A new US Commerce rule targets remote access to advanced AI chips through overseas data centres. Learn what cloud operators and customers should review.
Cloud computing has made advanced artificial intelligence infrastructure available to organisations that do not own specialist chips or operate large data centres. A business can rent processing capacity remotely, train or run models in the cloud, and serve customers across borders. That flexibility is now receiving closer regulatory attention. On 15 November 2026, the US Department of Commerce’s Bureau of Industry and Security (BIS) released an interim final rule addressing remote access to advanced semiconductors through overseas data centres.
The change matters because export controls have traditionally focused on where physical chips and servers are shipped. The new rule also examines who can access computing capacity after the equipment has arrived in another country. For cloud providers, data-centre operators, chip suppliers and companies buying large-scale AI services, customer checks and end-use records are becoming more important.
What the new US rule changes
The interim final rule adds controls for certain advanced computing items when exporters, re-exporters or transferors know that equipment is destined for a data-centre use that would provide significant computing capacity to a company headquartered in Macau or a destination listed in US Country Group D:5. The rule covers direct and indirect access, including arrangements in which separate facilities are digitally linked and share computing resources.
A central threshold is more than 12 million units of aggregate total processing performance, or TPP, available to the relevant customer or group of linked data centres. TPP is a regulatory measure of computing performance, not simply a count of machines. Connected facilities can be assessed together when their infrastructure allows them to share resources, balance workloads or replicate data.
The rule also introduces additional destination-based licensing requirements for specified advanced semiconductor items shipped to Indonesia, Malaysia, Singapore, Thailand and Brazil. It includes a limited exception for commercial recommender systems exclusively designed and used to rank, filter or suggest products, services or content. That exception should not be interpreted as a general exemption for generative AI systems.
When does it take effect?
The document was issued as an interim final rule, but its effective date is tied to publication in the Federal Register. The public-inspection copy lists 19 November 2026 as the scheduled publication date and says comments are due 60 days after publication. Existing overseas data centres already providing covered levels of computing capacity to relevant customers have a 90-day period after the effective date to obtain authorisation if they wish to continue that service.
This timing is important. The announcement is not the same as saying every AI workload or cloud service is immediately prohibited. The licensing test depends on the covered items, the destination and ownership of the customer, the level and purpose of remote access, and what the parties know about the intended end use. Businesses should read the actual rule and obtain qualified export-control advice before making compliance decisions.
Why remote access is now part of compliance
A data centre can be physically located in one country while its computing resources are used by a customer headquartered somewhere else. A customer may also obtain capacity through a reseller, affiliate or managed-service provider. If compliance teams only review the shipping address or the name on a purchase order, they may miss the organisation that ultimately benefits from the computing capacity.
BIS therefore places emphasis on due diligence. Data centres, foreign foundries and chip exporters are expected to examine customers, including direct or indirect ownership and control. The rule also addresses situations in which a data centre changes its end use after the original shipment. Customer identity, beneficial ownership, workload purpose and later changes in service can all become relevant.
For technology companies, the broader lesson is that cloud infrastructure is no longer separate from supply-chain governance. Server location, component origin, customer ownership and digital access arrangements can affect whether a transaction needs regulatory review.
What cloud and AI businesses should review
Companies do not need to assume every customer or workload is affected. They do need a process for identifying transactions that may fall within the rule. A practical review should include:
- Map advanced chips and servers in the supply chain, including where they are exported, re-exported or transferred.
- Identify ultimate parent companies and beneficial owners of customers, resellers and major service partners.
- Record which customers can access high-capacity computing environments and whether facilities are digitally linked.
- Document intended workloads and create a process for reviewing material changes in customer ownership or use.
- Ask legal and compliance teams to assess licensing requirements before committing to new infrastructure or service contracts.
- Keep customer screening, capacity records and export-control decisions together for audit purposes.
Cloud customers should ask providers how contracts address regulatory changes, service continuity, data location and notice periods. A customer may not control a provider’s export obligations, but it can request clarity about whether a service depends on particular chips, regions or third-party infrastructure. This is especially relevant for organisations relying on a single cloud environment for model training or critical applications.
What the rule does not mean
The rule does not establish that all AI services are restricted, nor does it automatically prohibit ordinary business use of cloud software. It is directed at specified advanced computing items and particular end-use and destination circumstances. It also formally rescinds the earlier AI Diffusion Rule, which BIS says had never been enforced. That rescission is separate from the new controls on remote access to advanced computing capacity.
Businesses should avoid relying on short summaries or social-media posts to decide whether a service is affected. The full text contains technical definitions, licensing provisions and reporting obligations. The impact on an individual transaction depends on its specific facts.
The next phase of AI infrastructure governance
AI competition is increasingly shaped by access to computing power as well as model design. Governments are trying to balance commercial innovation with national-security concerns, while cloud providers make specialised hardware available around the world. The new BIS rule brings those developments into closer contact.
For responsible operators, the strongest response is to improve visibility rather than abandon international cloud services. Knowing who uses infrastructure, where capacity is located, how facilities are connected and which rules apply can reduce uncertainty. Businesses should monitor the rule’s publication, review implementation guidance and update compliance processes where necessary.
Sources and related reading
BIS interim final rule on remote access to advanced semiconductors. BIS announcement, 15 November 2026.
TechCrunch reporting, 16 November 2026. Related Digismartiens coverage: AI governance, cloud continuity, and Pakistan IT exports.



