Nvidia’s limiting Chinese access to its chips

Nvidia is making it harder for Chinese companies to use its chips. The chipmaker has told its cloud computing clients that at least 51% of their services must go to non-Chinese companies, the Financial Times reports, citing five people it says have knowledge of the matter. There were no such restrictions before.

Why it matters: US export controls bar some of Nvidia’s most advanced chips, like the B300, from being sold in China. Many Chinese tech companies have instead relied on renting computing power from data centers abroad for advanced AI work, including training models.

Why now? Nvidia adopted the rule over the summer to head off criticism from US lawmakers over its approach to China and its sales teams have verbally passed the instructions to customers, two people told the FT. The rule only covers new contracts, not existing leases.

The squeeze: Data centers outside the US are scrambling to find non-Chinese clients and could be forced to shrink if they can’t. “If strictly implemented, it will definitely hurt business,” one Singapore-based group told The Financial Times.

Pricey in China: A rack of Nvidia’s B300 chips that sells for about USD 600k internationally fetched about USD 2.8 mn in China last month, the FT says. The US prohibits selling those chips directly to China, but there’s a thriving black market for them.

The Gulf angle: The rule applies to Nvidia’s clients worldwide, not just data centers in Asia, two of the people told the FT. In the Gulf, the US gave a handful of UAE firms license-free access to Nvidia’s top chips in July, and Saudi Arabia’s Humain has approval to buy 35k of them, as we’ve reported.