On November 16, 2026, the Commerce Department's Bureau of Industry and Security (BIS) released an interim final rule titled "Addressing Remote Access to Advanced Semiconductors." It was published in the Federal Register on November 19 at 91 FR 74970. The rule tries to close a gap that earlier controls left open. Companies headquartered in China and other Country Group D:5 destinations could not buy top-end accelerators, but they could rent the compute from data centers in third countries.
This matters to builders because the rule reaches past chip vendors. It covers anyone who runs or depends on GPU capacity outside the US: neoclouds, colocation operators and AI labs that rent capacity in Southeast Asia, the Gulf or Brazil. The rule also ends the never-enforced AI Diffusion framework, including its controls on model weights.
What we checked
- Can the interim final rule's operative provisions and effective dates be confirmed directly from its Federal Register text or an authoritative BIS publication? confirmed. The published rule (FR Doc. 2026-23529, RIN 0694-AK50) is effective November 19, 2026, takes comments until January 19, 2027, and contains the data center threshold, the five-country license requirement, a December 21, 2026 savings-clause cutoff and the rescission of the AI Diffusion Rule. The BIS press release is dated November 16, 2026. [1][2][4]
- Does the rule make providing remote access (cloud/IaaS) itself an export? refuted. No. The rule works through an end-use license requirement on chips going to data centers that serve covered companies. Baker McKenzie notes that it does not redefine export, reexport or transfer to include pure remote access. [5][1]
What we found
- The rule took effect on publication, November 19, 2026, and BIS takes comments until January 19, 2027. Items already en route under actual orders can still ship under their prior eligibility only if the export, reexport or in-country transfer is completed by December 21, 2026. [1][2]
- Separately, any export, reexport or in-country transfer of advanced semiconductors to Indonesia, Malaysia, Singapore, Thailand or Brazil now needs a license. The Secretary of Commerce can add countries in consultation with the Secretary of State or the national security adviser. Gibson Dunn reports an exclusion for US-headquartered recipients that is conditioned on monthly reporting. We did not confirm the reporting mechanics in the rule text ourselves. [1][3]
- The rule formally rescinds the regulations that implemented the Biden-era AI Diffusion Rule and removes ECCN 4E091, the control on model weights. Lab builders no longer face an export classification for closed model weights under this framework. The rule's text also replaces "Artificial Intelligence" with "Super Intelligence (SI)" throughout. [1]
- BIS's own press release, dated November 16, says the rule was written by and at the direction of Treasury and the National Security Council. That is unusual wording for a Commerce rule. The Federal Register text we read names no Treasury role. [4]
What it means
If you operate or resell GPU capacity outside the US, check now whether any customer is headquartered in China, Macau or another D:5 country, or has its ultimate parent there. Then add up aggregate TPP across linked facilities. Above 12,000,000, existing arrangements need a license within 90 days of November 19. New hardware deliveries to such sites need a license now. Expect your chip and server suppliers to demand KYC and ownership data before they ship. If you build in Indonesia, Malaysia, Singapore, Thailand or Brazil, every advanced chip import now goes through licensing, apart from the exclusion for US-headquartered recipients that Gibson Dunn describes.
For AI labs, two points matter. Renting compute is still not itself an export, so the burden falls on the data center and the chip supply chain rather than on API access. And the removal of the model-weights control lifts a classification worry for distributing closed weights abroad. Because this is an interim final rule written under unusual interagency authorship, plan for amendments after January 19, 2027.
Competing readings
- The rule is in force as published and materially raises compliance burdens for non-US data centers and chip supply chains that serve D:5-headquartered companies (likely, about 70%). What would tell it apart: The rule took effect on publication, with a 90-day licensing window for existing arrangements and a hard threshold [1]; Five new license destinations widen the controls beyond China itself [1].
- Interagency friction leads to substantial amendment, delay or narrowing after the comment period (unlikely, about 22%). What would tell it apart: BIS attributes authorship to Treasury and the NSC, and the interim final rule invites comments until January 2027 [4][1].
- The rule mostly formalizes the end of the AI Diffusion Rule and has little practical bite, because remote access itself stays uncontrolled (very unlikely, about 8%). What would tell it apart: Pure remote access is not redefined as an export, and the AI Diffusion and model-weight controls are removed [5][1].
Base rate
Of 25 earlier comparable stories about organisations in the New Runtime archive over the last 12 months that are at least 3 months old, 13 (52%) were followed within 3 months by a release, access-change or pricing story on the same company or product line. Of 25 comparable archive stories, 13 were followed within 3 months by a release, access-change or pricing story. That rate of about one in two fits our expectation that follow-on actions such as license policies, country additions or amendments are plausible but not assured.
What would change this view
- A Federal Register correction, delay or withdrawal notice for FR Doc. 2026-23529 (signal): The operative dates or provisions described here no longer apply as published.
- BIS issues a broad general license or a raised TPP threshold for existing data centers before the 90-day deadline (signal): The rule's practical bite on non-US operators is much weaker than its text suggests.
- Major non-US cloud providers in Malaysia or Singapore report halted GPU deliveries or cut off Chinese tenants (signal): This would confirm that the rule is reshaping where compute is located, beyond what its text alone shows.
Still open
- How will BIS measure 'aggregate TPP' across linked facilities in practice, and how will it treat shared multi-tenant clusters?
- Will BIS process licenses for existing data centers within the 90-day window, and on what policy?
- Which additional countries might the Secretary designate under the 'tolerate evasion' language?
- Did Commerce disagree with the rule internally? The press release attributes authorship to Treasury and the NSC, and that could shape later amendments.