A Reuters has found that over $3.2 billion (more than 23 billion yuan) of datacenter projects in China are tied to Nvidia accelerators that are subject to U.S. export restrictions.
What the investigation examined
The news agency reviewed more than 150 Chinese procurement notices, investment plans and contracts. From those documents it identified deployments based on Nvidia H100, H200 and H800 chips not only at major technology firms but also in municipal projects, universities and regional datacenters.
The total 23 billion yuan figure reported by Reuters covers both planned and completed investments and includes real estate and energy infrastructure costs, so it does not represent direct chip expenditures alone.
Timeline of export controls
The United States began restricting exports of advanced AI chips to China in 2022. Subsequent measures tightened controls on processors modified specifically for the Chinese market, such as the H800. In January of this year the U.S. government decided to adjudicate export licenses for H200 chips on a case-by-case basis for eligible Chinese customers under differing conditions.
Nvidia’s management has said the measures have effectively cut the company out of the Chinese datacenter market; Jensen Huang stated last year that Nvidia’s market share in China fell from about 95 percent to near zero. Still, an October report from IDC indicated that in the first half of 2026 the U.S. vendor still accounted for roughly 34 percent of AI-accelerator shipments to China.
Where the chips are being used — and how they arrived
Reuters’ documents show deployments across China, from coastal technology hubs to remote western provinces, built around the H100, H200 and H800 accelerators. The records do not make clear the supply routes by which hardware subject to restrictions reached Chinese users.
Reactions and analysis
The Reuters report emerged one day after Nvidia co‑founder and CEO Jensen Huang resigned amid a China‑linked controversy. Sunny Cheung of the Jamestown Foundation, cited by Reuters, argued that U.S. authorities should focus less on tracking the immediate buyers of chips and more on monitoring who ultimately uses the compute capacity and for what purposes.
Why this matters
The findings suggest that, despite export controls, advanced AI accelerators continue to appear in Chinese datacenter projects through channels that are not fully transparent. That has implications for international export policy and for understanding how global AI infrastructure and capabilities are being deployed across different sectors in China.
This article is based on Reuters’ investigation. It does not constitute investment advice.
