A company can train an AI model on chips it never buys or takes delivery of. It rents computing capacity from a data center and sends the work over a network. A new U.S. Commerce Department rule brings certain arrangements like that into the scope of advanced chip export controls, making the identity of the remote customer relevant even when the hardware stays put.
The Bureau of Industry and Security (BIS) announced the interim final rule on November 16, 2026. The Federal Register’s public inspection notice lists November 19 as the scheduled publication date. The rule text says it takes effect on publication and gives the public 60 days from that date to comment. BIS announcement Federal Register public inspection copy
When remote access can trigger licensing
The new end-use control targets transactions involving specified advanced chips and systems when the supplier knows a data center will provide significant computing capacity to certain customers. It covers customers headquartered in Macau or a destination in Country Group D:5, which includes China, as well as companies whose ultimate parent is headquartered in one of those places. Both direct and indirect access can count, and the rule accounts for capacity spread across linked data centers. BIS says it will review license applications case by case. Rule text, § 744.24
The rule also addresses changes after delivery. If a data center books a new customer that triggers the control, a shipment that previously went ahead without a license can still give rise to an authorization requirement for the new use. The text sets out a transition provision for certain data centers that already provide covered access. It also describes a narrow exclusion for systems used exclusively to recommend commercial products, services, or content, with information that must be submitted to BIS. A system that also generates new content falls outside that definition. Rule text, § 744.24(b), (d), and (e)
A separate rule for chip shipments to five markets
The same measure adds destination-based licensing for specified advanced items shipped or transferred to Brazil, Indonesia, Malaysia, Singapore, and Thailand. It provides an exclusion for certain U.S.-headquartered recipients or recipients with a U.S.-headquartered ultimate parent, subject to reporting conditions. This is a separate question from whether a data center serves a covered remote customer. A shipment may need review because of where the hardware is going, while an existing cloud arrangement may need review because of who uses the capacity. Rule text, amendments to § 742.6
The rule also removes provisions from the 2025 AI Diffusion Rule, including its country tiers and model-weight controls. That change and the new licensing requirements appear in the same document, so the removal of the earlier framework does not tell an operator whether any particular chip transaction or cloud arrangement is covered. Rule text, section IV
Why this matters for cloud operators and customers
Consider an operator with two facilities that pool accelerator capacity. One team manages the hardware inventory, another onboards customers, and a third schedules jobs across both sites. The rule brings those records together. Reviewing an arrangement may depend on the chips in use, which facilities share capacity, the customer’s ultimate parent, and how much compute the customer can access across the combined system.
That also changes what customers may be asked to provide when they reserve a large block of computing time. Operators may need more detail about corporate ownership and how a workload will use the capacity, especially when a reseller sits between the operator and the end customer. How often onboarding changes, and how long license decisions take, will depend on implementation. BIS’s stated aim is to close a gap between physical chip shipments and remote access. Its announcement does not measure how well the new controls will achieve that goal. BIS explanation of the rule’s purpose
Cover image: AI-generated conceptual illustration.
