The U.S. government is finalizing new export control rules for AI chips, planning to restrict Chinese AI companies from remotely using advanced chips such as Nvidia Blackwell through overseas data centers. The new rules could be announced as early as this month.
The new regulations bring Singapore, Malaysia, Thailand, Indonesia, Norway, Brazil, and South Africa under a stricter chip export licensing system. Data centers operated by non-US companies in these countries importing controlled AI chips such as Nvidia Blackwell and Rubin must apply for a license from the US and declare the end use. Data centers operated by US companies may be exempt if they prohibit remote use of the relevant chips by countries subject to arms embargoes, such as China and Russia, but they must report the identities of their customers and the scale of their computing power.
Previously, Nvidia required overseas cloud service providers to offer at least 51% of their services to non-Chinese customers, and Oracle canceled its approximately $7 billion computing power leasing contract with Tencent under pressure from the US government. Currently, Chinese cloud service providers and AI labs are accelerating their bookings of overseas Nvidia computing power.
