Nvidia sets a 51% threshold for global cloud providers: Chinese clients can no longer account for the majority of computing power business.

Beating AI News Flash: The Financial Times, citing people familiar with the matter, reported that Nvidia is requiring its global data center and cloud computing partners to sell at least 51% of their computing power services to non-Chinese customers. The requirement has been communicated verbally by sales teams since this summer, with no written documentation at present, and applies only to future business; already-signed leasing contracts are unaffected.

Previously, the U.S. mainly restricted the physical export of advanced GPUs to China, but Chinese companies could still rent high-end computing power through overseas cloud service providers. For example, chips placed in data centers in Singapore, Malaysia, and other locations could be used remotely by Chinese companies over the network without needing to ship the GPUs back to China. This was also an important channel for some Chinese AI companies to obtain advanced computing power.

Nvidia already tightened restrictions once in July this year, cutting by more than half the list of customers in Asia approved to purchase AI chips, with a focus on reviewing cloud service providers that specifically rent out GPUs. Staff would even conduct on-site inspections of server rooms, verify contracts, and interview end users.

Now the restrictions have further extended to the end buyers of computing power. Some Southeast Asian cloud providers rely heavily on Chinese customers and will find it difficult in the short term to find enough non-Chinese buyers. Some cloud providers interviewed worry that if they cannot meet the requirements, they may have to scale back their data center businesses.