Nvidia is reportedly ordering its data centre partners to reduce exposure to Chinese customers in a new threat to the country’s AI ambitions, according to the Financial Times.

Citing five people with knowledge of the matter, the paper reported that the world’s largest chipmaker has told its cloud computing clients they must sell at least 51 per cent of their services to non-Chinese companies. This is the first such restriction the company has imposed.

The policy has been communicated verbally by Nvidia’s sales force and regional teams, rather than made explicit through documentation, the people said.

China’s AI industry is currently heavily reliant on Nvidia chips, despite restrictions on their use by both China and the US. Domestic manufacturers such as Huawei are struggling to meet demand, leaving many Chinese AI developers bridging the gap with older or consumer-grade Nvidia GPUs.

According to the FT, many Chinese firms have also previously relied on leasing compute from foreign data centres for advanced AI tasks, including training new models.

These additional rules apply to Nvidia’s clients globally, not just the data centres in Asia, two of the people told the paper. They only affect future contracts, not existing lease arrangements, however.

A Singapore-based group told the FT that cloud providers were struggling to comply with this restriction because it is “difficult to find enough non-Chinese clients outside the US”, adding that if the policy is strictly implemented, “it will definitely hurt business”.

In a statement to the paper Nvidia said: “Datacentres and clouds built on American chips are the foundation that American model and application makers need to compete worldwide.

“Longstanding US policy encourages every cloud to build on American infrastructure and to win business everywhere they can. America’s longstanding policy has always supported the build-out of American infrastructure worldwide, bringing prosperity, influence, and jobs home.”

---