Nvidia Orders Data Center Partners to Cut Back on Chinese Customers

Nvidia has instructed its cloud computing partners to make non-Chinese companies the majority of their customer base, tightening a route Chinese AI developers use to access advanced American chips overseas.

Providers must sell at least 51% of their services to non-Chinese companies, according to five people cited by the Financial Times.

The requirement introduces a customer restriction that didn't previously exist.

The measure reaches beyond Asia.

Two people familiar with the instructions said they cover Nvidia clients worldwide, although existing leases are unaffected and the restrictions apply only to future contracts.

Overseas Computing Under Pressure

Chinese technology groups have used computing capacity rented abroad to train AI models and perform other demanding tasks.

These arrangements let them use advanced Nvidia processors without importing the hardware into China.

That access has become more important as U.S. export controls block direct sales of Nvidia’s most powerful products and tighter customer screening constrains supplies reaching China through intermediaries.

The new requirement could force cloud providers to turn away prospective Chinese business unless they secure enough customers from other markets.

Chinese groups have been the largest customers for data centers in parts of Southeast Asia, while China and the U.S. dominate global demand for computing power.

One Singapore-based provider said it was “difficult to find enough non-Chinese clients outside the U.S.” The group added: “If strictly implemented, it will definitely hurt business.”

Operators unable to attract sufficient alternative demand could have to reduce the size of their businesses, according to the report.

Instructions Delivered Verbally

Nvidia introduced the requirement over the summer, with sales staff and regional teams communicating it orally rather than through written rules, people familiar with the matter said.

Two people described the decision as a pre-emptive response to criticism from U.S. lawmakers and others over Nvidia’s dealings with China.

Washington is seeking to restrict both the intermediaries supplying a black market for Nvidia chips and Chinese companies’ access to advanced processors through overseas leases.

Nvidia did not confirm or deny the specific 51% threshold in its response. Instead, the company emphasized the importance of expanding American computing infrastructure internationally.

“Datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide,” Nvidia said.

Scarcity Drives Up Prices

The customer requirement follows an earlier tightening of Nvidia’s approved buyer network.

The company has more than halved the number of Asian customers authorized to purchase its AI chips after introducing a list of buyers that passed tougher compliance checks this year.

Those checks reduced smuggled supplies reaching China and pushed more Chinese technology groups toward overseas rentals, the report said.

The latest restrictions put that alternative under additional pressure.

The shortage is reflected in prices.

A rack of Nvidia’s B300 chips sold in China last month for Rmb18.5 million, or about $2.8 million, according to a person with direct knowledge of the transaction cited by the FT.

Its international price is roughly $600,000.

U.S. export controls prohibit sales of the B300 to China. The processor is among Nvidia’s most advanced products and is used by American AI companies including Anthropic and OpenAI.