Nvidia reportedly asks cloud partners for at least 51% non-Chinese revenue

Nvidia has reportedly asked some cloud partners to derive at least 51% of service revenue from non-Chinese customers. If linked to chip allocations, the guidance could turn customer identity into a constraint on AI infrastructure growth.

Nvidia has reportedly asked some data centre and cloud partners to generate at least 51% of service revenue from non-Chinese customers. Investing.com carried the Financial Times account.

Separate reporting describes the guidance as verbal, not yet mandatory, and says Nvidia has requested information about chip end users.

Neither report establishes a universal contractual requirement. The threshold is not a published US export regulation or a confirmed company-wide Nvidia policy.

Nvidia could make access to future GPU allocations depend partly on who ultimately purchases cloud computing services.

That would move commercial pressure beyond chip shipment destinations and into customer screening, revenue composition and infrastructure financing.

The remote access loophole is the target

Export controls can restrict shipments of advanced chips to China. Cloud services complicate that boundary because customers can access computing capacity without owning or importing the underlying hardware.

A Chinese developer could potentially rent capacity from servers in Singapore, Malaysia or another overseas market.

The Bureau of Industry and Security's May guidance describes licensing requirements involving advanced-computing items supplied to entities headquartered in China or Macau, including relevant entities operating elsewhere.

The guidance also permits bona fide data centre operators to continue using, storing and servicing covered items until further notice.

Those provisions do not establish a general exemption for remote access. Nor do they impose Nvidia's reported 51% customer-revenue threshold.

The responsibilities differ: BIS determines regulatory obligations; Nvidia could separately influence commercial access through supplier relationships and future hardware allocations.

A hypothetical operator earns 60 units of revenue from Chinese customers and 40 from non-Chinese customers.

If Chinese revenue stays constant, non-Chinese revenue must exceed approximately 62.45 units to represent at least 51% of the total.

That requires more than 22 additional units of non-Chinese revenue, an increase exceeding 56% from the original 40.

The arithmetic exposes the mechanism: an operator could face pressure to recruit new customers, decline otherwise profitable demand or change its business mix.

But the reported guidance leaves the measurement period, treatment of resellers and definition of customer nationality unspecified. Those unresolved details determine whether the threshold is operationally enforceable.

A sales rule can reshape the neocloud model

Specialist AI clouds commit capital to GPUs, facilities and power infrastructure before all future demand is secured.

For an operator dependent on Chinese model developers, a customer-mix condition could create a conflict between immediate utilisation and eligibility for additional Nvidia systems.

The pressure would build in stages.

First, concentrated Chinese demand raises exposure to the reported threshold. Second, the operator needs additional non-Chinese revenue or reduced Chinese business. Third, customer acquisition costs, utilisation or contract economics may deteriorate. Finally, uncertainty over future GPU allocations can complicate capacity expansion and financing.

The downside could be asymmetric. Nvidia may have alternative buyers for constrained supply, while an operator remains committed to its infrastructure.

That is a conditional economic inference, not evidence that Nvidia has withheld allocations.

Nvidia's fiscal 2026 annual report identifies cloud providers, neocloud builders and AI model makers among its indirect customers. It also separately reports China revenue.

The two measures cover different business activities.

Nvidia's geographic revenue disclosure concerns its own reported business. The proposed cloud threshold would concern the customer composition of downstream service providers.

Those measures cannot be substituted for one another. Revenue earned by Nvidia outside China leaves the nationality of customers ultimately consuming overseas cloud capacity unknown.

The supplied evidence contains no comparable public customer-mix series showing which operators would meet or breach 51%.

China exposure is already a planning variable

In its second-quarter fiscal 2027 results, Nvidia said its outlook assumed no data centre compute revenue from China.

That assumption concerns Nvidia's outlook. It does not demonstrate that Chinese demand for overseas cloud computing has disappeared.

The commercial comparison is therefore between constrained direct China-facing compute revenue and potential indirect exposure through overseas cloud customers.

Chinese developers could respond by using domestic accelerators, changing overseas providers or adapting workloads. The supplied evidence leaves the relative cost and technical equivalence of those options unresolved.

For cloud operators, the next potential cost is verification.

A locally registered reseller may serve a different ultimate customer. If Nvidia examines end-user identity, incorporation location alone may not settle compliance questions.

Operators could respond with beneficial-ownership disclosures, downstream-user identification, contractual warranties and workload-access controls.

Those requirements could increase onboarding costs and delay deployment. They could also favour larger providers with established compliance systems.

But the strongest counterargument is that the reported threshold may remain flexible commercial guidance. Diversified providers could satisfy it while continuing to serve Chinese customers.

A second counterargument is that customer identity may be difficult to establish consistently across intermediaries. Without a documented calculation method, the percentage risks becoming a discretionary allocation signal rather than a reproducible compliance test.

What matters is whether Nvidia actually links verified customer composition to future chip supply. Reaching 51% mathematically is only part of the question.

Possible outcomes

Over the next three months, private screening intensifies

The near-term base case is tighter end-user questioning without a published universal threshold.

Evidence would include authenticated partner instructions, new contractual disclosures or allocation notices referencing customer identity.

The scenario weakens if no such records emerge and partner onboarding practices remain demonstrably unchanged.

Over the next 12 months, customer mix affects allocations

If Nvidia applies the reported threshold commercially, diversified operators could gain an allocation advantage over China-focused competitors.

The test is documented allocation treatment explicitly linked to customer composition, rather than ordinary hardware scarcity.

The scenario fails if comparable China-focused operators continue receiving new systems without customer-mix restrictions.

Over the next 36 months, regulatory clarification changes the mechanism

BIS could establish more explicit requirements for remote access to advanced computing resources, reducing reliance on discretionary supplier screening.

A published rule defining covered users, provider obligations and relevant remote-access conditions would support this scenario.

Continued dependence on private allocation decisions without equivalent regulatory clarification would weaken it.

These are conditional scenarios, not probability estimates.

CTOL expects at least one major overseas AI cloud operator to publish stronger end-user or beneficial-ownership requirements by June 30, 2027.

The forecast fails if no qualifying public requirement appears by that date. A positive result would not independently establish Nvidia as the cause.

Evidence to watch

Three records would resolve the central uncertainty. The first is an authenticated Nvidia instruction defining the threshold and its scope. The second is a partner contract or allocation notice tying customer composition to chip access. The third is a BIS rule explicitly governing the relevant remote-computing arrangements.

Until then, the 51% figure remains a reported commercial threshold, not a verified universal requirement.

The investment implication is conditional but consequential. If Nvidia makes future GPU allocations depend on ultimate customer identity, overseas cloud operators will compete on capacity, price and performance. They will also need to prove who uses the compute.