More than $3 billion in Chinese AI-computing projects have been planned, contracted or built around Nvidia processors covered by U.S. export restrictions, according to a Reuters investigation published on November 13, 2026. The finding puts a number on a persistent tension: Chinese developers continue to design computing centres around American chips even when procuring or operating them may be difficult.

According to Reuters, the reporters examined more than 150 procurement notices, investment plans, construction contracts and corporate disclosures. Together, the identified projects and related equipment agreements carry reported values above 23 billion yuan, about $3.2 billion. Crucially, that is not a figure for actual Nvidia chip sales. It includes planned projects, contracts, operating facilities and expenditure on buildings, electricity and supporting infrastructure.

What is planned, and what is actually installed?

The distinction between a proposal and an installed cluster changes how the numbers should be read. Reuters describes a facility in Gansu that was expected to install thousands of Nvidia processors, while noting that their arrival remained unconfirmed. A procurement plan does not prove delivery, installation or a lawful export licence.

The potential cost of that uncertainty is tangible. If a site has been designed for a particular accelerator, switching to other hardware can require changes in software support, server design, power distribution and deployment schedules. Such redesigns are plausible operational risks, not confirmed outcomes for each project in the investigation.

Export controls are more complicated than a single ban

Washington began tightening restrictions on advanced AI accelerators in 2022. But U.S. policy is not identical for every Nvidia product and customer. In a January 13, 2026 statement, the U.S. Bureau of Industry and Security said certain H200 export applications would be reviewed case by case, subject to security and compliance conditions. That process is not blanket authorization to supply all Chinese buyers.

Reuters' findings do not establish the origin of every processor in operation, whether any particular shipment breached export laws or whether the full pipeline will be delivered. Procurement through domestic intermediaries and access to rented computing can further complicate tracing. Evidence that a facility specifies a chip is different from evidence of a lawful export license, arrival, installation and use.

Why the pipeline matters beyond China

For global AI infrastructure suppliers, the report suggests that demand can remain high in a market where direct sales are constrained. For Chinese cloud operators, the challenge is determining which hardware they can actually obtain and support over the life of a facility. For policymakers, a list of domestic chip sales alone may miss computing capacity available through intermediaries or service contracts.

The immediate next questions are measurable: which proposed centres receive their processors, which become operational, and which switch to domestically available accelerators. Until that evidence exists, the $3.2 billion should be treated as a mixed portfolio of plans and facilities—not as verified Nvidia revenue or an inventory of successfully delivered restricted chips.

Sources: Reuters investigation, November 13, 2026; BIS semiconductor licensing policy, January 13, 2026.