Universities, military affiliates and data centers have obtained thousands of restricted processors through grey market channels and rental arrangements
Universities, military affiliates and data centers have obtained thousands of restricted processors through grey market channels and rental arrangements
President Donald Trump’s recent decision to permit exports of Nvidia’s H200 artificial intelligence chip to China promises to legitimize a trade that has flourished in the shadows for months. A comprehensive review of more than 100 procurement documents and academic papers reveals that Chinese buyers have already acquired substantial quantities of the powerful processors through grey market channels, despite existing export restrictions.
The evidence paints a detailed picture of which Chinese organizations will likely pursue legal bulk purchases once Beijing approves sales. The H200 represents Nvidia’s second most powerful AI chip and vastly outperforms any processor the company currently has authorization to sell in China.
Academic institutions lead the charge
China’s premier universities have emerged as major consumers of restricted Nvidia hardware, driven by intense competition for top talent and breakthrough research. The availability of advanced chips directly influences these institutions’ ability to recruit leading researchers and conduct cutting edge artificial intelligence development.
A professor at Beijing Jiaotong University openly advertises that his laboratory possesses eight H200 chips for AI model research. Researchers from the state backed Shanghai Artificial Intelligence Laboratory collaborated with colleagues from Sun Yat Sen, Tsinghua and Shanghai Jiao Tong universities to train an AI model using four Nvidia H200 processors. Their project, detailed in a paper published last month, focused on detecting AI generated images.
In June, a government run AI institute in Hefei issued a procurement request for a server equipped with eight Nvidia H200 chips to power a quantum AI model project. The Reuters analysis identified dozens of universities and research institutes nationwide that have either purchased or attempted to acquire H200 chips through various channels.
Military connections raise national security concerns
China hawks in Washington have strongly opposed Trump’s reversal of previous export controls, warning that the Chinese military would leverage Nvidia’s most advanced AI chips to dramatically enhance its capabilities. The procurement review suggests these concerns have merit, as H200 chips appear to be reaching the People’s Liberation Army and closely affiliated universities.
In August, the PLA Air Force Medical University in Xian issued a tender for eight Nvidia H200 chips intended to train a large language model platform supporting medical AI and biosurveillance research. More recently, the School of Cyberspace Security at Beihang University sought suppliers capable of renting H200 level computing power. Beihang belongs to China’s Seven Sons, a group of universities under U.S. sanctions due to their extensive defense related research activities.
Chinese entities have increasingly adopted rental arrangements as a strategy for accessing restricted hardware without directly importing it. This approach allows organizations to utilize servers fitted with banned Nvidia chips while maintaining plausible deniability about ownership.
Data centers plan massive deployments
Even before Trump’s announcement, Chinese AI infrastructure projects had incorporated large quantities of H200 chips into their development plans. In Jiangsu province, a company owned by the Binhai county government issued a July tender for 48 servers equipped with 384 H200 chips, with deliveries scheduled before year end.
More ambitious projects have materialized in Xinjiang, where Chinese companies and authorities have pursued aggressive AI infrastructure expansion to capitalize on cheap land and electricity. The far western region has witnessed a massive buildout in recent years, attracting enormous data center developments.
A June tender by Urumqi Jiangsuan outlined plans for a 20,000 petaflop computing hub incorporating more than 8,000 H200 processors alongside 12,000 RTX 4090 units and 4,500 servers fitted with Huawei Ascend 910C processors, currently the most powerful domestically produced AI chip available in China.
Another project in Burqin county valued at 1.86 billion yuan detailed a green energy computing center dominated by 1,000 domestic chip servers but supplemented by a smaller cluster of 100 servers using H100 or H200 chips. The H100, the H200’s less powerful predecessor, has been banned from export to China since late 2022.
Nvidia chips power telecommunications expansion
In Hubei province, Xiaogan Yunqi Data Technology submitted an October regulatory filing for a computing project worth 307 million yuan. The initiative plans to deploy 128 H200 servers for telecommunications giant China Unicom by March, demonstrating how restricted chips have penetrated even state owned enterprises in strategic industries.
The widespread presence of H200 chips across Chinese institutions, military affiliates and commercial data centers suggests that export controls have proven largely ineffective at preventing technology transfer. Whether Trump’s policy shift will bring this underground market into the open remains uncertain, as Beijing has not yet confirmed whether it will permit official sales within China.