Reuters finds more than $3 billion in Chinese AI projects built on restricted Nvidia chips

Nvidia has said US export controls have effectively locked it out of China’s AI data-centre market. The paperwork in China tells a more complicated story.

A Reuters review published Nov. 13 identified more than 23 billion yuan, or approximately $3.2 billion, in Chinese AI computing projects and equipment agreements that explicitly reference Nvidia’s restricted H100, H200, and H800 processors. The chips are banned. The demand for them evidently missed the memo.

Where the chips are showing up

The records Reuters examined fall into three buckets: planned investments, contracts that have already been awarded, and computing facilities that are up and running. That spread matters.

Plans can be aspirational. Awarded contracts and active facilities are considerably harder to wave away.

Geographically, the projects span China’s wealthy coastal tech hubs as well as its remote western provinces. Reuters highlighted activity in Fujian, Shaanxi, Inner Mongolia, and Gansu.

The single largest example is a 6.2 billion yuan ($873 million) AI center in Fujian. Its requirements specifically call for H100 processors, the chip that became the workhorse of the global AI buildout.

In Shaanxi, a 3 billion yuan project linked to state-owned Sinomach involves both H100 and H200 chips. A state-owned enterprise appearing in a project built on export-controlled American hardware is the sort of detail that tends to travel quickly around Washington.

Then there is Gansu. A facility there was awarded a 1.16 billion yuan contract in July for 3,072 H100 and H200 processors.

The research does not establish how, or whether, the restricted chips have actually been delivered to each project. What the documents do show is that Chinese buyers are still designing computing infrastructure around hardware they are not supposed to be able to buy.

Three years of controls, and the demand persists

Washington began restricting exports of advanced AI chips to China in 2022. The rules have been tightened several times since, each round aimed at closing gaps that the previous one left open.

Nvidia’s own footprint in the country remains meaningful by some measures. An IDC estimate cited in the research put Nvidia’s share of Chinese AI-accelerator shipments at 34% in the first half of the year.

That figure sits awkwardly next to Nvidia’s public position. The company has cited regulatory uncertainty as the reason it excludes China data-center revenue from its forecasts.

What this means for Nvidia, Washington, and Beijing

For Nvidia, the most immediate issue is compliance exposure rather than lost sales. The company has already taken China out of its data-center guidance, so its forecasts do not depend on these projects.

Congressional calls for stricter controls are already intensifying, according to the research. A $3 billion-plus paper trail is the kind of evidence that tends to turn hearings into legislation.

Reuters flagged mounting US scrutiny as a risk for developments that are still trying to source banned Nvidia hardware.

Projects with chip-specific requirements baked into their plans may face delays, redesigns, or quiet reassessments if supply looks even less reliable.

The broader takeaway from the documents is that export controls can restrict a product without restricting the demand for it. Over 23 billion yuan in projects still built around Nvidia’s restricted chips suggests Chinese buyers are not treating the ban as the final word.