Nvidia has reportedly told its cloud and data-centre partners to keep Chinese customers below half of their business under new contracts, drawing a quiet line through the global market for its most powerful chips. The reported rule would require at least 51% of services to go to companies outside China, leaving Chinese firms at no more than 49%. It would apply worldwide, not only in Asia, and it would cover future agreements while leaving existing contracts untouched.
What the reported rule would cover
The instruction was passed verbally through Nvidia sales teams and regional offices recently, rather than set out in a written document, according to the Financial Times report. That detail matters because partners must interpret how strictly the balance will be policed, and how it will be measured as their customer base changes. No penalty or audit mechanism has been described.
The rule would reach beyond the companies that buy Nvidia chips. Cloud providers and data-centre operators that rent out computing capacity would have to watch the mix of their customers, even when hardware stays abroad. That indirect route has become central to the export-control debate, because renting capacity abroad can offer access without importing a restricted chip.
What Nvidia did and did not say
Nvidia neither confirmed nor denied the report when asked about it. In a statement, the company said data centres and clouds built on American chips are the foundation that American model and application makers need to compete worldwide. It added that longstanding United States policy encourages every cloud to build on American infrastructure and to win business everywhere it can.
China’s foreign ministry said its spokeswoman was not aware of the reported restrictions. Beijing said it hopes all parties can provide a fair, transparent and non-discriminatory business environment for companies from all countries, including Chinese companies. That response stops short of a formal protest, but it signals that Beijing is watching how partners apply any new balance.
Why partners, not just buyers, are the pressure point
Direct chip sales to China have already faced tightening controls, site visits, contract reviews and tougher compliance checks. Reports say Nvidia more than halved its list of approved Asian customers in July, after applying stricter checks. A customer-mix rule would extend that scrutiny one step further, from who buys the hardware to who rents the computing power built on it.
The distinction between a hard rule and a nudge is unresolved. A follow-up report said Nvidia’s instructions were intended to encourage a broader customer mix, and were not a mandatory restriction. Until partners or Nvidia clarify that point, the market is left with two readings: a firm cap on business, or a strong steer that still leaves room for exceptions.
That ambiguity sits against a wider pattern of diversion cases and investigations. Researchers have previously reported export-controlled chips found on Chinese networks, while United States scrutiny has also touched an alleged spy plot aimed at Nvidia’s chief. Neither case proves the new rule, but both help explain why customer identity now matters as much as chip destination.
Why It Matters
For Chinese AI companies, the reported rule could narrow a remaining route to computing power without an import. If partners must reserve a majority of new capacity for non-Chinese customers, Chinese firms may face higher prices, longer waits or fewer choices renting capacity abroad. That would not stop chip development, but it could slow work that depends on American systems.
For operators, it would turn customer mix into a compliance question, not a sales target. Operators that built growth on a customer base would need to track nationality and capacity allocation in a way not done before. The cost of that tracking, and the risk of getting it wrong, could reshape which partners win the next AI infrastructure work.
Our read
The most important word in this story is reportedly, and the second is verbally. A written rule creates a clear test; a verbal steer creates a signal that everyone must interpret. Nvidia can push partners toward diversification without publishing a policy that invites a legal challenge, a diplomatic row or a compliance test it cannot yet meet.
The timing matters. Nvidia’s next results will test whether investors see shifting partner demand. A $115 billion optical deal already shows how much capital is chasing data-centre capacity. Against that backdrop, a customer-mix steer is less about one quarter’s revenue and more about who gets to use capacity being built.
Chinese customers are unlikely to disappear from these clouds overnight, because existing contracts are untouched and enforcement is open. The real test will come with the next contract cycle, when a partner must choose between a familiar Chinese customer and the balance Nvidia reportedly wants. That choice, repeated across operators, is where the rule would either bite or fade.
Background
Nvidia has become the central supplier of the chips that train and run large AI models, and Washington has steadily tightened the rules on advanced processors reaching China over national-security concerns. The reported customer-mix guidance, if confirmed, would mark a shift from controlling shipments to shaping the customer base of the clouds that rent those chips out.
Source: AFP report on the FT claim and Invezz summary of the reports
0 Comments