Nvidia reportedly asks cloud clients to sell at least 51% of services outside China
On November 3, 2026, the Financial Times reported that Nvidia has asked its cloud computing customers to pull back their exposure to Chinese companies. Bloomberg carried the account the same day, attributing it to unidentified people. According to the FT, customers must make sure at least 51% of their services are sold to companies outside China. AFP, citing the FT, said the figure came from five people with knowledge of the matter. Reuters could not immediately verify the report. The instruction reportedly passed through Nvidia's sales and regional teams in recent months, and the company issued no written notice.
The requirement reportedly applies only to future contracts, so existing agreements stay in place. Nvidia neither confirmed nor denied the report. In a statement to AFP, it said clouds built on American chips underpin the work of American model makers, and that longstanding US policy encourages every cloud to win business wherever it can. The FT also reported that Nvidia has cut its authorized customers in Asia by more than half after tighter compliance checks, and that it is asking cloud clients to identify the end users of its chips. A Singapore-based cloud provider reportedly warned that finding enough non-Chinese customers could be difficult.
The reports do not explain how the 51% share is measured or how Nvidia would enforce it, so the exact scope of the instruction remains unclear.
Why it matters
For AI companies renting Nvidia-based capacity from cloud providers, the reported rule would make customer mix a condition of access to chips. Operators with Chinese customers may need to reprice, rewrite contracts, or turn away business, and new capacity may depend on finding non-Chinese buyers. Startups using overseas GPU clouds should expect tighter questions about who their end users are and where their workloads run.
For investors, the reported rule adds a diligence item: customer geography and export-control exposure for GPU cloud operators, especially those built on Nvidia allocations. Because the rule reportedly covers only future contracts, the first effect would likely fall on new bookings rather than existing revenue, which matters for how growth forecasts are modeled. Founders renting capacity should confirm where compute physically runs, who owns the hardware, and what customer-location limits apply, and keep a second region or vendor ready.
Key points
- 1The Financial Times reported the instruction on November 3, 2026, citing people with knowledge of the matter.
- 2Nvidia reportedly told cloud clients that at least 51% of their services must go to non-Chinese companies.
- 3The reported rule applies to future contracts and leaves existing agreements unaffected.
- 4No written notice was issued, and the instruction reportedly passed through Nvidia's sales and regional teams.
- 5
