Nvidia China Curbs Hit Cloud Clients With 51% Non-Chinese Rule

Nvidia has told its cloud computing clients that at least 51% of their services must now be sold to companies that are not Chinese, the Financial Times reported on Tuesday, citing five people familiar with the matter. The Nvidia China instruction targets the data centers that rent out its AI chips and narrows one of the last routes Chinese tech firms use to reach advanced US hardware.

According to the report, the requirement was delivered verbally through Nvidia’s sales staff and regional teams rather than in written documents. Two of the people said it applies to Nvidia’s customers worldwide, not only to data centers in Asia. It covers new contracts only and does not affect existing leases.

What the Nvidia China Cloud Rule Changes

US export controls block the sale of Nvidia’s most advanced chips, such as the Blackwell-based B300, to China. But a chip installed in Singapore, Malaysia or Japan can still be rented over the internet, and many Chinese companies have leased overseas computing power to train AI models. Chinese firms are already among the largest customers of data centers in Southeast Asia.

The 51% floor does not ban Chinese customers outright. Instead, it caps how much of any single provider’s business they can account for. The reports did not say whether the share is measured by revenue, computing capacity or number of customers, or how Nvidia defines a Chinese company.

Earlier this year, Nvidia introduced a whitelist of buyers that passed stricter compliance checks, which cut the number of Asian customers allowed to buy its AI chips by more than half, according to the FT. The new rule follows US Commerce Department guidance issued in May aimed at keeping advanced chips away from Chinese entities operating through third countries.

Black Market Prices and Pressure on Cloud Providers

The tighter controls are already showing up in prices. A person with direct knowledge of the trade told the FT that 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. That is about 4.7 times the global price, and it echoes the record black market premiums seen earlier this year.

A Singapore-based cloud provider told the FT that it is hard to find enough non-Chinese customers outside the US and that strict enforcement would clearly hurt its business. Providers that cannot replace Chinese demand may have to scale back operations.

In a statement, Nvidia said data centers and cloud services built on American chips are the foundation US AI developers need to compete globally, and that US policy has encouraged cloud companies to win as many customers as possible on American infrastructure. The company did not directly confirm or deny the 51% figure. Bloomberg and The Information carried similar reports, while Reuters said it could not immediately verify the story.

What It Means

For Chinese AI labs, renting overseas GPUs is about to get harder and more expensive, which adds to pressure to switch to domestic chips from Huawei and others. For neocloud operators in Asia, the rule could force a rapid hunt for Western and local clients. It also lands as governments tighten their grip on data centers, as Russia’s new capacity decree showed this week.

Nvidia’s own ties to China remain under a spotlight too, following the recent Jensen Huang spy report.