On January 13, 2025, the US Department of Commerce’s Bureau of Industry and Security (BIS) published an Interim Final Rule (IFR) titled “Framework for Artificial Intelligence Diffusion.” The IFR builds on a long series of recent export control restrictions on advanced computing hardware, technology, and components critical for artificial intelligence (AI)1 and seeks to establish an ambitious framework for managing export and security terms for the spread of AI going forward. It does so through a number of measures, including (i) tightening existing controls on advanced computing integrated circuits (IC), computing equipment, and related technology; (ii) introducing new export controls on closed AI model weights; (iii) requiring export licensing for most countries; and (iv) establishing a tiered security framework for securing authorization to receive the most advanced US AI technology.
From both a regulatory and business planning perspective, the IFR offers both compliance challenges and significant business opportunities. It has broad implications for companies across the AI landscape, including AI cloud providers, hyperscale data center developers, owners and operators, advanced chip and computing manufacturers, and their suppliers. Affected companies should review the IFR carefully to assess the potential impact on their operations.
The IFR requires compliance starting May 15, 2025.2 It sets a deadline of May 15, 2025, to submit comments, although it is possible that a further comment period could be adopted to allow for consideration of industry comments before implementation.
Most significantly, the IFR imposes a global license requirement for exports of the most advanced computing ICs, computing equipment and commodities containing those ICs, and certain production equipment necessary for their production, as well as related software and technology (“Advanced AI Chip Items”).
The framework establishes a three-tiered country structure for these items:
The IFR defines several new conditional license exceptions authorizing license-free exports of Advanced AI Chip Items in Group 3 destinations:
License exceptions are highly conditional and in certain cases subject to limitations and should therefore be reviewed carefully in order to ensure compliance.
The IFR also includes two categories of the Data Center Validated End-User (DC VEU) Authorization intended to enable eligible and vetted data center owners and operators to procure and establish large quantities of Advanced AI Chip Items for the establishment and operation of data centers around the world:
The IFR also adds new controls on exports, reexports, or transfers (in-country) of advanced closed-weight AI models trained with 1026 or more computational operations (“Controlled Model Weights”). 7 Published model weights are excluded from the scope of the control. The IFR imposes export control licensing requirements on these advanced model weights for all destinations. End users who are, or whose ultimate parent is, headquartered in the United States or another Whitelisted Country will be able to receive Controlled Model Weights in any destination outside the Prohibited Country Group, subject to important limitations and requirements.
Notably, BIS issued new “red flag” guidance involving scenarios where US subsidiaries of entities headquartered outside a Whitelist Country Group country use a US Infrastructure as a Service (IaaS) provider to assist in training an advanced AI model and seeks to obtain the resulting model weights. BIS cautions US IaaS services providers that, under those circumstances, the domestic transfer of a Controlled Model weight to the US subsidiary of the foreign-headquartered entity could expose the provider to liability under the Export Administration Regulations depending on the facts and, at a minimum, raises an obligation to inquire with respect to the intended destination prior to the domestic transfer.
The framework set out under the IFR impacts a broad range of companies across the AI supply chain, including chip designers and chip manufacturers, as well as developers, owners, and operators of AI data centers. While the IFR will introduce additional complexity and compliance requirements, it is also intended to establish clear pathways for:
(i) License-free procurement and deployment of unlimited AI computing power in the US and other Whitelisted Countries.
(ii) Companies headquartered in those Whitelisted Countries to aggressively expand and deploy anywhere else (other than the small handful of arms-embargoed countries) upon applying for and receiving “validated” status and to receive the most advanced model weights, subject to security obligations.
(iii) For the vast majority of destinations outside Whitelisted Countries, continued substantial, though more limited, deployment of AI computing power subject to quantity limitations and the ability for companies headquartered in those countries to access significantly higher computing power upon applying for and receiving “validated” status.
Potentially impacted companies should carefully consider the potential application of these rules to their operations, including whether there may be potential opportunities to leverage the pathways above for strategic advantage.
1 See Mayer Brown Legal Updates discussing requirement related to AI models and advanced computing items here and here.
2 Paragraphs 14, 15, and 18 of Supplement No. 10 to Part 748 have a delayed compliance date of January 15, 2026 and cover security for chips and data, AI cybersecurity, and personnel security respectively.
3 Supplement No. 5 to Part 740 lists Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, Republic of Korea, Spain, Sweden, Taiwan, the United Kingdom, and the United States.
4 Country Group D:5 countries include Afghanistan, Belarus, Burma, Cambodia, Central African Republic, China, DRC, Cuba, Cyprus, Eritrea, Haiti, Iran, Iraq, North Korea, Lebanon, Libya, Nicaragua, Russia, Somalia, South Sudan, Sudan, Syria, Venezuela, and Zimbabwe.
5 The Technical Notes to ECCN 3A090 state that TPP is defined as “2 x ‘MacTOPS’ x ‘bit length of the operation’, aggregated over all processing units on the integrated circuit.” Further, attrition of ICs due to loss, damage, failure, relocation, and resale will not count toward the country allocation.
6 Qualifying countries will be listed in paragraph (b) to Supplement No. 5 to Part 740.
7 The IFR defines “parameters” for these purposes as “any value learned during training (e.g., network weights, biases, etc.).”
Mayer Brown is a global legal services provider comprising associated legal practices that are separate entities, including Mayer Brown LLP (Illinois, USA), Mayer Brown International LLP (England & Wales), Mayer Brown Hong Kong LLP (a Hong Kong limited liability partnership) and Tauil & Chequer Advogados (a Brazilian law partnership) (collectively, the “Mayer Brown Practices”). The Mayer Brown Practices are established in various jurisdictions and may be a legal person or a partnership. PK Wong & Nair LLC (“PKWN”) is the constituent Singapore law practice of our licensed joint law venture in Singapore, Mayer Brown PK Wong & Nair Pte. Ltd. More information about the individual Mayer Brown Practices and PKWN can be found in the Legal Notices section of our website.
“Mayer Brown” and the Mayer Brown logo are the trademarks of Mayer Brown.
Attorney Advertising. Prior results do not guarantee a similar outcome.