Blacklisted Chinese University Bought Nvidia H200 Chips Through Domestic Supplier, Records Show

Two chips and five rental contracts reveal access to restricted computing

Northwestern Polytechnical University, a Chinese institution under US export restrictions since 2001, bought two Nvidia H200 artificial intelligence processors for 874,700 yuan, about US$123,000, according to a procurement record posted on its website on October 21, 2026. The purchase came after Washington loosened some restrictions on H200 sales to China while keeping separate limits on blacklisted institutions.

The university's School of Management purchased the processors from Shaanxi Bosuan Network Technology, a domestic supplier, rather than directly from Nvidia. The university, based in Xi'an, has close ties to China's People's Liberation Army and conducts research connected to military aircraft, missiles and satellites.

Procurement records also show five contracts during 2026 giving the university access to powerful Nvidia processors through rented computing services. Together, the purchases and rentals expose a central enforcement problem: an institution can obtain computing power through a supplier inside China, even when US rules restrict exports to that institution.

The stakes extend beyond the ownership of two processors. Washington's revised licensing policy includes safeguards intended to keep blacklisted institutions and military users from accessing approved exports remotely. The university's contracts put both physical sales and access to computing services at the centre of the debate.

What the October purchase records show

The October 21 notice identifies two H200 processors, the total price and the supplier. Northwestern Polytechnical University belongs to the group of Chinese institutions commonly called the Seven Sons of National Defense, which have strong connections to defence research and industry.

The purchasing department was the School of Management. Other records identify computing contracts awarded by the university's cybersecurity school, showing that access involved more than one academic department.

The route by which the purchased processors reached the domestic supplier, and any licensing history associated with them, remain unknown. Nvidia, the university and the US Commerce Department's Bureau of Industry and Security did not respond to requests for comment, and contact information for Shaanxi Bosuan Network Technology could not be found.

Rental contracts opened another route to H200 computing

The university's cybersecurity school awarded two contracts in January 2026 for seven months of access to servers containing H200 processors. Each contract specified access to eight chips. One cost 441,700 yuan and the other 441,000 yuan, bringing their combined value to 882,700 yuan.

Those contracts covered fine tuning AI models and training through reinforcement learning. Fine tuning means adapting an existing model to a particular task or set of information. Reinforcement learning trains a system using feedback or rewards, helping it favour responses or actions that meet a chosen goal.

Renting access allows a customer to submit computing tasks to servers operated by another company. The customer can use the processors without buying them or installing them on its own premises. That makes the identity of the person or organisation using a server important alongside the identity of its owner.

The two detailed H200 rental contracts together cost 8,000 yuan more than the university's purchase of two H200 processors. The comparison shows how much the university committed to temporary computing access, although the contracts bought different services from a hardware purchase.

Other rental records covered servers containing eight Nvidia A100 chips and eight A800 chips. Those models are also subject to US restrictions barring their export to Chinese entities. Across the five computing contracts, the university sought access to several generations of Nvidia hardware, with H200 the most advanced model identified in the procurement records.

Why the H200 matters for AI development

Nvidia's H200 is designed for the large numbers of simultaneous calculations needed to train and run advanced AI systems. It belongs to a category called graphics processing units, or GPUs. These processors can divide demanding work into many smaller calculations and perform them at the same time.

Training is the process in which an AI model learns patterns from data. Running a trained model, often called inference, is what happens when the system answers a question, analyses an image or produces another result. Both can require substantial computing capacity, especially when models are large or many people use them at once.

The H200 has substantial memory capacity and bandwidth, allowing it to hold and move the information needed for demanding workloads. It is described as Nvidia's second most advanced AI processor after Blackwell and can offer up to six times the computing capacity of the H20, a less powerful chip designed for the Chinese market. Actual gains depend on the task and the surrounding computer system.

Chinese companies including Alibaba Group and ByteDance have shown interest in H200 processors because greater computing capacity can help them develop and deploy advanced models. Nvidia's appeal also rests on its software ecosystem, which developers use to build and operate AI applications.

Chinese manufacturers, including Huawei, are working to compete with that combination of hardware, production scale and software. For customers choosing equipment now, access to Nvidia processors can affect which models they can run and how quickly they can complete demanding workloads.

China sales became easier, but the university stayed restricted

The US Commerce Department placed Northwestern Polytechnical University on its Entity List in 2001 over alleged involvement in missile proliferation and military research. The list imposes licensing requirements on transactions involving specified organisations. In 2019, Washington tightened the university's restrictions by adopting a presumption of denial for most applications to export controlled US technology to it.

A presumption of denial means officials generally begin from the position that an application should be rejected. That restriction remained in place when President Donald Trump's administration changed its broader approach to H200 sales to China.

On December 8, 2025, Trump announced that approved Chinese customers could buy H200 processors under conditions. In January 2026, the Commerce Department changed its licensing policy so applications involving H200 and comparable AMD processors could be considered individually.

The revised policy included independent testing in the United States, safeguards against unauthorised use and assurances that exports would not reduce supplies available to American customers. It also required protections against remote access by blacklisted entities and military users. Blackwell processors did not receive the same approval.

The sequence shows how the university's purchases fit into a changing policy:

• 2001: The Commerce Department added Northwestern Polytechnical University to the Entity List.
• 2019: Washington adopted a presumption of denial for most controlled technology exports to the university.
• December 8, 2025: Trump announced conditional H200 sales to approved Chinese customers.
• January 2026: Licensing rules changed, and the university awarded two contracts for access to H200 servers.
• October 21, 2026: A university procurement notice recorded the purchase of two H200 processors.

Permission to consider H200 exports to some Chinese customers left the university's individual restrictions intact. A licensing review therefore involves both the processor being sold and the organisation that will ultimately use it.

Military concerns collide with Nvidia's China strategy

US lawmakers concerned about advanced chip sales argue that greater computing power could accelerate Chinese capabilities in weapons design, intelligence analysis and military modernisation. AI can process large volumes of information and assist research involving complex calculations, giving the same hardware uses in both civilian and military work.

Northwestern Polytechnical University's research connections make its access particularly sensitive. Its work on aircraft, missiles and satellites is central to the reasons Washington has restricted technology exports to the institution for more than two decades. The procurement records identify academic departments and AI training tasks, while the export rules apply to the university as a restricted organisation.

Nvidia chief executive Jensen Huang has argued that continued sales to China would help preserve American technological leadership. His position is that Chinese customers using US chips and software remain connected to an American technology ecosystem, whereas tighter exclusion could push them towards domestic competitors.

Beijing has its own concerns about dependence. Chinese authorities have encouraged companies to use domestic alternatives, and officials reportedly told companies in January to purchase H200 processors only when necessary. Changes in US export policy create a practical risk for Chinese buyers: equipment available under one policy may become harder to obtain or expand under another.

The university's spending illustrates the continuing demand beneath those competing policies. It committed money both to ownership of H200 hardware and to temporary access to servers operated by Chinese companies.

A separate inquiry adds scrutiny around Nvidia's leadership

Nvidia is also facing scrutiny over allegations that Jensen Huang was targeted in a Chinese espionage operation involving Alice Huang, a former executive at Singapore based data centre company Megaspeed. Jensen Huang has acknowledged a personal relationship with her, and Nvidia's board has reportedly begun an independent investigation into whether company secrets or sensitive technology were exposed.

Nvidia has said it has no reason to believe that either was compromised. Alice Huang has pledged to cooperate with the inquiry, while China has dismissed the espionage allegations as baseless.

Further allegations concern Alice Huang and associates helping arrange at least US$7.5 billion in Nvidia chip shipments to Chinese companies through a network of businesses. Those claims remain under investigation. The university procurement records identify Shaanxi Bosuan Network Technology as the seller in the October purchase.

Enforcement reaches beyond the shipment itself

The university's procurement activity shows why export controls must account for what happens after processors enter a domestic market. A chip may pass through a supplier before reaching a purchaser, while a server operator can provide computing access to customers who never take possession of its equipment.

The January licensing conditions address that second route by requiring safeguards against remote use by restricted organisations. For companies handling approved processors, checking the original buyer is only one part of the task. The rules also require attention to who can use the machines and for what purpose.

Northwestern Polytechnical University's records bring those two routes together: a purchase through a Chinese supplier and multiple contracts for computing services. Washington's decision to permit some H200 exports created a conditional route for approved customers, while the university's longstanding restrictions continued to apply.

Key Points

• An October 21, 2026 procurement notice recorded Northwestern Polytechnical University's purchase of two Nvidia H200 processors for 874,700 yuan.
• The seller was Shaanxi Bosuan Network Technology, rather than Nvidia directly.
• Five computing rental contracts also provided access to powerful Nvidia hardware during 2026.
• Two January contracts each specified access to eight H200 chips for seven months, at a combined cost of 882,700 yuan.
• The university has been on the US Entity List since 2001, with tighter licensing restrictions introduced in 2019.
• January 2026 rules allowing individual review of some H200 exports retained restrictions on blacklisted institutions and military users, including remote access safeguards.