Nvidia Corp. has told cloud computing and data-center customers worldwide that at least 51% of their services must be sold to non-Chinese companies, a verbal directive that for the first time imposes a ceiling on how much revenue those partners can derive from Chinese buyers1,3.
The instruction was delivered through Nvidia's sales organization and regional teams in recent months but was not put in writing, according to five people with knowledge of the matter cited by the Financial Times4. No comparable restriction had existed before. The rule applies globally, not only to data centers in Asia, and covers future agreements only; existing rental contracts are unaffected.
How the directive works
Nvidia's cloud clients must now ensure that a majority of their compute capacity goes to non-Chinese firms, effectively requiring any partner whose Chinese business exceeds 50% to restructure its customer mix going forward. The requirement adds a private compliance layer on top of U.S. government restrictions that already limit the sale and export of advanced AI chips to China.
Many Chinese technology companies have relied on offshore cloud rentals to access high-performance computing for AI model training, circumventing direct chip-export controls. Washington has been working to shut down both the intermediaries that have fueled a black market for Nvidia chips inside China and the overseas rental route.
Nvidia's response and Beijing's reaction
Nvidia neither confirmed nor denied the report to AFP. In a statement, the company said that "datacenters and clouds built on American chips are the foundation that American model and application makers need to compete worldwide". It added: "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".
Chinese foreign ministry spokeswoman Mao Ning said she was not aware of the report.
ANALYSIS The 51% threshold turns Nvidia's dominant market position in AI accelerators into a de facto compliance mechanism, extending Washington's chip-control strategy beyond government export rules and into private commercial terms. Data-center operators that have built their businesses primarily around Chinese demand now face a choice between diversifying their customer base or losing access to future Nvidia hardware.
The directive lands as the shortage of advanced computing power at Chinese technology companies is, by the Financial Times' account, deepening. Marvell Technology's recent agreement to acquire Lumentum Holdings in an approximately $115 billion deal, in which Nvidia invested $2 billion in each company, underscores how tightly the AI infrastructure supply chain is consolidating around a small number of U.S.-linked players[2].