Nvidia Squeezes China’s Backdoor Access To US AI Chips With This Unwritten Rule

China’s back door to America’s best artificial intelligence chips just got narrower — but it is not shut.

Nvidia is capping Chinese customers at 49 percent of sales for companies that rent out its chips, the Financial Times reported Monday. The company is passing along the rule to clients with an unwritten, verbal agreement, which could restrict Chinese firms’ ability to utilize chip power from American data centers to train their own AI models.

“America cannot win the SI race unless America’s SI infrastructure wins worldwide — if the world’s clouds rely on foreign-made chips, the model and application layers will be foreign-built as well,” an Nvidia spokesperson said in a statement to the Daily Caller News Foundation.

“SI” refers to President Donald Trump’s push to rebrand AI as “Super Intelligence,” which was made official on Sept. 29 with the signing of the “Inaugurating the Era of Super Intelligence” executive order.

The Chinese Embassy in Washington, D.C. did not immediately respond to the DCNF’s request for comment. The Commerce Department did not provide a comment on the record by time of publication.

American policy has long encouraged data centers around the world to run on U.S. chips, which American AI companies rely on to compete globally, the spokesperson told the DCNF. The spokesperson did not confirm or deny the verbal requirement.

While the rule applies to new contracts with Nvidia clients worldwide, the company separately created a “white list” to cut back its Asian clients by more than 50 percent, according to the Financial Times.

Nvidia began giving these verbal instructions during the summer, as an effort to get ahead of critiques on how the company handles China, individuals with knowledge on the subject told the outlet.

However, it is “difficult to find enough non-Chinese clients outside the US,” so enforcing the rule runs the risk of harming business, a Singapore-based group told the Financial Times.