Nvidia has told cloud-computing providers that use its chips to sell at least 51 per cent of their services to companies outside China, according to the Financial Times, in an effort to close a route by which Chinese firms gain access to advanced American AI hardware.
The requirement, conveyed verbally through Nvidia's sales teams rather than in writing, applies only to new contracts, according to five people familiar with the matter cited by the newspaper. Nvidia had not previously imposed such a limit.
Nvidia neither confirmed nor denied the report. In a statement it said: "America cannot win the AI race unless America's AI infrastructure wins worldwide." Mao Ning, a spokeswoman for China's foreign ministry, said she was not aware of the report and that Beijing hoped for a "fair, transparent and non-discriminatory business environment".
The overseas loophole
America has restricted exports of Nvidia's most advanced chips to China since 2022. But many Chinese AI developers rent computing power from data centres in other countries, particularly in South-East Asia and the Middle East, to train their models. Washington has tried for years to close that route, and smuggling persists: earlier this year several senior employees of Super Micro, a server maker, were charged with smuggling about $2.5bn of Nvidia AI servers into China.
Beijing, for its part, is pushing in the opposite direction. A year ago it banned foreign AI chips from state-funded data centres to support domestic suppliers such as Huawei.
The cap could weigh on cloud providers that have built businesses serving Chinese clients, and on demand for Nvidia's own chips in those markets. It also shows how far the world's most valuable chipmaker is now acting as an instrument of American industrial policy, policing its customers' customers on Washington's behalf.
