NVIDIA has told its cloud computing clients they must sell at least 51% of their services to companies outside China on future contracts, according to reports. The instruction applies to customers worldwide, not only those in Asia. Existing agreements reportedly remain unaffected.
NVIDIA is yet to confirm or deny the reports. In a statement the company said: “America cannot win the AI race unless America’s AI infrastructure wins worldwide – if the world’s clouds rely on foreign-made chips, the model and application layers will be foreign-built as well.” It added that “datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide.”
The condition targets a known workaround. US export controls already restrict the shipment of NVIDIA’s most advanced chips to China. Chinese AI developers have responded by renting capacity in overseas data centres so the chips never cross the Chinese border. A 51% non-Chinese floor on new contracts is designed to limit how much of that rented capacity can flow back to Chinese end users.
Mao Ning, Chinese foreign ministry spokeswoman, said she was not aware of the report when asked. She restated Beijing’s position that “as a matter of principle, we consistently hope that all parties can provide a fair, transparent and non-discriminatory business environment for companies from all countries, including Chinese companies.”
How Domestic Alternatives Compare
The timing lands as China’s domestic AI chip industry continues to scale. Huawei expects 2026 AI chip revenue of roughly $12 billion. NVIDIA generated about $17.1 billion in China revenue in 2025. The gap is narrowing, and the new contractual condition further reduces the ability of Chinese developers to treat overseas NVIDIA capacity as a reliable substitute for restricted hardware.
We have previously examined how domestic developers are already shifting toward local alternatives, including the Ascend line that Huawei is positioning as a CUDA-free stack.
What This Means for European Operators
The rule is global. That includes European neoclouds and data-centre operators that buy NVIDIA chips in bulk and rent the compute.
Nebius and Nscale are the two largest European-headquartered neoclouds by live power. Both, like their peers elsewhere, run primarily on NVIDIA silicon. Any future capacity they contract with NVIDIA may now need to demonstrate that a majority of the services sold from that capacity goes to non-Chinese customers. Operators that already have Chinese AI labs among their larger tenants, or that have used China-linked demand to underwrite new sites, will have to adjust either the customer mix or the contracts they sign.
This adds another constraint to a sector already limited by power access, grid queues, and capital. Finland’s data-centre pipeline, the UK’s grid delays (including Microsoft’s exit from Nscale’s Loughton site), and the broader European competition for power are well documented. Nscale’s prospectus listed more than $100 billion in contracted value against far less live capacity, a pattern common across the sector. A contractual customer-mix rule does not replace those bottlenecks, but it does change the commercial terms under which new NVIDIA-backed capacity can be financed and filled.
For European AI sovereignty efforts the effect is mixed. On one hand, it reinforces the preference for American infrastructure that NVIDIA itself describes. On the other, it makes European-based NVIDIA capacity slightly less available as a flexible outlet for Chinese demand, which may accelerate Chinese developers’ shift toward domestic chips such as Huawei’s Ascend series. European operators that want to stay fully compliant while still serving a global customer base will need clearer internal tracking of end-user nationality on new deals. We have mapped the different definitions of sovereign AI that European policymakers and companies are using.
What to Look For Next
Existing contracts will continue, at least for the moment. The impact depends on how strictly NVIDIA enforces the threshold in new negotiations and whether customers restructure ownership or contracting arrangements to stay under the 51% line. NVIDIA has not published the rule in writing; reports indicate it was passed verbally through sales and regional teams.
For Europe, the immediate question is which operators signing new NVIDIA capacity in 2026 and 2027 already have, or expect, a material share of Chinese end users, and how they will document compliance.
Author: Grace Sharp
See Also:
Can Huawei Replace NVIDIA in China’s AI Sovereignty Race?
China’s Semiconductor Industry No Longer Needs NVIDIA
What Is Huawei Ascend? The CUDA-Free AI Stack, Explained
Frequently Asked Questions:
NVIDIA has instructed cloud and data-centre clients that on future contracts at least 51% of services must be sold to companies outside China. This limits China-related business to a maximum of 49% on new deals. The rule applies worldwide and does not affect existing contracts.
No. Reports indicate the 51% non-Chinese requirement applies only to future contracts. Existing agreements continue under their current terms.
No. The instruction applies to NVIDIA’s cloud and data-centre clients globally, not only those operating in Asia.
European operators such as Nebius and Nscale that buy NVIDIA chips in bulk will need to ensure that any new capacity they contract meets the 51% non-Chinese threshold. Operators with significant Chinese end-user demand will have to adjust customer mix or deal structure on future contracts.
