- Nvidia has reportedly told cloud and data center partners worldwide that at least 51% of services on new contracts must be sold to non-Chinese companies, leaving existing agreements untouched.
- Nvidia did not confirm the threshold, and it is unclear whether the share is measured by revenue, compute hours or customer count, or how a buyer is classed as Chinese.
- Nvidia's own 10-Q shows the leverage behind the request: $36 billion of cloud service commitments to AI cloud partners and $3.5 billion of guarantees on their data center leases.
Nvidia has told cloud providers and data center operators around the world that at least 51% of the computing services they sell under new contracts must go to customers outside China, according to a Financial Times report on November 3 that cites five people familiar with the instruction. The rule leaves Chinese companies with no more than 49% of a partner's new business and closes part of the route Chinese AI developers use to reach advanced Nvidia hardware without importing it.
Nvidia's 51% rule targets Chinese AI labs renting GPUs abroad
US export controls bar Nvidia's most advanced chips from shipment into China, yet a Chinese company can still rent them in a data center in Singapore or Malaysia and train models remotely. Tencent alone has leased about 100,000 advanced AI chips in Oracle's Southeast Asia data centers for roughly $7 billion. The reported requirement covers the GPU clouds that buy Nvidia systems in bulk and resell the capacity, in every region.
Nvidia neither confirmed nor denied the 49% figure. In a statement issued after the report, it framed the issue as one of global market share for American hardware.
“Datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide. Longstanding US policy encourages every cloud to build on American infrastructure and to win business everywhere they can.”
NVIDIA, statement on the Financial Times report, November 3, 2026
China's foreign ministry said it was not aware of the report. Operators in Southeast Asia, where Chinese firms are among the largest tenants, face the hardest arithmetic, and one Singapore-based provider told the FT that strict enforcement would hurt its business.
Nvidia's own filings show why partners are likely to comply
The instruction carries weight because many of these clouds depend on Nvidia for more than chips. In its quarterly report for the period ended July 26, Nvidia disclosed $36 billion of AI cloud agreements, under which partners buy its data center systems and Nvidia commits to buy cloud capacity back from them, with a share of their third-party revenue if certain criteria are met. It also carries up to $3.5 billion of guarantees on those partners' land, power and shell leases.
| Cloud service commitments to AI cloud partners | $36 billion |
| Maximum exposure on partners' data center lease guarantees | $3.5 billion |
| Revenue from China-headquartered customers, first half of fiscal 2027 | $12.4 billion |
| China's share of first-half revenue of $177.8 billion | 7.0% |
The same filing explains the gap the rule is meant to cover. Nvidia books revenue by the headquarters of its direct customer and warns that “the end customer and shipping location may be different.” It also states that because it has no physical control of products after sale, it “must also rely on the compliance programs of our customers and partners.”
| Customer headquarters | Q2 fiscal 2027 revenue | Share of total |
|---|---|---|
| United States | $60.1 billion | 62.4% |
| Taiwan | $27.0 billion | 28.0% |
| China (including Hong Kong) | $7.9 billion | 8.2% |
| Other | $1.3 billion | 1.3% |
Source: NVIDIA Corporation, Form 10-Q, quarter ended July 26, 2026. Revenue by direct customer headquarters location.
The China line counts only direct sales to Chinese-headquartered buyers, so Chinese demand served through a Singapore cloud appears under that cloud's headquarters. That hidden share is what a partner-level cap reaches, and it sits alongside direct demand such as the $35 billion of chip orders placed by ByteDance, Alibaba and Tencent last month.
Nvidia reports third-quarter results on November 17, the first chance for analysts to ask how the requirement is measured and policed. By writing customer-mix terms into its own commercial relationships, Nvidia is applying a limit to rented AI infrastructure that US export law has yet to impose, and the clouds that rely on its financing have little room to refuse.
Santage is committed to independent, transparent journalism. This article is produced in accordance with Santage's Editorial Standards and aims to provide accurate and timely information. Readers are encouraged to verify information independently.