Nvidia tells cloud partners to sell most of their capacity outside China, FT reports

Nvidia has told the data-centre and cloud companies that buy its AI chips to sell at least 51% of their computing services to customers that are not Chinese, the Financial Times reported on Tuesday, citing five people with knowledge of the matter. The rule covers new contracts only, not existing ones, according to the report.

Nvidia, the world's most valuable company, neither confirmed nor denied the report. In a statement quoted by AFP, it said that "datacentres and clouds built on American chips are the foundation that American model and application makers need to compete worldwide", and that "America cannot win the AI race unless America's AI infrastructure wins worldwide."

What the report says

According to the FT's account, as summarised by AFP and The Next Web, the instructions were passed on by word of mouth through Nvidia's sales force and regional teams rather than in writing. Nvidia had no such limit before. In practice, the rule caps the share of a partner's business that can go to Chinese customers at 49%.

The practical target is a workaround that Washington has tried to close for years. Since 2022 the US has barred sales of Nvidia's most advanced chips to China. Many Chinese AI developers have instead rented computing power from cloud providers outside China to train their models. Smuggling has also continued: earlier this year several senior employees of the server maker Super Micro were charged with smuggling about $2.5 billion worth of Nvidia AI servers into China, The Next Web noted.

Beijing's response

Asked about the report at the Chinese foreign ministry's regular briefing on Tuesday, spokesperson Mao Ning said she was "not familiar with the specifics". According to the ministry's transcript, she said China hoped that all parties would provide "a fair, transparent and non-discriminatory business environment" for companies from all countries, including Chinese ones. At the same briefing she said China opposed "erecting digital iron curtains", in an answer about the World Internet Conference.

Beijing has pushed in the other direction too. A year ago it barred foreign AI chips from state-funded data centres in China to help domestic suppliers, The Next Web reported. The US and China have recently accused each other of stealing AI technology and threatening each other's security.

What it means for India

The FT report does not mention India, and Nvidia has not said which regions the rule covers. But Indian cloud and data-centre companies are building large Nvidia-based clusters, many of them supplying computing power to the government's IndiaAI Mission, and they sell capacity to foreign customers too. If the rule applies to them, their contracts will now have to keep Chinese customers in the minority.

The rule also points to a wider risk for India's AI plans. Access to the most advanced chips is now shaped by US export policy and by the commercial terms of a single American company. At the same time, India depends on China for much of the electronics hardware and rare earths that data centres and devices need. The pull between the two powers could raise costs and add conditions for Indian firms.

What happens next

Nvidia has not published the policy. Nor has it said how it will check partners' customer lists, or what happens to a partner that misses the threshold. The first test will come as operators sign new contracts, especially in markets where Chinese companies are large buyers of rented computing power.