The US government has extended its advanced semiconductor export-control framework to cover certain forms of remote access to computing capacity, closing a route through which customers could use powerful chips installed in data centres outside their home country. The Bureau of Industry and Security’s interim final rule, titled Addressing Remote Access to Advanced Semiconductors, was published in the Federal Register on November 19 and took effect the same day. It establishes restrictions on exports, reexports and in-country transfers to overseas data centres that remotely provide significant computing capacity to companies headquartered in specified restricted jurisdictions, including Macau.
The rule treats access to computing power as a potential control point even when the physical chips do not cross the border to the end customer. That distinction reflects the way AI workloads are increasingly provideed. A company may rent access to accelerators in a foreign data centre and submit training or inference jobs over a network, rather than purchase and operate the hardware itself. The US Commerce Department says the new restrictions aim to prevent companies in countries of concern from using US-origin advanced semiconductors to train advanced super-intelligence models through remote infrastructure.
The rule targets compute services and chip shipments
The Federal Register text places due-diligence obligations on data centres, foreign foundries and chip exporters. They must examine customers, including direct and indirect ownership or control by persons in the specified jurisdictions. This moves compliance beyond checking the name on an order form. Providers may need to understand who ultimately controls a customer, who benefits from the computing service and whether a nominally eligible company is acting on behalf of a restricted entity. The requirements could affect customer onboarding, contractual representations, screening systems and the records that providers retain to show compliance.
The rule also introduces licensing requirements for advanced semiconductors destined for Indonesia, Malaysia, Singapore, Thailand and Brazil, which the department identifies as destinations with a higher-than-normal risk of diversion. The text provides an exception for certain shipments to companies headquartered in the United States, or whose ultimate parent is headquartered there, subject to reporting requirements. The measure So does not impose a blanket ban on all chip exports to those markets. Its application depends on the item, destination, end user, ownership and the relevant licensing conditions.
A separate part of the rule changes presumptions applied to front-end chip fabricators and outsourced semiconductor assembly and test providers when they export advanced-node logic chips. The changes reduce the routes available to overcome certain presumptions for chip designers located in destinations subject to licensing requirements for advanced semiconductors. That provision connects the remote-compute restrictions with the physical manufacturing chain. It gives US authorities additional oversight over attempts to use production capacity in third countries for customers that would face restrictions if they sourced chips directly.
Providers face new customer-screening and licensing work
The rule is an interim final rule rather than a final, permanently closed regulatory process. The Federal Register notice says comments on its revisions and additions must be submitted by January 19, 2027. Companies affected by the requirements will need to assess the text as published while monitoring any changes that follow the comment period. The notice also formally rescinds the regulations implementing the Biden-era AI Diffusion Rule, which the department says had never been enforced and whose rescission it had announced in May 2025.
For cloud operators, the practical question is how to identify workloads and customers that trigger the new requirements without disrupting unrelated services. Providers may need more detailed customer declarations, beneficial-ownership checks, location controls and escalation procedures when a customer’s ownership or usage pattern changes. A data-centre operator that sells ordinary hosting to a broad customer base may have to distinguish those services from access to advanced semiconductor capacity at a scale covered by the rule. The distinction is important because the notice does not treat every remote computing session as controlled activity.
The policy also creates compliance questions for international customers that rely on overseas cloud infrastructure. A company may not own an advanced accelerator, yet its ability to train a model can depend on access to a large cluster operated by a third party. Under the new framework, the location of the data centre alone is not enough to determine whether the service is permitted. The customer’s headquarters, ownership, the amount and type of compute, and the provider’s due-diligence process can all become relevant.
The rule’s reach also depends on the definitions and thresholds used by regulators. The Federal Register describes restrictions for overseas data centres that remotely supply significant volumes of compute to companies headquartered in specified destinations. That wording leaves operational questions for providers that serve customers through subsidiaries, joint ventures or intermediaries. The ownership checks aim to reduce the risk that a restricted company can obtain access through a nominally separate entity, but they require providers to maintain accurate information and update it when corporate structures change. Compliance teams will need documented decision-making rather than a one-time check at account creation.
The rule’s immediate effect is to make remote access part of the semiconductor-control conversation. Its longer-term impact will depend on how BIS interprets significant computing capacity, how providers implement screening, and whether the comment process changes any provisions. For now, chipmakers, cloud providers, data-centre operators and customers using large AI clusters have a new compliance document to evaluate before entering or renewing cross-border compute arrangements.