Chinese AI Projects Worth Over $3 Billion Are Counting on Nvidia Chips

Chinese developers are planning and building AI data centers around Nvidia chips subject to U.S. export restrictions. Reuters found projects and equipment contracts worth more than 23 billion yuan, or about $3.2 billion, combined. It remains unclear whether the chips have been delivered for some of the projects.

For its document review, Reuters examined more than 150 Chinese tenders, investment plans, construction contracts and company records. The news agency found plans involving regional governments, universities and businesses, among others. The projects span the country, from the coastal province of Fujian to Gansu in the northwest.

The documents describe projects at different stages of development. One proposal specifies the type of chip sought but still needs partners. Another has already awarded a construction contract. Reuters also found examples of data centers that, according to local governments or company statements, are already running on Nvidia processors.

The total of more than $3 billion combines these different categories. It also includes buildings, power supplies and other infrastructure. The figure offers a sense of the scale of the projects—and of Chinese demand for Nvidia chips—but cannot be used to estimate Nvidia’s chip revenue. The documents do not provide the necessary breakdown.

A plan for a computing center in Xianyou County, Fujian, puts the project’s cost at 6.2 billion yuan, or about $873 million. The project is seeking commercial partners and lists Nvidia’s H100 chips as part of its planned setup. Elsewhere in the province, listed property developer Damingcheng said a company had already built a cluster using Nvidia H800 chips. The contract for computing services at the cluster was terminated early.

Gansu shows what a construction contract can—and cannot—confirm. In July, a consortium was awarded a 1.16 billion yuan contract for a computing center designed around about 3,072 Nvidia processors, according to Reuters’ project overview. The documents name the H100 and H200. Reuters could not determine from the records whether the processors had arrived.

The status is clearer in the eastern province of Jiangsu. In March 2025, the city of Wuxi announced that a cluster using H800 processors had gone into operation. But the announcement does not provide a full account of how or when the supplier delivered the chips.

Washington has restricted exports of certain advanced AI processors to China since 2022. In January, the U.S. government shifted to a case-by-case review of H200 chip exports to eligible Chinese customers. The chip model and delivery conditions therefore matter when interpreting a tender.

AI Wereld previously reported that Nvidia excluded Chinese data-center computing revenue from its outlook. That reflects the revenue the company expects to book. The investigation into Chinese projects shows the hardware developers are planning around.

The investigation appeared a day after Jensen Huang announced his departure from Nvidia. Reuters drew no connection between the Chinese projects it examined and the case involving Huang. Nvidia and China’s Ministry of Commerce did not immediately respond to questions about the findings, the news agency reported.