Nvidia tells data center operators to limit business from Chinese customers with new global restrictions, report claims

Nvidia reportedly told data center operators that at least 51% of cloud-service business must go to non-Chinese companies. Chinese customers would therefore account for no more than 49%, marking a new restriction on customer mix rather than a simple location ban.

The rule reportedly applies worldwide, including data centers outside Asia, and covers new or future contracts. An operator serving Chinese customers in Europe would still need to track its overall customer split. It would need to keep Chinese business below half. Nvidia’s sales and regional teams reportedly communicated the policy, but it has not appeared in official documentation.

This comes as U.S. policy seeks to slow China’s AI progress by restricting access to Nvidia’s high-performance chips. Existing Nvidia hardware-leasing agreements are reportedly unaffected. New deals could reduce the pool of customers and squeeze data center growth. Enforcement may be difficult: China Telecom was recently found with Nvidia hardware despite existing restrictions.

Data center operators run the buildings, power, cooling, networks, and servers that process digital workloads. Cloud providers use those facilities to rent computing power, including Nvidia-equipped servers, instead of customers buying machines themselves. This matters because Nvidia chips are widely used for AI servers and data centers.

For example, a cloud company can install Nvidia hardware in its data center and offer customers remote access through online services. Customers send workloads to those servers, while the provider owns or leases the equipment and manages it. Under Nvidia’s reported new policy, providers must keep Chinese customers below 49% of new business, leaving at least 51% for non-Chinese companies.

The policy applies globally, not just in Asia, but reportedly excludes existing Nvidia hardware-leasing agreements. Nvidia sells chips only to whitelisted companies, yet China Telecom was recently found using Nvidia hardware. That shows how access can persist through black-market channels or existing arrangements, while tighter rules may shrink operators’ customer pool.

The reported rule would cap Chinese customers at 49% of a provider’s overall business. In practice, a data-center operator would need to ensure that at least 51% of its services went to non-Chinese companies, rather than serving Chinese customers without a stated ceiling.

For example, if a provider had 100 equivalent service contracts, no more than 49 could involve Chinese companies, while at least 51 would need non-Chinese customers. The provider would have to track that customer mix, apparently for new and future contracts; existing Nvidia hardware leasing agreements would not be affected.

The policy reportedly applies worldwide, not only to Asian data centers, and Nvidia’s sales and regional teams communicated it without public documentation. The article says Nvidia chips remain highly sought after: even older servers tripled in value, and China Telecom was found with Nvidia hardware. Providers may therefore face fewer eligible customers and more competition for non-Chinese contracts, potentially shrinking operations.

The policy is global because Nvidia is trying to control access to its high-performance data-center hardware wherever customers operate. The reported rule requires cloud providers to keep Chinese business at 49% or less, leaving 51% for non-Chinese companies. That matters because cloud services can deliver computing power across borders, not just from Asian facilities.

For example, a data center outside Asia could still face the rule when signing new contracts with Chinese companies. Nvidia’s sales and regional teams reportedly communicated the policy, although no official document appears to spell it out. The article gives no stated reason for excluding existing leases; it reports that the restriction is prospective.

The policy fits U.S. efforts to slow China’s AI progress, since Nvidia chips power many AI servers. Yet restrictions have not fully closed access: China Telecom was recently found with Nvidia hardware. Going forward, operators may face fewer customers and tougher competition for contracts.

Operators could lose scale and face tougher economics if Chinese customers disappear. Nvidia’s reported policy applies worldwide to new contracts, requiring 51% of business from non-Chinese companies. That could leave costly capacity underused, pressuring prices upward through higher costs or downward through spare capacity.

China Telecom was recently found with Nvidia hardware despite existing restrictions. That example shows how difficult enforcement can be. If providers lose Chinese contracts, they may spread fixed facility costs across fewer customers, making expansion less attractive.

Sales teams reportedly communicated the policy, but Nvidia has not published it formally. Existing leasing agreements are unaffected, while older Nvidia servers reportedly tripled in value this year. Future facilities could be smaller or delayed as operators seek replacement customers, though no price change is confirmed.

Chinese companies can still encounter Nvidia hardware through cloud providers, older servers, and a reported black market. The restrictions control eligible sales and customer relationships, but they have not eliminated hardware already circulating or services sold by global operators.

A striking example is China Telecom, a state-owned Chinese telecoms giant. It was recently found to have Nvidia hardware in its data centers, despite the existing limits. The article says Nvidia’s older servers became extremely sought-after. Their values tripled, showing how scarcity can drive resale activity.

Nvidia’s reported new policy does not close these routes. It applies globally to new and future leasing contracts, requiring providers to keep Chinese business at 49% or less. Existing agreements are unaffected, while black-market demand and already deployed equipment remain ongoing challenges.

Nvidia’s chips are the go-to hardware for AI servers and data centers worldwide. That makes them central to building and running advanced AI systems. Their high performance supports the huge calculations needed to train models and produce answers quickly.

The article notes that Nvidia chips are the go-to option for AI servers and data centers around the globe. In practice, their specialized computing power handles many calculations simultaneously, accelerating model training and responses produced by AI applications. That usefulness explains why even older Nvidia servers reportedly tripled in value when access became harder.

Access is now politically important, not merely technical. Nvidia reportedly told global customers to cap Chinese business at 49% in new contracts, leaving 51% for non-Chinese companies. Black-market demand and China Telecom’s reported hardware suggest restrictions have not eliminated access, but customer pools may shrink.