DGCity
— A Window to China

Chinese AI Centers Still Rise on Restricted Nvidia Chips

Chinese developers are still breaking ground on artificial-intelligence computing centers built around Nvidia processors that Washington bars from being sold to China, according to a review of more than 150 procurement notices, investment plans, construction contracts and corporate disclosures. The documents identify projects and equipment agreements with a combined reported value of more than 23 billion yuan, roughly $3.2 billion, stretching from the wealthy coastal technology hubs to inland and western provinces.

The filings show how deeply American hardware remains woven into China’s AI buildout. A proposed 6.2 billion yuan computing center in Fujian province openly seeks partners and specifies Nvidia H100 chips for large-scale model training. In Shaanxi, state-owned engineering group Sinomach said a subsidiary won a 3 billion yuan contract for a center calling for H100 and H200 processors. A 2.26 billion yuan project in Jilin pairs H100 chips with domestic processors, while an Inner Mongolia development worth 1.5 billion yuan proposes 625 H200 servers, with first-phase construction reportedly under way.

The scale sits awkwardly with Nvidia’s own statements. The company has said U.S. export controls effectively shut it out of China’s data-center market, and its chief executive last year put its Chinese market share at effectively zero, down from 95 percent. Yet an industry report estimated Nvidia still accounted for about 34 percent of AI accelerator shipments into China in the first half of 2026. The $3.2 billion total covers buildings, power systems and infrastructure as well as hardware, not chips alone, and includes signed contracts and planned investments alongside completed installations.

Restricted H100, H200 and earlier H800 processors have continued to reach Chinese users through stockpiles, third-country hubs and offshore data centers, where Chinese firms rent computing power without taking delivery of the chips. Washington began restricting advanced AI chip sales to China in 2022 and later closed loopholes for slower China-specific models; this year it moved to review H200 export licenses case by case.

The projects now face an exposed future. Tighter enforcement could delay construction, disrupt supply or force expensive redesigns around domestic accelerators, analysts caution, while some planned installations may never receive the processors they were designed around. Regional governments and universities that backed the hubs as anchor infrastructure would be left carrying those costs. The findings add urgency to debates over whether controls should target not just who buys chips, but who ultimately uses the computing power.

Like(0)
未经允许不得转载:Dgcity » Chinese AI Centers Still Rise on Restricted Nvidia Chips

Sign In

Forgot Password

Sign Up