US restricts remote access to advanced AI chips to limit China

Covered by 14 sources
Read onNew York Times· 6 min
The US Commerce Department has imposed new export controls requiring licenses for overseas data centers that enable remote access to advanced American AI chips by Chinese and other restricted entities. This rule aims to close loopholes that allowed China to use US-origin AI hardware remotely, extending restrictions to Southeast Asia and tightening enforcement after a related Nvidia scandal.

What happened

The US Commerce Department has issued new export controls that require licenses for overseas data centers providing remote access to advanced US AI chips, specifically targeting Chinese companies. The rule closes a loophole that previously allowed restricted entities to use high-performance AI chips via third-party data centers outside the US. The new controls also extend to several Southeast Asian countries, including Singapore and Malaysia, following a recent scandal involving suspected chip diversion to China. Data centers already serving affected companies have 90 days from the rule's effective date to comply by obtaining the necessary licenses.

Key details

  • The rule targets data centers exceeding 12 million aggregate TPP (tera operations per second).
  • Licenses are now required for data centers in Indonesia, Malaysia, Singapore, Thailand, and Brazil if serving restricted entities.
  • The rule applies globally to any data center granting remote access to Chinese-headquartered companies.
  • The new export controls are effective immediately upon publication in the Federal Register, with a 90-day compliance window for existing arrangements.
  • Congress is considering additional laws to codify and expand these export controls.

Why it matters

US tech companies, overseas data center operators, and Chinese firms are directly affected. Data centers must now obtain licenses to provide advanced AI chip access to Chinese companies, increasing compliance costs and operational complexity. Chinese companies face new barriers to accessing US AI hardware, even indirectly, which may limit their AI development capabilities.

Also covered by 13 sources

More on this story


More in Chips & Compute