Nvidia Reportedly Tells AI Cloud Clients to Reduce China Exposure

Nvidia has reportedly asked some data-centre customers to reduce their exposure to Chinese buyers, a sign that the contest over AI computing is now reaching beyond chip exports and into the business models of cloud providers. The Financial Times reported that Nvidia verbally told cloud customers in recent months that at least 51% of their services should be sold to non-Chinese companies. The reported condition applies to future contracts globally, rather than deals already signed.

The figure has not been independently confirmed. Reuters said it could not verify the account, and Nvidia’s response to the FT did not expressly confirm the reported threshold. That distinction matters because a verbal instruction to customers is not the same as a published policy or a government rule.

US export controls have limited direct sales of advanced AI chips to China. But a company can also seek access to computing services delivered by a cloud operator outside China. If Nvidia is pressing customers to limit how much of that capacity ultimately serves Chinese firms, it would extend commercial scrutiny from the destination of a chip to the customers using the data centre built around it.

For cloud providers, the reported 51% test would raise practical questions. They would have to know which businesses are buying their services, how resellers account for capacity, and whether a customer’s headquarters or ultimate ownership determines its classification. The FT report does not establish how Nvidia would enforce the requirement, or whether every customer received identical terms.

The move would also fit a market already shifting under geopolitical pressure. TechBooky reported that Nvidia’s AI chip shipments in China fell sharply in the first half of 2026 as Huawei narrowed the gap. That shipment data describes the market inside China; the new FT account concerns cloud services sold globally, so the two should not be conflated.

Nvidia has said US policy supports American infrastructure around the world. For now, the reported customer threshold is best treated as a significant claim requiring a clearer public explanation. If confirmed, it could affect where AI cloud providers seek revenue and how they document who uses their capacity. If not, the episode still shows how sensitive the boundary between hardware sales and cross-border compute services has become.