Nvidia caps Chinese business at 49% for data center clients

What did Nvidia reportedly do?

The Financial Times reported on Tuesday that Nvidia is ordering its data center partners to scale back work with Chinese customers. Citing five people with knowledge of the matter, the FT said Nvidia's cloud computing clients must now sell at least 51% of their services to non-Chinese companies.

The instructions were delivered verbally through Nvidia's sales force and regional teams rather than in written documents, according to the report. Nvidia previously had no such restrictions in place.

Tom's Hardware framed the same reporting as a 49% cap: data center operators who sell services to Chinese companies must track the share of their business those customers represent and limit it to 49% of the total, ensuring at least 51% goes to non-Chinese outfits.

Which contracts does it apply to?

The report says the new rules only affect future contracts and not existing agreements. That distinction matters for data center operators with large Chinese customer bases: they will not be forced to drop current Chinese customers, but new business must be rebalanced toward non-Chinese demand.

The policy applies globally rather than only to data centers located in Asia, according to Tom's Hardware.

What did Nvidia and China say?

Nvidia neither confirmed nor denied the report to AFP, but issued a statement: "data centers and clouds built on American chips are the foundation that American model and application makers need to compete worldwide."

The company added: "Longstanding U.S. policy encourages every cloud to build on American infrastructure and to win business everywhere they can," and "America cannot win the AI race unless America's AI infrastructure wins worldwide—if the world's clouds rely on foreign-made chips, the model and application layers will be foreign-built as well."

Asked about the report, Chinese foreign ministry spokeswoman Mao Ning said she was not aware of it. "As a matter of principle, we consistently hope that all parties can provide a fair, transparent and nondiscriminatory business environment for companies from all countries, including Chinese companies," she said.

Why does cloud access matter for China?

Chinese technology companies have relied on leasing advanced computing capacity through overseas cloud arrangements for AI work, including model training, rather than importing restricted hardware directly. Washington has been tightening controls on that channel alongside chip export rules in place since at least 2022.

The reported Nvidia policy adds a commercial restriction beside existing US export controls. By limiting how concentrated a partner's customer base can be in Chinese demand, it could reduce the pool of overseas Nvidia-powered capacity available to Chinese AI developers.

What is the enforcement backdrop?

Restrictions on Nvidia hardware reaching China have been tested by smuggling cases. Yahoo Finance reported that several senior employees at server maker Super Micro, which assembles Nvidia-based servers, were charged earlier this year in connection with the alleged smuggling of about $2.5 billion worth of Nvidia AI servers to China. The charges concern alleged misconduct and do not establish guilt.

Tom's Hardware also noted that China's state-owned telecoms giant China Telecom was recently found to have Nvidia hardware in its data centers amid a thriving market for the chips outside whitelisted sales channels.

What does this mean for the AI infrastructure market?

If the FT reporting is accurate, Nvidia is pushing its data center partners to make customer geography a compliance consideration, potentially affecting how capacity is marketed in Asia and beyond.

For cloud providers that historically depended on Chinese demand, the 51% threshold creates an incentive to diversify toward other markets. For Chinese AI developers, it could make overseas compute access less predictable. Nvidia, which AFP notes is the world's most valuable company thanks to the data center buildout, has not formally confirmed the policy, and the report says no written documentation of it exists.

How would the reported policy be enforced?

If the FT's reporting is correct, the requirement is not written into contracts or formal documentation. Yahoo Finance reported the instructions were delivered verbally through Nvidia's sales force and regional teams, which makes the policy harder to audit and easier to change.

For data center operators, that means customer geography becomes a commercial consideration in new deals even without a contractual clause. A partner whose pipeline is heavy on Chinese demand would need to either find non-Chinese customers or accept less Nvidia capacity, because the reported threshold is a ceiling on the mix rather than a ban on any single customer.

Tom's Hardware noted the reporting applies globally rather than only to data centers in Asia, and that even older Nvidia servers have become so sought-after that used values have reportedly tripled this year, a sign of how constrained supply is outside approved channels.

What does this mean for AI development in China?

Chinese AI developers have increasingly relied on overseas cloud capacity to run large model training jobs, since importing the underlying chips directly has been restricted since at least 2022. Yahoo Finance said the reported Nvidia policy could make that leasing route less accessible, particularly for providers that historically concentrated on Chinese demand.

The report does not say Chinese customers would be banned outright from Nvidia-powered clouds; the 51% requirement limits the share of services a partner can sell to them. Still, any shift in cloud strategy takes time, and the uncertainty itself hits planning for training clusters that are booked months in advance.

The backdrop remains tense: the US has been tightening cloud-adjacent rules alongside chip export controls, and Chinese officials have pushed back on restrictions as discriminatory. Nvidia, meanwhile, has said its American-built infrastructure is what lets US model and application makers compete worldwide.