Nvidia Requires Cloud Providers to Sell at Least 51% of Services to Non-Chinese Customers
Nvidia has told cloud providers worldwide that at least 51% of their services must go to non-Chinese customers, the Financial Times reported, citing five people familiar with the matter. The requirement, communicated verbally by Nvidia sales and regional teams over recent months, amounts to a privatized enforcement of U.S. export controls by making chip access contingent on customer demographics.
The Rule and Its Scope
The 51% threshold applies to all new contracts for Nvidia's AI chips globally, not only in Asia, according to two of the FT's sources. Existing lease agreements are exempt. Nvidia did not put the rule in writing; it was conveyed through its sales organization, the sources said [1][2].
A company spokesperson told the FT that "data centers and clouds built with U.S.-made chips serve as the foundation enabling U.S. AI companies to compete in the global market, and that expanding domestic infrastructure worldwide has been a longstanding U.S. policy" [2].
Closing the Cloud Loophole
U.S. export controls have barred direct shipment of Nvidia's most advanced chips to China since 2022. Chinese firms have since rented capacity from overseas data centers — often in Singapore, Malaysia, and Japan — to train models on hardware they cannot buy. The 51% rule directly targets that workaround, according to the FT's evaluation of the move as an effort to make it even more difficult for China to catch up [2].
Nvidia has also cut its list of approved Asian customers by more than half in July 2026 and introduced hands-on compliance checks: physical inspections of data centers, contract verification, and interviews with end users [3]. Those steps follow U.S. Commerce Department guidance issued in May 2026 aimed at curbing diversion of advanced technology to China [3].
Pressure on Cloud Providers
The requirement puts neocloud operators — newer providers specializing in GPU rental — in a bind. "It is difficult to find enough non-Chinese customers outside the U.S.," a Singapore-based executive told the FT [1][2]. In Southeast Asia, Chinese tech companies have been major buyers of AI capacity. Providers must now rebalance their customer base or risk losing access to Nvidia hardware.
Nvidia's list of approved Asian customers shrank by more than 50% in July 2026, turning away some buyers even as demand for its chips remains high [3].
Chinese Access and Pricing
Tencent is reportedly securing AI chips through data centers in Japan and Australia, while Alibaba is obtaining hardware through overseas arrangements described as legally compliant, though the locations are not specified [3]. With the new cap, even those channels narrow.
Scarcity is already visible in pricing. A rack of B300 chips that sells for roughly $600,000 internationally was recently priced at the equivalent of $2.8 million in China, according to an insider [1].
What Remains Unclear
- Whether the 51% rule will be formalized in written contracts or remain verbal.
- How Nvidia will audit compliance at scale across hundreds of cloud providers.
- Whether the U.S. government directed the specific threshold or Nvidia set it independently to demonstrate compliance.
- The revenue impact on Nvidia if providers cannot meet the threshold and return chips.
Nvidia's move shifts enforcement from government licensing to commercial leverage. Whether it holds depends on whether cloud providers can find enough non-Chinese demand — and whether the verbal rule survives legal and commercial pushback.