At least seven Chinese universities with documented links to China’s armed forces are actively seeking access to Nvidia’s H200 graphics processing units through leasing arrangements, Bloomberg reported, citing findings that expose a significant vulnerability in the United States’ semiconductor export control architecture. The institutions identified include Beihang University and Northwestern Polytechnical University, both of which appear on the US Commerce Department’s Entity List, meaning American companies are legally prohibited from supplying them with controlled technology without a specific license.
The report lands at one of the most sensitive moments in the ongoing US-China technology rivalry, raising uncomfortable questions about whether leasing structures, where a third party technically owns the hardware, are being used to circumvent the intent, if not always the letter, of Washington’s export restrictions.
A Chip That Far Exceeds What China Is Permitted to Buy
To understand why this matters, it is necessary to appreciate the gap between the H200 and what China’s technology sector has been permitted to access. Bloomberg notes that the H200 carries more than six times the processing power of the H20, the downgraded chip that Nvidia developed specifically to comply with US export rules targeting the Chinese market. The H20 itself was the subject of further US government scrutiny earlier this year, with the Biden-era restrictions and subsequent Trump administration reviews leaving Chinese buyers in prolonged uncertainty about which chips they could legally purchase.
The H200, by contrast, sits firmly at the frontier of AI computing infrastructure. It is the chip powering the most advanced large language model training runs and scientific research workloads globally. For military-linked research institutions, access to that level of compute would represent a qualitative leap in capability, not simply a faster version of what they already have, but entry into an entirely different tier of AI development.
This context is critical. The ongoing effort to smuggle advanced AI chips into China, documented in a $2.67 billion federal crackdown, demonstrates that demand for restricted chips is intense and that bad actors are already probing every available channel. Leasing arrangements represent a more legally ambiguous pathway than outright smuggling, making them harder to prosecute and easier to obscure within complex corporate structures.
The Entity List and Its Limits
Beihang University, formally known as Beijing University of Aeronautics and Astronautics, and Northwestern Polytechnical University, based in Xi’an, are among China’s most prominent defense-affiliated research institutions. Both have been on the Commerce Department’s Entity List for years, a designation that requires US exporters to obtain a license, which is almost never granted for advanced semiconductors — before shipping controlled items to them.
The problem, as the Bloomberg report highlights, is that export controls are primarily designed to regulate sales transactions. When a leasing company, potentially domiciled outside the United States, using chips that may have been legally purchased before restrictions tightened, provides computing access to a blacklisted entity, the legal chain becomes murkier. It is this ambiguity that appears to be driving the strategy documented in the report.
US officials and trade compliance experts have long warned that access to cloud computing and hardware leasing represent the next frontier of enforcement challenges. When physical chips cross a border, customs agencies can intercept them. When compute is delivered as a service, the transaction may be invisible to traditional border controls, residing instead in contract law, financial flows, and corporate registration records spread across multiple jurisdictions.
A Broader Pattern of Access-Seeking
The Bloomberg findings fit within a broader pattern as Chinese institutions and companies develop increasingly sophisticated strategies to maintain access to frontier computing power despite tightening restrictions. Anthropic’s widely discussed internal analysis warned explicitly that chip controls remain one of the few levers capable of slowing China’s AI development trajectory, a conclusion that the H200 leasing story seems to validate rather than challenge.
The Commerce Department’s Bureau of Industry and Security has previously penalized US firms for illegal chip exports to Entity List companies, but enforcement actions have struggled to keep pace with the creativity of evasion strategies. The leasing model, if it becomes widespread, could prove significantly harder to police than direct sales.
Implications for US Export Control Policy
The revelations arrive at a moment of genuine policy flux in Washington. The Trump administration has signaled a willingness to revisit some of the Biden-era chip restriction framework, including the AI Diffusion Rule, which imposed tiered restrictions on chip exports across a wide range of countries. At the same time, national security hawks within the administration have pushed for stricter enforcement rather than any relaxation.
The H200 leasing story is likely to energize those calling for tighter controls, potentially accelerating moves to close loopholes around leasing and cloud access that current regulations do not fully address. It may also intensify pressure on allied governments and jurisdictions, including those in Southeast Asia and the Middle East, where leasing intermediaries could plausibly be based, to align more closely with US enforcement priorities.
For Nvidia, the situation creates familiar reputational and regulatory risk. The company has consistently maintained that it complies with all applicable export laws, and there is no suggestion in the Bloomberg report that Nvidia itself has done anything improper. But the association between its flagship chips and Chinese military-linked institutions keeps the company entangled in geopolitical debates it would almost certainly prefer to avoid, particularly as it navigates the delicate balance of maintaining access to China’s massive AI market while staying on the right side of US regulators.
The Stakes of the Compute Race
Ultimately, the H200 leasing story is a reminder that the US-China AI competition is not confined to model benchmarks and research publications. It is also being fought through supply chains, legal structures, and the grinding bureaucratic machinery of export enforcement. AI safety discussions between Washington and Beijing have opened narrow diplomatic channels at the summit level, but the underlying competition for compute supremacy continues unabated beneath the surface.
For the seven universities identified in the Bloomberg report, access to H200-class hardware would meaningfully accelerate research programs that US policymakers have specifically sought to constrain. Whether existing legal frameworks are adequate to prevent that outcome is now a question that lawmakers, regulators, and Nvidia itself will be forced to answer publicly, and soon.