Making sense of crypto.
NBTCNEWS

Washington tightens AI chip rules: what risks for Nvidia?

The United States wants to require licenses for Chinese customers' access to advanced chips in certain overseas data centers, and to regulate deliveries to five countries in order to limit diversion. This extension of controls could delay Nvidia orders or steer customers toward Huawei, but its financial impact remains uncertain, and a license is not equivalent to a general ban.

By NBTC3 min read
Washington tightens AI chip rules: what risks for Nvidia?

Washington is toughening restrictions on AI chips in order to limit their use by Chinese companies through overseas data centers. This measure could reduce Nvidia's opportunities and encourage its Chinese customers to turn to competitors such as Huawei.

Washington toughens restrictions on AI chips

The United States is strengthening controls on artificial intelligence chips to limit their remote use by Chinese companies. A new rule targets overseas data centers that rent computing power to them, while more deliveries to five countries, including Singapore and Malaysia, will require licenses. For Nvidia, this tightening threatens an indirect opportunity in the Chinese market.

Since October 2022, Washington has restricted exports to China of the most powerful AI processors. Chinese developers could, however, still use this equipment without importing it, by renting servers installed abroad and accessing them over the Internet.

An excerpt from the document announcing tougher rules on AI chips.

This use of cloud services notably allowed them to train their models on Nvidia chips located in Southeast Asia. The restrictions governed the movement of hardware but imperfectly covered its remote use. The new rule aims to reduce this possibility of circumvention.

Licenses to govern access to chips

The Bureau of Industry and Security (BIS), responsible for US export controls, is planning two complementary measures: one concerns the customers of data centers, the other their location.

The first requires a license to provide advanced chips to centers that give access to significant computing power to companies based in China or in other restricted destinations. Companies whose parent firm is located in these territories are also affected.

The threshold set exceeds 12 million cumulative TPP, an indicator of processor performance. The capacities of several interconnected centers can be added together. Spreading machines across different sites is therefore not enough to escape control. The text also covers intermediaries who would train models on behalf of targeted customers.

The second obligation concerns Singapore, Malaysia, Thailand, Indonesia and Brazil. Deliveries of advanced chips to these countries will require a license, with controlled exemptions for American groups. Washington thus intends to reduce the risks of diversion to prohibited destinations.

For Nvidia, more constrained opportunities

For Nvidia, the stakes go beyond direct sales in China. The group also supplies foreign operators who rent its processors to Chinese developers. These developers therefore indirectly contribute to demand for American chips.

The new requirements could delay certain orders or prompt operators to review their customer base. Their financial impact on Nvidia remains to be assessed, however. A license requirement is not a general ban.

This tightening could also encourage Chinese developers to favor domestic manufacturers, including Huawei. Nvidia thus risks seeing part of this demand shift toward competing solutions.

The measure comes after Jensen Huang's departure, amid a governance crisis. Even before the announcement, Chinese companies were seeking to reserve more Nvidia capacity abroad, anticipating a tightening of US controls.

Sources: US Department of Commerce, Milk Road IA on X

share preview

Sharing is disabled in this private preview. Nothing has been sent.