For years, Washington has banned the sale of America's most powerful AI chips to China. On Monday, it moved to close the back door that let Chinese firms use those same chips anyway — by renting the computing power from data centres abroad.
The US Commerce Department's Bureau of Industry and Security (BIS) issued an interim final rule on 16 November that targets remote access to advanced AI chips and servers. According to the department, the rule is designed to make sure existing restrictions "cannot be circumvented through remote computing arrangements involving data centers in third countries."
What the new rule actually does
The rule works in two main ways. First, it adds a licensing requirement for data centres anywhere in the world that provide large amounts of computing power to companies headquartered in China, Macau or other restricted destinations. The threshold is more than 12 million aggregate TPP — a measure of total processing performance that reports estimate is roughly the output of around 750 of Nvidia's H100 chips.
Second, it requires export licences for shipments of advanced AI chips and servers to five countries the department says carry a higher-than-normal risk of diversion: Indonesia, Malaysia, Singapore, Thailand and Brazil. Data centres already serving covered customers will have 90 days from the rule's effective date to obtain a licence.
The rule also formally scraps the Biden administration's January 2025 AI Diffusion Rule, which sorted countries into tiers for chip access and was never enforced. The Trump administration had already said in May that it would discard that framework.
Why it matters now
The timing is hard to ignore. The rule is the administration's first policy response to the scandal surrounding former Nvidia chief executive Jensen Huang, who stepped down days ago amid reports of an alleged espionage case, according to NextGov. Bloomberg tied the move directly to the increased scrutiny of semiconductor exports that followed.
For the cloud computing industry, the stakes are real. Singapore in particular has grown into a major hub for AI data centres, and the new licensing requirements will force operators there — and in the other four named countries — to look much harder at who is renting their computing power. American-headquartered companies are exempt from the new destination licensing requirement, according to reports, although the remote-access limits would still apply to them.
The rule takes effect once it is published in the Federal Register, with comments due 60 days after publication. Until then, data centre operators and chipmakers will be reading the 132-page text closely to work out exactly where the new lines are drawn.
This is a developing story, and we will update this article as the rule is published and companies respond.
