What happened
The U.S. Bureau of Industry and Security (BIS) published an interim final rule (IFR) titled “Addressing Remote Access to Advanced Semiconductors” on November 16, 2026. The rule, which takes effect upon publication in the Federal Register, adds two primary controls aimed at preventing U.S. advanced AI chips and services from being used to support Chinese entities via third‑country data centers. First, a new § 744.24 requires chip exporters to have “knowledge” that the chips will be used in a foreign data center that provides more than 12 million aggregate tera‑operations per second (TPP) to a covered Chinese or Macau‑based entity, triggering a license requirement. Second, BIS re‑interprets an existing § 744.23(a)(3) to apply a 50 percent usage test retroactively, meaning any foreign data center where a Chinese‑headquartered company rents at least half of the compute capacity now requires a license, regardless of whether the hardware physically passes through the United States. The IFR also adds a destination‑based license requirement for advanced computing items shipped to Indonesia, Malaysia, Singapore, Brazil, and Thailand, labeling them “higher‑than‑normal risk of diversion.” Finally, the rule tightens foundry‑level controls, requiring that chips designed by entities in certain high‑risk jurisdictions be either approved or authorized before a foundry can export them without a license. The AI Diffusion Rule, rescinded in 2025, is formally removed from the Code of Federal Regulations.
On November 16, 2026, the Bureau of Industry and Security (BIS) released an interim final rule (IFR) called “Addressing Remote Access to Advanced Semiconductors.” The rule is slated to become effective the day it appears in the Federal Register, with a 60‑day public comment window thereafter.
The IFR introduces a new control, § 744.24, which mandates that exporters of advanced computing items obtain a license if they know the items will be sent to a foreign data center that will provide more than 12 million aggregate tera‑operations per second (TPP) to a company headquartered in China, Macau, or a Country Group D:5 jurisdiction. The rule does not apply to data centers located within the United States.
BIS also re‑interprets an existing provision, § 744.23(a)(3), to apply a 50 percent usage test retroactively. Under this test, if a Chinese‑headquartered entity rents at least half of a foreign data center’s capacity, a license is required, even if the hardware never physically enters China. This interpretation covers scenarios where a third‑country data center hosts Chinese customers, effectively treating the shipment as “for an entity” in China.
A separate provision adds a destination‑based license requirement for advanced computing items shipped to Indonesia, Malaysia, Singapore, Brazil, and Thailand, citing a higher‑than‑normal risk of diversion. BIS reserves the right to impose similar requirements on other destinations by notice.
The rule also tightens foundry‑level controls. Chips designed by entities in Country Group D:1, D:4, D:5, or the five newly listed countries must be either approved or authorized before a foundry can export them without a license. This removes previously available self‑attestation pathways for such designs.
Source details: natlawreview.com
Why it matters
The new BIS rule represents a significant escalation in U.S. export‑control policy for advanced semiconductor and AI resources. By extending the definition of “for an entity” to include remote compute services, the rule closes a loophole that allowed Chinese firms to access cutting‑edge AI hardware indirectly through data centers in Southeast Asia and other third‑country locations. This retroactive interpretation means that past transactions that were previously considered compliant may now be deemed violations, exposing exporters, data‑center operators, and foundries to potential penalties and requiring extensive due‑diligence documentation. The added destination‑based licensing for five countries signals heightened scrutiny of supply‑chain routes that could be used for transshipment, potentially slowing the flow of advanced chips to emerging markets and increasing compliance costs for manufacturers and distributors. For U.S. chip makers and cloud providers, the rule could reshape market strategies, prompting a shift toward more tightly controlled sales channels or the development of compliance infrastructure to track end‑use and end‑user information. The exclusion for “commercial recommender systems” suggests a nuanced approach, but the reporting requirement for such exemptions adds another layer of administrative burden. Overall, the IFR tightens the U.S. government’s ability to enforce national‑security objectives in the AI hardware domain, with ripple effects across global semiconductor supply chains and AI service markets.
By extending export‑control jurisdiction to remote services, the IFR closes a loophole that allowed Chinese firms to access U.S.‑origin AI chips via overseas data centers. This retroactive stance could expose past transactions to enforcement, creating legal and financial risk for exporters and data‑center operators who previously believed they were compliant.
The 12 million TPP threshold and the 50 percent usage test introduce quantifiable metrics that companies must monitor, likely increasing compliance costs and requiring new reporting systems to track end‑use and end‑user data across multi‑jurisdictional supply chains.
The destination‑based licensing for five countries signals a broader shift toward geographic risk assessments, potentially slowing chip shipments to emerging markets and prompting firms to reconsider supply‑chain routes that involve those jurisdictions.
The exemption for commercial recommender systems, while narrow, indicates BIS’s willingness to carve out limited use‑cases, but the required reporting to the End‑User Review Committee adds administrative overhead and may deter firms from seeking the exemption.
Overall, the rule strengthens U.S. national‑security objectives by tightening control over advanced semiconductor technology, but it also creates uncertainty for global AI infrastructure providers and could reshape market dynamics for AI services.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
Why can ethical evaluation not be reduced to one model score?
What to watch next
Stakeholders will monitor how quickly the IFR is published in the Federal Register and the ensuing 60‑day comment period, which could lead to revisions or clarifications. Companies with existing contracts for remote AI services to China should assess whether their arrangements now fall under the 50 percent test and prepare licensing requests if needed. Enforcement actions or civil penalties announced by BIS or the Department of Commerce will indicate how aggressively the retroactive provisions are being applied. Additionally, watch for industry responses, especially from Southeast Asian data‑center operators and chip foundries, as they may lobby for adjustments or seek exemptions. Finally, the impact on AI model‑ transfers remains unaddressed, so future rulemaking may target that area if the current approach proves insufficient.
The exact publication date in the Federal Register and the content of the 60‑day public comment period, which may lead to adjustments in thresholds or definitions.
How quickly companies with existing remote‑ contracts to China move to obtain licenses or restructure agreements to avoid violations.
Any enforcement actions or civil penalties announced by BIS, which would clarify the agency’s willingness to apply the retroactive provisions.
Industry lobbying efforts, especially from Southeast Asian data‑center operators and chip foundries, that could result in future amendments or additional exemptions.
Potential follow‑up rulemaking addressing AI model‑ transfers, which remain unregulated under the current IFR.