AI news
AI model reviews
AI tools
How-to guides
Our staffAboutContact

Reuters finds $3.2 billion of Chinese AI projects built on restricted Nvidia chips

A Reuters review found over 23 billion yuan of Chinese data-centre projects specifying Nvidia H100, H200 and H800 chips, a day after Jensen Huang quit as CEO.

Short answer: a Reuters review has identified more than $3 billion in Chinese computing projects built around Nvidia's restricted processors, even though Nvidia says US export controls have effectively shut it out of China's data-centre market. The findings landed a day after Jensen Huang, a vocal advocate of easing export restrictions, resigned as chief executive over a scandal.

I think the headline number is softer than it looks. Plenty of it is plans and buildings rather than chips.

The projects, in numbers

Reuters looked at more than 150 Chinese procurement notices, investment plans, construction contracts and corporate disclosures, and the combined reported value tops 23 billion yuan ($3.2 billion). Some of those projects are only plans. The estimate includes planned investments and signed contracts as well as completed installations, and it covers buildings and power systems, so it isn't spending on Nvidia chips alone.

LocationReported valueHardware and status
Xianyou county, Fujian6.2 billion yuan ($873 million)Proposed; seeks commercial partners; specifies H100 chips
Shaanxi3 billion yuanSinomach says a subsidiary won a contract; specifies H100 and H200
Jilin2.26 billion yuanH100 mixed with Chinese-developed chips, per an April government prospectus
Inner Mongolia1.5 billion yuanProposes 625 H200 servers; construction reportedly began on two buildings in June
Gansu1.16 billion yuanConstruction contract awarded in July for about 3,072 Nvidia processors

Some sites are further along. The Wuxi municipal government said in March 2025 that an H800 computing cluster had entered operation, and Shanghai-listed Damingcheng disclosed an H800 cluster in June. Smaller buyers show up too: Shenzhen University bought three H100s, and US-blacklisted Northwestern Polytechnical University bought two H200s.

Who pays and who gains

The money is coming from several places. In Jilin, the developer expects bank loans to finance about 80% of the investment. State-owned groups are doing the building: Sinomach in Shaanxi, and a consortium led by a Shaanxi Construction Engineering subsidiary in Gansu. Fujian's proposed centre is still hunting for commercial partners.

Who gains is murkier. Construction firms and local governments clearly do. Nvidia's position is less clear, because the figures don't establish how the restricted hardware reached Chinese users. The Gansu documents, for instance, don't show whether the processors were ever delivered.

That's a real gap in the story. A tender that names H200s isn't the same as H200s in a rack.

The Huang factor

Reuters says Huang's departure could deepen these vulnerabilities by weakening Nvidia's lobbying leverage as US lawmakers press for tighter controls. That's a possible consequence, not a documented one, and I'd treat it that way. Operators already face potential delays, supply disruptions or costly redesigns around domestic chips.

The scandal itself is separate from these projects. The Chinese projects Reuters identified haven't been linked to Megaspeed or the allegations involving Huang, and there's been no public finding that Huang violated export-control laws. Megaspeed, the Singapore-based Nvidia cloud partner, has denied illegally diverting chips to China. Nvidia named CFO Colette Kress interim CEO and Harvey C. Jones as its first-ever board chair.

Nvidia's own numbers don't line up neatly

Two data points pull in different directions. Nvidia's August filing said it was effectively excluded from China's data-centre computing market, yet an IDC report put its share of Chinese AI-accelerator shipments at 34% in the first half of 2026. Reuters points out the two figures measure different things. Nvidia is talking about what it can sell directly; IDC is counting chips that actually shipped.

On the policy side, the Trump administration moved in January to review H200 export licences for eligible Chinese customers individually, subject to conditions. So the rules haven't been a flat ban on everything.

What to watch next

Nvidia's next test comes soon: it reports third-quarter fiscal 2027 results on Tuesday, November 17. Earlier, the company had excluded China data-centre computing revenue from its forecasts, citing regulatory uncertainty, so any change in that wording is worth reading closely. China's commerce ministry and Nvidia hadn't responded to Reuters' requests for comment.

Honestly, I'd read the $3.2 billion as a measure of demand, not of lost Nvidia revenue. A good chunk of it is concrete, power systems and planned buildings, and I'd guess plenty of it ends up on Chinese chips instead. The projects most at risk are probably the early-stage ones with no signed equipment deal, like the Fujian proposal.

If you're tracking the AI chip market, I'd wait for the earnings call and any export-licence news before drawing conclusions about China. The Reuters list shows the appetite is still there. It doesn't show Nvidia is the one getting paid.

Keep reading