Loading page…
The requests cover future contracts and end-user identities. Accounts differ on whether the guidance is voluntary or a required customer-share limit.
Listen to this story
Nvidia is reportedly steering cloud operators—including neoclouds it backs—away from taking as much business from Chinese firms and toward identifying who ultimately uses its chips. That would extend pressure beyond hardware exports to rented compute, a route that generally lets Chinese AI developers access advanced Nvidia chips from overseas data centers. The policy’s force remains unsettled: reports differ on whether it is voluntary guidance or a binding global limit, and the reported instructions are not formally documented.
The reported change covers future contracts, not deals already in place.
The Financial Times account, as summarized by Tom’s Hardware, puts Chinese customers at a maximum 49% of an operator’s business; The Decoder describes voluntary guidance.
Nvidia said current U.S. policy encourages cloud providers to build on American infrastructure and accept business from everywhere they can.
Nvidia has reportedly asked some cloud providers to take less business from Chinese companies and identify the end users accessing its chips. The requests would reach beyond hardware shipments to rented computing capacity. The Information reported the push to reduce Chinese business, citing three people familiar with the matter, according to The Decoder.
The outreach includes Nvidia-backed neoclouds, providers that rent out AI computing capacity. The reported goal is a more diverse customer mix. The guidance concerns future contracts, rather than agreements already in place.
How firmly Nvidia is pressing the change remains unclear. The Decoder describes voluntary guidance. Tom’s Hardware, citing the Financial Times, describes a required ceiling: Chinese companies would supply no more than 49% of an operator’s business, leaving at least 51% to non-Chinese customers. It says the direction applies globally, not only in Asia.
Nvidia communicated the instruction verbally through sales representatives and regional teams, the Financial Times reported, as summarized by The Decoder. Tom’s Hardware says the policy does not appear to have been set out in official documentation.
American chips are the foundation that American model and application makers need to compete worldwide.
Nvidia, in a statement carried by Tom’s Hardware
Nvidia’s statement also emphasized broad access: it said current U.S. policy encourages cloud providers to build on American infrastructure and accept business from everywhere they can.
The request addresses a distinction between moving a chip and using one. U.S. export controls restrict chip shipments but generally do not cover remote access through cloud services, The Decoder explains. Chinese AI developers can therefore rent computing capacity at foreign data centers and use advanced Nvidia chips without importing the hardware into China.
Overseas access had already drawn government attention. U.S. authorities had been examining Chinese AI companies’ use of chips through foreign data centers after allegations surfaced involving Moonshot AI, the developer of the Kimi chatbot, according to The Decoder. That scrutiny followed allegations, not an established finding of wrongdoing by Moonshot.
Loading discussion...
Join the conversation
Explain where a chip supplier’s oversight should end.
Be the first to share a perspective or an experience.
Reader comments
Newest comments first. Replies stay oldest first.