Nvidia restricts Chinese companies' access to AI capabilities via overseas clouds
According to the Financial Times, Nvidia has informed cloud partners that at least 51% of their services must be allocated to companies not linked to China. Existing contracts are not affected by the new conditions.
Nvidia has urged partners providing cloud computing services on its accelerators to reduce their dependence on Chinese clients. This is reported by the publication, citing five sources familiar with the situation.
According to the publication, in recent months, the company has notified cloud providers that at least 51% of their services must be sold to companies not linked to China. The conditions were communicated verbally through sales department employees and regional Nvidia divisions. The company has not presented a public document with detailed rules.
As FT writes, the change applies to future contracts and extends to partners worldwide, not just Asian data centers. It does not affect existing agreements.
Renting overseas computing power allows Chinese developers to work with advanced accelerators that cannot be directly supplied to China due to U.S. export restrictions. Such resources are used, in particular, for training and deploying artificial intelligence models.
The new condition may complicate this approach, though it does not imply a complete rejection of Chinese clients. Providers with a high proportion of such customers will need to expand their client base.
The change may be particularly sensitive for data center operators in Southeast Asia. A representative of one Singaporean company told FT that finding enough clients not linked to China is challenging, and strict enforcement of the condition could harm their business.
In a comment, Nvidia emphasized the importance of infrastructure based on U.S. chips for the global competitiveness of American developers. However, the published statement does not clarify how the company will verify the share of clients or enforce the new conditions.
