Nvidia is ordering its cloud computing and data centre partners to scale back work with Chinese customers, the Financial Times reported on Tuesday, citing five people familiar with the matter. Under the new condition, at least 51 per cent of a partner's services must be sold to non-Chinese companies — a threshold the industry is already calling a "de-Sinicization" rule, with no precedent of its kind from a chipmaker.
The instruction has not been issued as a written policy. According to the report, it has been passed verbally to customers by Nvidia's sales staff and regional teams since the summer, and it applies to partners worldwide, not only to data centres in Asia. It covers only future contracts: existing leasing agreements are not affected retroactively.
A rule about who buys the compute, not where it ships
The move targets one of the last open routes Chinese firms have to advanced American AI hardware. United States export controls already block the sale of Nvidia's most advanced chips, such as the Blackwell-based B300, into China. But Chinese AI companies can still rent computing power hosted in data centres abroad — in Singapore and Malaysia, for example — through remote cloud connections, without any chip ever crossing into China. The 51 per cent threshold adds the identity of the buyer to the question of where the hardware is shipped.
It follows a pattern of tightening. In July, Nvidia cut by more than half its list of Asian customers approved to buy its AI chips, the Financial Times reported. The new rule also follows United States Commerce Department guidance issued in May aimed at keeping advanced chips away from Chinese entities operating through third countries. The report cited a person with direct knowledge of the trade saying a rack of B300 chips, which sells for about $600,000 internationally, changed hands in China last month for 18.5 million yuan, or roughly $2.8 million.
Southeast Asia's cloud providers feel the squeeze
The immediate pressure falls on cloud providers in Southeast Asia, where Chinese firms are among the largest customers of rented GPU capacity. Some providers fear they may have to scale back data centre operations if they cannot attract enough non-Chinese customers, the report said. A Singapore-based cloud provider told the Financial Times that it is hard to find enough non-Chinese customers outside the United States to fill the gap.
Markets reacted quickly. Chinese tech stocks slid on Tuesday, with Shanghai's STAR 50 index down 3.8 per cent after the report emerged. The reports did not say whether the 51 per cent share is measured by revenue, computing capacity or number of customers, or how Nvidia defines a Chinese company — details that will decide how painful compliance actually is.
What Nvidia will and won't say
Nvidia neither confirmed nor denied the report. In a statement, the company said that "datacentres and clouds built on American chips are the foundation that American model and application makers need to compete worldwide," and that United States policy has encouraged cloud companies to win as many customers as possible on American infrastructure. It did not confirm the 51 per cent figure or any enforcement details. China's foreign ministry spokeswoman Mao Ning said she was not aware of the report, and Reuters said it could not immediately verify it, though Bloomberg and The Information carried similar reports.
For now, the rule's power lies in its ambiguity: a verbal threshold, delivered without documents, that still forces every Nvidia cloud partner to weigh its next contract against its Chinese client base.



