The world’s largest AI chip designer is facing criticism over whether it can realistically police its supply chain to stop foreign labs from utilizing high-end hardware. Despite public efforts to curb Chinese access to its technology, Nvidia continues to list Megaspeed—a company currently under federal investigation—as a cloud partner on its website. Critics also point to Nvidia’s relationship with Aivres, a U.S. subsidiary of the entity-listed Chinese firm Inspur, which has reportedly supplied chips to companies that provide server access to Chinese state-owned enterprises.
These concerns follow reports that Nvidia verbally instructed its data center and cloud clients since the summer to ensure no more than 49% of their compute services are sold to Chinese customers. The mandate, which applies to new contracts, is intended to close a remote access loophole that allows Chinese firms to rent GPU capacity when they cannot legally import chips under U.S. export controls. To enforce the rule, Nvidia has requested that partners disclose the identities of their end users, a move that comes as the market for AI hardware in China has surged; B300 racks have sold for $2.54M in China last month, compared to roughly $600,000 in international markets.