Nvidia sets a 51% threshold for global cloud providers, barring Chinese clients from dominating the bulk of its computing power business.

动察 Beating AI Flash News: The Financial Times cites anonymous sources saying Nvidia is requiring global data center and cloud computing partners to sell at least 51% of their computing power services to non-Chinese clients. The requirement was communicated verbally to sales teams starting this summer, with no written documentation currently in place, and applies only to future business—existing signed leases remain unaffected. Previously, the U.S. restricted the export of advanced GPUs to China, but Chinese firms could still access high-end computing power via overseas cloud providers. For instance, chips are deployed in data centers in Singapore, Malaysia and other regions, with Chinese companies accessing them remotely via the internet, no need to ship GPUs back to China. This has been a key channel for some Chinese AI enterprises to obtain advanced computing power. Nvidia already tightened rules once in July this year, cutting the list of Asian customers approved to purchase AI chips by half, with a focus on reviewing cloud service providers that specialize in renting out GPUs. Staff even conduct on-site inspections of data centers, verify contracts and interview end users. Now the restrictions have expanded to the end buyers of computing power. Some Southeast Asian cloud providers are heavily dependent on Chinese clients, making it difficult to secure enough non-Chinese buyers in the short term. Some surveyed cloud providers worry they may have to scale back their data center operations if they fail to meet the requirement.