Chinese university acquires banned Nvidia chips via domestic suppliers
Procurement records show a military-linked Chinese university bypassed U.S. export controls by renting high-end AI hardware through local intermediaries.
The short version
- Northwestern Polytechnical University spent 874,700 yuan on two Nvidia H200 chips through a domestic supplier.
- The university is one of the 'Seven Sons of National Defence' and is on the U.S. trade blacklist.
- Records show the school rented servers with A100 and A800 chips, both of which are currently banned for export to China.
- Other institutions including Shenzhen University and Beihang University also appear to have made similar purchases.
Why it matters
This highlights significant gaps in U.S. export controls, as domestic suppliers can act as intermediaries to move restricted technology. It shows that even institutions on the 'Entity List' are finding ways to access the hardware needed for advanced AI training.
Bypassing the Entity List
Procurement records from Northwestern Polytechnical University revealed that its School of Management spent 874,700 yuan on two Nvidia H200 chips in October. While the U.S. government has placed this specific university on its trade blacklist, the school managed to acquire the hardware through a domestic company called Shaanxi Bosuan Network Technology Co Ltd.
The university is a member of the 'Seven Sons of National Defence,' a group of institutions with close ties to the Chinese military. Because of these connections, the U.S. Department of Commerce has restricted its access to high-end technology since 2001. However, the recent acquisition suggests that domestic suppliers are successfully navigating these trade barriers.
The U.S. government did begin allowing some H200 chip exports to China in January of this year. Crucially, however, those permissions did not apply to institutions on the Entity List. The university's ability to secure these chips indicates that the blacklist is not preventing the flow of hardware to sensitive research centers.
Renting restricted capacity
Beyond direct purchases, the university has also utilized a rental model to gain access to high-performance computing. The School of Cybersecurity signed two contracts in January for server access, with each server housing eight H200 chips. These resources were specifically used for reinforcement learning training and fine-tuning AI models.
The university also rented access to servers equipped with Nvidia A100 and A800 chips. Both of these specific models are currently under a total ban for export to any entity within China. By renting this capacity, the university can perform complex simulations without having to own the physical hardware directly.
One research institute at the university spent 375,000 yuan in April for computing services related to fluid dynamics and structural simulations. These specific services required at least 24 A100 chips. The university signed these contracts with several small, little-known Chinese technology suppliers.
The domestic supplier network
The procurement records do not provide clear information on the original source of the processors or the physical location of the rented servers. This lack of transparency allows the university to utilize advanced hardware while maintaining a degree of separation from the primary manufacturer. It suggests a sophisticated supply chain where domestic firms act as the final link.
Reuters identified several other Chinese universities and research institutions that have made similar purchases through domestic distributors. These include Shenzhen University, Beihang University, and Wuhan University of Technology. Many of these institutions also share the same close ties to the Chinese military as Northwestern Polytechnical University.
Reports indicate that these restricted chips are often embedded in servers from major brands like Dell, Gigabyte, and Super Micro Computer. There is also evidence of an 'underground market' in China where restricted chips are sourced from overseas inventories and moved into the country via intermediaries in third countries.
Pressure on Nvidia and U.S. policy
The ability of Chinese entities to acquire these chips is putting pressure on Nvidia's leadership in Washington. CEO Jensen Huang has faced scrutiny over his lobbying efforts to ease restrictions on AI chip sales. This comes at a time of increased oversight regarding the company's interactions with Chinese interests.
The findings highlight a persistent gap in Washington's efforts to restrict China's access to advanced technology. While the U.S. aims to slow China's AI development, the use of domestic suppliers and rental services provides a workaround. These methods allow Chinese researchers to continue training large-scale models despite official trade barriers.
The situation suggests that as long as domestic distributors can source and host these chips, the effectiveness of the Entity List may be diminished. Policymakers may need to look beyond the primary manufacturer to address the secondary markets and rental services that facilitate these transactions.
What to watch
- Whether the U.S. government will target domestic Chinese distributors to close the hardware loophole.
- If more 'Seven Sons of National Defence' universities are found to be using similar rental contracts.
Try it
Search for the 'Entity List' on the U.S. Department of Commerce website to see which other institutions are currently restricted from receiving high-end technology.
Sources and links
- Reporting联合早报: the article we sourced
Liked it? Say why
Add a line for other readers.
No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.
Listen to the AI Wire headlines
Listen to the AI WireListen to our daily AI shows every morning: on your commute, in the shower, wherever you are
Listen to the daily showsFirst reported by 联合早报 · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.
