Reuters Finds $3.2B in China AI Projects Tied to Nvidia Chips

More than $3.2 billion worth of Chinese computing projects and equipment agreements have been designed around Nvidia processors subject to US export restrictions, according to a new Reuters review. The finding does not mean $3.2 billion of banned chips have been delivered. It shows how much planned AI infrastructure may be exposed to changing supply and enforcement rules.

Reuters examined more than 150 procurement notices, investment plans, construction contracts and corporate disclosures. Together, they describe over 23 billion yuan in projects involving Nvidia hardware, from regional computing centres to facilities backed by local authorities and universities. The total includes buildings, power systems and other infrastructure, as well as agreements at different stages of completion. It is not a measure of Nvidia chip revenue.

Some Chinese operators have planned data centres around H100, H200 or H800 processors. Those names matter because export conditions have changed over time and differ by product and end user. A project designed for one processor may need new hardware, software work or a revised timetable if the expected supply becomes unavailable. Reuters said some facilities are operational or under construction, while others remain plans. At one site in Gansu province, it was unclear whether the thousands of expected chips had arrived.

The review therefore establishes demand and dependence more clearly than it establishes deliveries. It does not show how restricted chips reached every operator, nor does it prove that each project breached US law. Treating plans, signed contracts and running installations as a single completed stock of processors would overstate the evidence.

Nvidia has said export controls effectively prevent it from competing in China’s data-centre computing market. Reuters notes that an IDC estimate still gave the company 34% of Chinese AI accelerator shipments in the first half of 2026. Those statements are not necessarily contradictory. Nvidia’s current access to a particular high-end market is different from shipments across a broader accelerator category, including products sold under different rules.

Washington has progressively tightened controls on advanced AI processors since 2022. The rules aim to limit access to high-end computing capability, but customers can plan projects through intermediaries, older inventory or hardware located abroad. TechBooky recently reported that some Chinese firms were seeking Nvidia-powered capacity outside China amid concern about further US restrictions on remote access. That is a separate route to computing power and should not be confused with the installations in Reuters’ latest review.

The latest findings arrive amid heightened scrutiny of Nvidia following Jensen Huang’s resignation as chief executive. Reuters said the Chinese projects it reviewed have not been linked to Megaspeed or to allegations surrounding Huang. Nvidia’s board has opened an investigation, but there has been no public finding that Huang violated export-control laws or that company secrets were compromised. The procurement documents stand on their own as evidence of a wider policy problem.

For Chinese developers, the immediate risk is that a data centre planned around scarce imported chips can be delayed or forced to switch suppliers. Domestic alternatives may reduce dependence, but changing hardware also means adapting software and operations. For US regulators, the challenge is to understand who ultimately uses restricted computing power, not simply who appears on a purchase order. Reuters’ review puts a number on the projects at stake, while leaving the most consequential questions about actual deliveries and compliance open.

Source: Reuters’ investigation, published on 13 November 2026.