Key points
- Reuters tallies Chinese AI projects calling for Nvidia chips
- Nvidia's filings say it's shut out of China's data centers
- What Nvidia's forecast assumes before its Nov. 17 results
Chinese buyers are still planning and building AI projects around restricted Nvidia chips, even though the company says US export controls have effectively shut it out of China's data center market.
A Reuters review found Chinese AI projects and equipment agreements worth more than 23 billion yuan ($3.2 billion) that call for or use Nvidia (NVDA) processors subject to US restrictions.
Reuters examined more than 150 procurement notices, investment plans, construction contracts, and corporate disclosures. The total includes planned investments, signed contracts, and completed installations, as well as spending on buildings, power systems, and other infrastructure. It doesn't measure spending on Nvidia chips alone.
The review comes after Nvidia announced Thursday that Jensen Huang had resigned as CEO, amid a scandal over his relationship with Alice Huang, whom US authorities suspect of intelligence ties to Beijing. The projects haven't been linked to Megaspeed, the Singapore-based Nvidia cloud partner she was associated with, or to the allegations involving her. There has been no public finding that Jensen Huang violated export-control laws.
Nvidia and China's commerce ministry didn't immediately respond to requests for comment. Nvidia shares closed at $166.19 on Thursday, down 4.3%, and traded at $169.75 at 9:13 a.m. ET Friday, before the market opened, up 2.1%. We're tracking key developments in our Jensen Huang scandal timeline.
The projects run from Fujian to Inner Mongolia
The largest project in a Reuters factbox is a proposed 6.2 billion yuan ($873 million) AI computing center in Xianyou county, Fujian province. Its plans specify Nvidia H100 chips, and the project is seeking commercial partners.
Other plans specify H100 or H200 processors:
- In Shaanxi, state-owned engineering group Sinomach said a subsidiary won a contract as part of a 3 billion yuan ($422 million) computing center that specifies H100 and H200 processors.
- In Jilin, a 2.26 billion yuan ($318 million) project would combine H100 chips with Chinese-developed chips, according to an April government prospectus. Its developer expects bank loans to cover about 80% of the investment.
- In Inner Mongolia, a 1.5 billion yuan ($211 million) development proposes installing 625 Nvidia H200 servers.
Some projects are further along. In July, a consortium led by a subsidiary of state-owned Shaanxi Construction Engineering won a 1.16 billion yuan ($163 million) construction contract for a computing center in Gansu province. Tender documents specify H100 and H200 hardware and say the center plans to install about 3,072 Nvidia processors. They don't establish whether the processors have been delivered.
Other facilities already use restricted Nvidia hardware. The Wuxi municipal government said in March 2025 that a computing cluster using H800 processors had entered operation. Damingcheng, a Shanghai-listed company, disclosed in June that a computing venture had built an H800 cluster. The venture signed a computing-services agreement originally worth 759 million yuan ($107 million), but the contract ended early, according to the factbox.
Smaller buyers show up too. Shenzhen University awarded a contract for three H100 processors in September, and Northwestern Polytechnical University, which is on a US blacklist, bought two H200 chips in October, according to a separate Reuters report.
Nvidia's own filings describe a closed market
As of the end of its second fiscal quarter, Nvidia was "effectively foreclosed from competing in China's data center computing/compute market," the company said in its Aug. 26 quarterly report. Shipments of Data Center Hopper products to China were less than 1% of Data Center revenue in that quarter, according to the CFO commentary released the same day. Data Center revenue was $89.0 billion, so those shipments came to less than $890 million, by our calculations.
The US government granted licenses allowing Nvidia to ship small amounts of H200 products to specific China-based customers, the quarterly report says. The Chinese government restricted those sales. Nvidia took a $0.4 billion charge in the first half of its fiscal year for excess H200 inventory and purchase obligations. Separately, about 600,000 H200 chips are stored in Hong Kong because Beijing hasn't approved large-scale imports, Reuters reported on Nov. 5.
Huang said last year that Nvidia's market share in China had fallen from 95% to zero.
Separately, an IDC report cited by Reuters in October estimated that Nvidia accounted for 34% of Chinese AI accelerator shipments in the first half of 2026. That estimate covers a later period and measures shipments. Neither figure establishes how the restricted hardware in the review reached Chinese users, the report said.
Nvidia's forecast assumes no China compute revenue
Nvidia's outlook for its third fiscal quarter calls for revenue of $108.0 billion, plus or minus 2%. The company isn't assuming any Data Center compute revenue from China in that outlook. Nvidia reports those results on Nov. 17.
The Chinese developers face their own risk. Tighter US curbs could bring construction delays, higher costs, or forced redesigns around domestic chips, according to the review. Huang's departure could deepen those risks by weakening Nvidia's lobbying leverage as US lawmakers press for tighter controls, the report said.
Sunny Cheung, an analyst at the Washington-based Jamestown Foundation, said the findings support his view that Washington should look past the buyers named on paper and "start examining who ultimately uses the computing power."
Dollar figures for the Shaanxi, Jilin, Inner Mongolia, Gansu, and Damingcheng amounts are our conversions at the rate Reuters used for the Fujian project, about 7.1 yuan to the dollar. Share prices are from Yahoo Finance.
