinese customers reliant on US technology is a strategic advantage for the company.

However, this stance has met significant pushback from US lawmakers who warn about the risks of military modernization. Three Republican congressional committee chairmen recently requested a briefing with Huang regarding his reported contacts with an alleged Chinese intelligence operative. This controversy has further fueled the debate over whether the current export policies are sufficient to protect national security.

The tension between corporate interests and national security remains a central theme in the debate over AI hardware. While Nvidia seeks to maintain its market position, the government is struggling to close the loopholes used by Chinese universities. The next phase of policy may involve stricter requirements for tracking chips once they leave the factory floor.