The US Commerce Department has formally rescinded the Biden administration's AI Diffusion Rule, a tiered export control framework issued in January 2025 that would have placed dozens of US allies in a second-tier classification restricting their access to advanced AI chips. The rule's compliance requirements were set to take effect in May 2025. BIS Under Secretary Jeffery Kessler directed enforcement officials not to apply the rule while a formal rescission is published, characterizing it as "ill-conceived and counterproductive" and saying it would have "undermined US diplomatic relations with dozens of countries."
The rescission does not mean looser controls. BIS issued three guidance documents alongside the announcement targeting Huawei Ascend chips specifically, warning US companies against allowing their chips to be used for training or inference of Chinese AI models, and providing guidance on protecting supply chains from diversion. The SMCI indictment unsealed days later, which alleged a $2.5 billion scheme to route AI servers to China through a Southeast Asia transshipment network, illustrated precisely the diversion risks that guidance addresses.
A replacement rule is in development and the direction is tighter, not looser, in one specific respect: the emerging framework would move away from country-tier classifications toward a transactional model that conditions export licenses on what foreign buyers agree to do. Officials are weighing requirements that would tie chip export approvals to overseas investment commitments in US AI data centers or to security guarantees, according to reporting from Reuters. One proposal under consideration would trigger these requirements for orders above 200,000 chips.
The practical consequence for semiconductor supply chains is significant uncertainty. Lead times for high-end GPUs and custom AI ASICs are already constrained by wafer capacity at TSMC and Samsung. Export licensing rules determine which customers can actually receive allocations already in production queues. When those rules are in flux, orders placed under previous eligibility assumptions may require reassessment, particularly for deployments in the Middle East and Southeast Asia, two regions that have seen substantial AI data center investment from both US hyperscalers and sovereign wealth funds.
The policy also has a design implication. An earlier generation of China-specific restrictions prompted chip vendors to ship modified variants with artificially limited interconnect speeds or memory bandwidth to remain below regulatory thresholds. The incoming framework's emphasis on specific performance metrics, including compute density and interconnect bandwidth, signals that similar design workarounds could re-emerge as vendors try to maintain export flexibility for constrained markets.
The replacement rule has not yet been published and no timeline has been given. Until it is finalized, the gap between the Biden-era framework's rescission and the new rule's implementation leaves the export control posture for advanced AI chips operating under the older pre-diffusion-rule baseline, supplemented by the three guidance documents issued alongside the rescission.
Sources: Bureau of Industry and Security, Reuters
–
By the Control Plane Editorial Team