On 16 November 2026 the US Bureau of Industry and Security (BIS) released an interim final rule on remote access to advanced AI chips. BIS says the rule was written by and at the direction of the Treasury and the National Security Council. Its purpose is to stop companies in countries of concern from renting through the cloud the computing power they are not allowed to buy.
What the rule does
According to the rule text filed for public inspection, a licence will be needed to export, re-export or transfer advanced AI chips to an overseas data centre that provides more than 12,000,000 "aggregate TPP" to companies headquartered in Country Group D:5 (which includes China) or Macau, or whose ultimate parent is headquartered there. The test can be met at a single site or across linked data centres. Data centres, foreign foundries and chip exporters must carry out "robust due diligence" on customers, including indirect ownership and control.
Data centres that already supply significant compute to such companies must obtain a licence within 90 days of the rule taking effect or face future export restrictions. Computing used only for "commercial recommender systems", such as those run by online retailers and social media companies, is carved out, subject to reporting.
The rule also:
- requires licences for advanced chips going to Indonesia, Malaysia, Singapore, Thailand and Brazil, which BIS describes as presenting a higher-than-normal risk of diversion. Chips destined for companies headquartered in the US are exempt if reporting conditions are met;
- tightens the presumptions that chip foundries and packagers face when making advanced-node logic chips;
- formally rescinds the Biden-era AI Diffusion Rule.
BIS warns that countries tolerating evasion could face import restrictions, new licence requirements, a downgraded export-control status or Entity List designations.
Timing
The rule takes effect when it is published in the Federal Register, scheduled for 19 November. Comments are due 60 days after publication. XenoSpectrum calculates that, if publication goes ahead as planned, existing operators would have until 17 February 2027 to obtain licences. It also estimates that the 12 million TPP threshold equals about 758 Nvidia H100 chips.
Ars Technica links the rule to the scandal around Megaspeed, a Singapore data centre company, and Jensen Huang's resignation as Nvidia chief executive.
Why it matters financially
The rule reaches data centre operators anywhere in the world, not just chip sellers. Operators in South-East Asia and Brazil face new licence requirements for chips they had been buying without one. Any facility serving Chinese-headquartered customers above the threshold must license that business within 90 days or lose it. For chipmakers and cloud providers, the due-diligence and customer-screening duties add compliance costs. They may also cut off a source of revenue that, according to The New York Times as reported by Ars Technica, was legal until now. No official estimate of the revenue at stake has been published.