U.S. export controls on AI chips may soon extend to customers at overseas data centers.
The U.S. Department of Commerce is preparing to tighten export licenses for advanced AI chips, potentially affecting seven countries: Malaysia, Singapore, Thailand, Brazil, Indonesia, Norway, and South Africa.
The rules are expected to focus on Nvidia Blackwell chips and higher-end models.
The most noteworthy aspect this time isn’t the addition of a few more countries.
It’s that the U.S. is trying to close a loophole: instead of buying restricted GPUs directly, Chinese companies rent data centers in places such as Singapore and Malaysia to access computing power remotely through cloud services.
The chips don’t enter China, but the computing power can be used across borders.
If the new rules take effect, overseas data centers may have to prove not only where their GPUs came from, but also who is ultimately using them.
This could affect GPU procurement, computing-power rentals, AI cloud services, and cross-border computing-power intermediaries in Southeast Asia.
It could especially affect overseas data centers that rely on Chinese customers while also buying high-end U.S. GPUs.
Until now, the business model has been: place GPUs overseas, then sell computing power to customers.
Going forward, the key question may be whether you can prove that the computing power is not ultimately being provided to restricted users.
Of course, these are still proposed rules; it’s not a blanket ban on sales to all seven countries.
The U.S. wants to restrict not just where chips go, but who ultimately gets access to advanced computing power.
The U.S. Department of Commerce is preparing to tighten export licenses for advanced AI chips, potentially affecting seven countries: Malaysia, Singapore, Thailand, Brazil, Indonesia, Norway, and South Africa.
The rules are expected to focus on Nvidia Blackwell chips and higher-end models.
The most noteworthy aspect this time isn’t the addition of a few more countries.
It’s that the U.S. is trying to close a loophole: instead of buying restricted GPUs directly, Chinese companies rent data centers in places such as Singapore and Malaysia to access computing power remotely through cloud services.
The chips don’t enter China, but the computing power can be used across borders.
If the new rules take effect, overseas data centers may have to prove not only where their GPUs came from, but also who is ultimately using them.
This could affect GPU procurement, computing-power rentals, AI cloud services, and cross-border computing-power intermediaries in Southeast Asia.
It could especially affect overseas data centers that rely on Chinese customers while also buying high-end U.S. GPUs.
Until now, the business model has been: place GPUs overseas, then sell computing power to customers.
Going forward, the key question may be whether you can prove that the computing power is not ultimately being provided to restricted users.
Of course, these are still proposed rules; it’s not a blanket ban on sales to all seven countries.
The U.S. wants to restrict not just where chips go, but who ultimately gets access to advanced computing power.