Nvidia Urges Cloud Partners to Limit China Exposure
Nvidia is asking cloud customers to diversify away from Chinese clients, tightening a remote-access loophole in US AI-chip controls.
Nvidia has begun urging some cloud-computing customers to reduce their commercial exposure to Chinese companies, according to reporting by The Information and the Financial Times. The guidance reportedly applies to future business and asks providers to identify the end users of Nvidia-powered capacity. It has been communicated through sales and regional teams and, so far, is described as non-mandatory rather than a formal contract prohibition.
The move targets an important gap in US export controls. While Washington restricts the shipment of advanced AI accelerators to China, overseas cloud providers can often rent access to those chips remotely. That arrangement has allowed Chinese developers to obtain compute without physically importing restricted hardware. The new pressure would make that route harder by encouraging Nvidia-backed neoclouds and other infrastructure operators to build a broader customer mix.
The commercial consequences are not yet clear. A reported 51-percent non-Chinese customer threshold would be materially stricter than ordinary compliance screening, but the available reporting does not establish that Nvidia has imposed a universal rule or that every cloud partner has accepted it. Existing leasing agreements also appear to be treated differently from new contracts.
Why it matters: this is a shift from regulating chips at the border to influencing who is allowed to consume them in the cloud. If enforced broadly, it could raise compute costs for Chinese AI labs, complicate access for international neoclouds, and push more demand toward domestic accelerators. It also places Nvidia in a difficult position: the company must protect access to its hardware while asking customers to walk away from a potentially significant market.