HomePoliticsBIS Issues a Rule on Remote Access to Advanced AI Chips

BIS Issues a Rule on Remote Access to Advanced AI Chips

The US Commerce Department’s Bureau of Industry and Security has issued an interim final rule intended to stop firms in countries of concern from using remote computing arrangements in third countries to reach advanced American chips. The 16 November press release said the rule was “written by and at the direction of the Department of Treasury and the National Security Council,” and is designed to stop existing restrictions being circumvented through data centres outside countries already subject to US export limits.

The interim final rule, RIN 0694-AK50, creates a licence requirement for exports, re-exports or transfers of advanced “super intelligence” items when there is knowledge that they are destined for use in data centres, “wherever located,” that will provide direct or indirect access to “a significant volume of computing power” to an entity headquartered, or whose ultimate parent is headquartered, in Macau or Country Group D:5. Licence applications will be reviewed on a case-by-case basis. The operative text sets the computing-power threshold at more than 12,000,000 “aggregate TPP,” measured at one data centre or through “linked data centers,” including digitally linked infrastructure. TechCrunch, in a 16 November article by Ram Iyer, compared that figure with the peak computing power of about 750 Nvidia H100 or H200 chips. That comparison is TechCrunch’s, not a figure in the rule.

Existing providers that were already supplying such remote access to companies headquartered in Macau or Country Group D:5 must obtain a licence within 90 days of the effective date, the rule says, or face future export restrictions. The DATES section of the display copy leaves the effective date as the date of publication in the Federal Register, which had not yet been filled in on the version published with the announcement. The licence requirement does not apply to commercial recommender systems, such as those used by online retailers or social media companies.

New country licences, and a formal rescission

The rule also adds a licence requirement for exports, re-exports or transfers of advanced semiconductors to Indonesia, Malaysia, Singapore, Thailand or Brazil, countries which, it says, “present a higher-than-normal risk of diversion.” That requirement will not apply to semiconductors destined for companies headquartered in, or whose ultimate parent is headquartered in, the United States, provided certain reporting requirements are met. Commerce said it would monitor countries that tolerate evasion and could consider further measures, including new licence requirements or Entity List designations.

The same instrument formally rescinds the regulations that implemented the Biden-era AI Diffusion Rule, the “Framework for Artificial Intelligence Diffusion” published on 15 January 2025. The IFR says that rule has never been enforced, and that the changes formalise a rescission BIS announced in May 2025, including the revocation of the model-weights controls published in that earlier framework.

The announcement comes in the same week as Nvidia’s board opened an independent inquiry after intelligence reports and Jensen Huang resigned as chief executive. Those are neighbouring TSN stories, not a cause stated in the BIS rule. A separate recent TSN account of Taiwan’s plan to add AI servers and advanced chips to its own export-control list sits on the same wider controls file, but is a different government’s proposal.

A licence rule is not a ban on remote computing

The official record is an interim final rule, not a finding that any named data centre has already diverted chips, and not a blanket ban on foreign cloud computing. A licence reviewed case by case can be granted or refused. The 12,000,000 aggregate TPP figure is a threshold in the rule, not a measurement of any operator named in the announcement, and TechCrunch’s 750-chip comparison is a journalistic illustration of that threshold, not a BIS count.

The 90-day clock for existing providers runs from the effective date, which the display copy still ties to publication in the Federal Register. Until that date is fixed, the rule should not be read as if the 90 days have already started. The country-level licences for Indonesia, Malaysia, Singapore, Thailand and Brazil are likewise a new requirement with a stated exception for US-headquartered companies, not a finding that diversion has already occurred in those markets.

The rescission of the AI Diffusion Rule is a formal legal change in this IFR, including the revocation of its model-weights controls. It is not, by itself, a decision about how any future model-weight transfer will be treated under other authorities. The next documented steps to watch are the Federal Register publication date, which starts the 90-day clock, and the first licence decisions under the new remote-access control.

Sources

Related stories

Follow @tsnmediahub on X for breaking AI news as it happens.

Follow on X

Share this story

Latest stories

More in this category

Latest stories

Explore the TSN Atlas: Every topic we cover, mapped. Click a bubble to dive in.