 |
| AI-generated illustration of a modern data centre hall with glowing server racks and a world map split between two sides, suggesting a customer-mix limit. |
Nvidia, the company whose chips power much of the world's artificial intelligence boom, is reportedly putting a new number on its cloud partners: at least 51% of their services must go to customers outside China. The report, published by the Financial Times on 3 November 2026 and picked up the same day by Bloomberg, Reuters and Agence France-Presse, says Nvidia has told data centre and cloud partners to scale back their exposure to Chinese customers under future contracts.
Read that sentence carefully, because the most important word in it is "reportedly". Nvidia has neither confirmed nor denied the report. Reuters said it could not immediately verify it. The guidance was reportedly delivered verbally, through sales and regional teams, rather than in a written notice. That does not make the story unimportant — the Financial Times cited five people with knowledge of the matter, and three major newsrooms treated it as a top story on the same day — but it does mean every detail below should be read as reported, not as an official Nvidia policy announcement, until Nvidia itself says more.
As an Amazon Associate, we earn from qualifying purchases.
Prices and savings can change. Check the retailer for the current price.
This article explains what the reported rule would actually require, who it would affect, why it is being reported now, and what it does and does not mean for ordinary buyers, cloud customers and investors. Where a figure comes from a single report or a market projection, that is stated plainly.
What is reportedly changing
In plain terms, the reported rule is a customer-mix test for Nvidia's cloud and data centre partners. Under future contracts, those partners would need to ensure that at least 51% of the services they sell go to non-Chinese companies. Put the other way round, Chinese companies could account for no more than 49% of a partner's services under the new arrangements.
Three limits on that claim matter as much as the headline number.
First, the rule reportedly applies only to future contracts, not to existing agreements. A partner with an existing contract is not, on the reporting so far, being asked to rewrite it or to drop current customers overnight. The change would bite as contracts renew or as new deals are signed.
Second, the rule reportedly applies to Nvidia's cloud computing clients and data centre partners — the companies that buy Nvidia chips in bulk and rent out computing power — rather than to every company that buys a single graphics card. These are sometimes called neoclouds and other data centre operators. They sit between Nvidia and the end customer: they buy the hardware, build the data centre capacity, and sell AI computing as a service.
Third, the guidance was reportedly communicated verbally, not in writing. That detail, repeated across the coverage, is worth noting because it affects how verifiable the rule is. A verbal instruction leaves no document for a partner or a journalist to quote directly, and it leaves room for differences in how individual partners understood it. It is also, notably, not the same as a government export control. This is reported as Nvidia's own instruction to its partners, not as a new law.
Nvidia's public response, as quoted to Agence France-Presse, did not confirm or deny the report. The company said that "datacentres and clouds built on American chips are the foundation that American model and application makers need to compete worldwide." In separate coverage, a Nvidia spokesman was quoted saying that "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." Those statements explain Nvidia's general position on American AI infrastructure. They are not a confirmation of a 51% threshold.
China's foreign ministry spokeswoman Mao Ning, asked about the report, said she was not aware of it. Alibaba, Tencent and ByteDance — three of the largest Chinese buyers of AI computing — have not publicly addressed the specific 49% guidance in the coverage so far.
Why a customer-mix rule, and why now
To understand why Nvidia might impose a rule like this, it helps to step back from the 51% figure and look at the position Nvidia is in. It is the world's most valuable company, built on the global rush to construct data centres packed with specialist chips that run AI systems. At the same time, it operates inside one of the sharpest technology rivalries of the decade: the contest between the United States and China for dominance in AI.
The United States has restricted exports of advanced AI semiconductors to China, concerned that they could accelerate China's AI development and aid its military. Those controls have already reshaped Nvidia's business in China. According to market research firm IDC, cited by Reuters, Nvidia's share of China's data centre AI chip market fell to 34% in the first half of 2026, down sharply as shipments fell. Huawei, by contrast, has gained ground with its own AI accelerators, narrowing the gap considerably.
Nvidia has also, according to the Financial Times reporting, already tightened its own controls this year: in July it cut its list of approved Asian buyers by more than half, imposing stricter compliance checks aimed at preventing its products from reaching China through intermediaries. The reported 51% rule would be a further step in the same direction, but aimed at a different point in the chain. Instead of only controlling who can buy chips directly, it would control who the buyers of those chips can sell computing services to.
That distinction matters. A data centre partner that buys Nvidia chips and rents out computing power could, in principle, serve customers anywhere. If most of that rented capacity ends up being used by Chinese companies, then the practical effect is similar to selling the chips to China directly, even if the formal buyer is elsewhere. A customer-mix rule is a way of closing that gap without banning any single transaction outright.
There is also a commercial logic. Coverage of the report notes that China and other restricted markets made up somewhere around 20% to 25% of Nvidia's data centre revenue before Washington tightened the rules. Nvidia therefore has a large commercial interest in China, and its chief executive Jensen Huang has repeatedly criticised export controls, arguing that they hand a large market to competitors such as Huawei. A partner-level customer-mix rule, rather than a blanket ban, would let Nvidia keep selling to partners while limiting how much of the resulting capacity flows to Chinese customers. Whether that balance satisfies either Washington or Beijing is a separate question, and one that the reporting so far does not answer.
Who would feel it, and who would not
The most directly affected group would be Nvidia's cloud and data centre partners, especially neoclouds whose business model is built on renting out Nvidia-powered computing. For a partner whose customer base is heavily Chinese, a 51% non-Chinese requirement would mean reshaping its sales pipeline: winning more customers outside China, turning away some Chinese demand on new contracts, or both. For a partner whose customers are already mostly outside China, the rule might change little in practice.
Chinese AI companies would feel it indirectly, as reduced access to rented Nvidia-powered capacity on new contracts. That does not mean their existing capacity disappears — existing agreements are reportedly unaffected — but it would narrow one route to additional computing power at a time when domestic alternatives are still scaling up. Market research firm TrendForce has projected that Nvidia's share of high-end AI chips used by Chinese cloud companies could fall from around 80% to somewhere between 50% and 60%. Treat that as a projection, not a measured fact: it is a forecast about market direction, and forecasts about a fast-moving, policy-driven market are frequently revised.
Ordinary consumers and small businesses are unlikely to feel a direct effect in the short term. The reported rule is about wholesale data centre services, not about retail graphics cards or consumer software. If you buy a gaming graphics card, use a chatbot, or run a small website, nothing in the reporting suggests your service changes tomorrow. The indirect effects, if any, would arrive slowly, through the price and availability of AI computing in the cloud — and even there, the market has many moving parts, including rival chipmakers, domestic Chinese chips, and the sheer pace of new data centre construction.
Investors, meanwhile, reacted calmly on the first day of the story. Nvidia shares advanced modestly in overnight trading after a four-session decline, according to market coverage, though a single day's share movement proves very little about a story whose details remain unconfirmed. The more useful investor signal in the coming weeks will be whether Nvidia confirms, denies, or clarifies the report, and whether any partner or customer speaks publicly about it.
What is confirmed, what is reported, and what is still unknown
Because this story rests on anonymous sourcing and an unconfirmed report, it is worth separating the layers clearly.
Confirmed, in the sense of being on the public record: Nvidia is the world's most valuable company on the back of AI data centre demand; the United States restricts advanced chip exports to China; IDC data reported by Reuters shows Nvidia's China data centre AI chip share fell to 34% in the first half of 2026; and Nvidia's public statements emphasise American-built data centres and clouds as the foundation for American AI companies to compete worldwide.
Reported, but not confirmed by Nvidia: the 51% non-Chinese services threshold itself; that it was communicated verbally through sales and regional teams; that it applies to future contracts only; that it applies to Nvidia's cloud and data centre partners globally, not only in Asia; and that Nvidia cut its approved Asian buyer list by more than half in July.
Unknown, on the reporting so far: whether the threshold is a firm requirement or a guideline; how compliance would be measured — by revenue, by computing capacity, by number of customers, or by some other metric; what happens to a partner that falls short; whether there are exceptions for particular countries, customers, or use cases; and whether governments in Washington or Beijing have been involved in shaping the guidance. None of those questions can be answered from the current reporting, and any article that pretends otherwise would be guessing.
That uncertainty is not a reason to ignore the story. It is a reason to read it for what it is: a well-sourced report from a major financial newspaper, corroborated as a news story by other major newsrooms, about a commercially and geopolitically significant shift — but not yet an official policy. If Nvidia confirms it, the story becomes a concrete rule to analyse. If Nvidia denies it, the story becomes a lesson in how quickly a plausible-sounding report can travel. Either way, the underlying tension it describes — between selling AI infrastructure globally and limiting its flow to China — is real and will not go away.
What to watch next
Five things will clarify this story in the days and weeks ahead.
First, Nvidia's own response. A confirmation, denial, or clarification would replace reporting with fact, and would likely include the measurement and enforcement details that are currently missing.
Second, any comment from the partners themselves. Data centre operators and neoclouds are the ones who would have to live with the rule, and their silence so far is itself informative: either they have not been briefed in the terms reported, or they are choosing not to discuss a commercially sensitive instruction publicly.
Third, any comment from major Chinese customers. Alibaba, Tencent and ByteDance have not addressed the specific guidance in the coverage so far. Their response, if any, would indicate whether they see this as a material constraint or as a manageable adjustment.
Fourth, the policy context. Watch for any statement from Washington or Beijing that touches on cloud services, chip exports, or AI infrastructure. The reported rule sits at the intersection of company policy and government policy, and official statements from either side could change its meaning.
Fifth, the market data. IDC's half-year figures, TrendForce's projections, and Nvidia's own future results will show, over quarters rather than days, whether Chinese access to high-end AI computing is actually narrowing, and whether domestic Chinese chips are filling the gap.
Checklist: how to read this story without being misled
- Do not treat the 51% figure as an official Nvidia policy yet. It is a reported threshold from the Financial Times, not confirmed by Nvidia.
- Do not assume existing contracts change. The reporting says future contracts only.
- Do not assume this is a consumer story. It concerns wholesale cloud and data centre partners, not retail graphics cards.
- Do not confuse a market projection with a measurement. TrendForce's 50% to 60% range is a forecast, and IDC's 34% share is a half-year market measurement reported by Reuters — they are different kinds of evidence.
- Do not read one day's share price movement as a verdict on the story.
- Do watch for Nvidia's own statement, partner comments, and official policy signals — those are what would turn a report into a rule.
Bottom line
Nvidia is reportedly telling its cloud and data centre partners to keep Chinese customers to no more than 49% of their services on future contracts, with at least 51% going to non-Chinese companies. The report comes from the Financial Times, was picked up the same day by Bloomberg, Reuters and AFP, and fits a wider pattern of tightening controls around AI chips and China — but Nvidia has neither confirmed nor denied it, and Reuters said it could not immediately verify it.
For the AI industry, the story matters because it shows how the US-China chip contest is moving beyond direct export bans into the business models of the companies that rent out computing power. For everyone else, the practical message is simpler: nothing changes for existing contracts or for ordinary consumers on the basis of this report alone, and the details that would matter most — how the rule is measured, enforced, and applied — remain unknown until Nvidia speaks. Treat the 51% figure as a serious, well-sourced report to watch, not as a settled fact, and judge it by what Nvidia and its partners say next.
Reference sources
- Financial Times report on Nvidia's data centre partners, via Agence France-Presse: https://www.malaymail.com/news/money/2026/11/03/nvidia-tells-data-centre-partners-to-curb-business-with-chinese-clients-ft-reports/241835
- Reuters report on Nvidia's data centre clients: https://www.thestandard.com.hk/innovation/article/347343/Nvidia-orders-data-centre-clients-to-reduce-China-exposure-FT-reports
- Bloomberg-cited coverage of the 49% cap: https://startupfortune.com/nvidia-tells-cloud-clients-to-cap-china-business-at-49-of-their-services/
- Agence France-Presse syndication of the report: https://www.freemalaysiatoday.com/category/business/2026/11/03/nvidia-orders-data-center-clients-to-reduce-china-exposure
- Background on Nvidia's China market share and the wider chip contest: https://dig.watch/updates/nvidia-clients-to-scale-back-chinese-relations
Leave a public comment