Nvidia reportedly tells cloud partners to sell at least 51% outside China

Nvidia is reportedly telling its cloud partners to sell at least half their services to customers outside China. The Financial Times reported the policy on November 3, 2026, and AFP's summary in the Taipei Times a day later gives the one number that matters: 51%. Nvidia did not confirm or deny it.

What was reported

According to AFP's account of the Financial Times story, Nvidia is telling its data center partners to scale back work with Chinese customers. The FT, citing five people with knowledge of the matter, said Nvidia's cloud computing customers must sell at least 51% of their services to non-Chinese companies. The new rules apply only to future contracts, so existing agreements are not affected.

The report does not say when the rules take effect or which partners are covered. Those gaps matter, because a rule that applies to a handful of neoclouds is a different thing from one that applies to every hyperscaler. We could not open the FT original, so everything here is attributed to AFP's summary.

What Nvidia said

Nvidia neither confirmed nor denied the report. In a statement quoted by AFP, it said: "datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide." It added: "Longstanding US policy encourages every cloud to build on American infrastructure and to win business everywhere they can."

The statement argues for the opposite of retreat: "America cannot win the AI race unless America's AI infrastructure wins worldwide," and "if the world's clouds rely on foreign-made chips, the model and application layers will be foreign-built as well." Read carefully, that defends selling American chips broadly, and says nothing either way about a 51% target.

China's Foreign Ministry spokeswoman Mao Ning said she "was not aware of it," and said Beijing hopes all parties offer a "fair, transparent and nondiscriminatory business environment" to companies from all countries.

What this does not tell us

The story is about who Nvidia's partners sell to, not how many chips Nvidia can make. Supply is a separate constraint, and Bloomberg published its own November 4 report on Chinese orders for Nvidia's RTX Pro line. We could not open that article, so we do not use its figures here.

Two things are clear from what we did open. Nvidia is under political pressure over China, with AFP noting that the US and China have each accused the other, over recent months, of stealing AI technology and jeopardizing security. And Nvidia is the world's most valuable company, driven by the global buildout of data centers, so a change to how its biggest customers contract is a market event as well as a policy one.

The export-control background

The chips Chinese buyers can legally get have changed this year. Bloomberg Law reported on February 26, 2026 that Nvidia had said it "secured a license to ship a small number of its less advanced H200 chips to customers in China," subject to inspection in the US and a 25% duty. The article noted it was "still unclear whether Beijing will allow even that limited return," and that Nvidia was not including any China data-center revenue in its first-quarter outlook.

Tom's Hardware later headlined that Chief Executive Jensen Huang said Nvidia had received orders from Chinese customers for H200s and had licenses from the US government, with H200 manufacturing restarting. We only saw the headline, so we use it for that claim alone.

For the underlying flow of hardware, see Epoch AI's estimates in our summary of its chip imports explorer, which put the US share of imported AI compute well above China's as of June 2026.

Why a customer-mix rule would matter

A requirement that cloud partners sell most of their capacity outside China does three things, if the report is accurate.

• It could limit the rental route. If a partner must sell most of its capacity outside China, less of any new cloud is left for Chinese customers. That is our inference from the 51% figure, not something the FT or Nvidia said.
• It gives Nvidia a political answer. With each country accusing the other of AI theft, a visible customer-mix rule is something Nvidia can point to without waiting for a statute. Again, that is analysis.
• It does not stop existing capacity. The rule covers future contracts only, so capacity already contracted is untouched.

None of this has been confirmed by Nvidia, and the 51% figure comes from five unnamed sources.

What would settle it

Confirmation would come in one of three ways: a letter to partners that surfaces, a customer saying publicly that its contract changed, or Nvidia describing a policy in its next earnings call. We would also look for a Commerce Department comment. Without one of those, this stays at "reported."

Our position: this is a credible story with a thin evidentiary base. The 51% threshold is concrete enough to test and vague enough that Nvidia could deny it later. Readers following the compute race should hold it as a lead, and watch whether Chinese cloud pricing or rental availability shifts over the next two quarters, which would be an outside signal that something changed.

Questions for the next two quarters

If the rule is real, several details will decide how much it bites. The first is how Nvidia measures a cloud's customer mix: by revenue, by compute hours, or by contract count. The second is who audits it. A 51% share is easy to state and hard to verify across resellers, subsidiaries and multinational customers whose models are trained in one country and sold in another.

The third is what counts as a "Chinese" customer. A company headquartered elsewhere but with large Chinese operations, or a startup financed from China but incorporated abroad, may or may not be inside the line. The AFP summary does not say.

Finally, there is the question of the model labs themselves. A tighter cloud channel would test whether Chinese labs depend on rented Nvidia capacity or on domestic chips and their own data centers. That is a testable question, and it is one of the strongest reasons to follow this story even while it is unconfirmed.