Nvidia is asking some cloud customers to limit new business with Chinese companies, a move that addresses access to its advanced AI chips through overseas data centers rather than direct hardware shipments. The Information reported, citing three people familiar with the matter, that the company wants those providers to develop a more diverse customer base.
The request includes Nvidia-backed neoclouds, providers focused on renting out computing capacity for artificial intelligence workloads. It is not mandatory and applies to future contracts rather than agreements already in place. That distinction makes the initiative a push to change the composition of future business, not an instruction to terminate existing customers.
Cloud providers are also being asked to identify the end users accessing Nvidia chips. The Financial Times reported that sales representatives and regional teams delivered the guidance verbally. The reported approach relies on Nvidia’s relationships with the companies buying and operating its hardware, rather than a newly announced government restriction.
At issue is a distinction between exporting a processor and selling remote access to one. U.S. export controls restrict shipments of advanced chips to China, but generally do not cover access through cloud services. Chinese AI developers can therefore rent computing capacity at data centers outside China without importing the underlying hardware.
For AI development, remote access can provide a practical alternative to owning processors. Cloud customers submit workloads to machines operated by a provider, with the calculations performed wherever those machines are located. The developer does not need physical possession of the chips to use their processing power for tasks such as training or running AI models.
That operating model complicates oversight. The company purchasing Nvidia hardware may be a cloud operator in one country, while the organization ultimately using its capacity is based elsewhere. Requests for end-user identities would give Nvidia more visibility into that relationship, although the reports do not establish how providers would verify those identities or how Nvidia would respond to the disclosures.
U.S. authorities had already been examining Chinese AI companies’ access to advanced chips through foreign data centers after allegations involving Moonshot AI, the developer of the Kimi chatbot. Those allegations form part of the backdrop to the scrutiny; they do not, on their own, establish that Moonshot or a cloud provider violated export controls.
The reported guidance leaves an important boundary intact: existing contracts are not covered. Any shift in the customer mix would therefore depend on the new agreements cloud providers choose to pursue and whether they follow Nvidia’s nonbinding request. The available reporting does not specify a numerical limit on Chinese business or a timetable for reducing it.
Sources: The Decoder