Introduction
The US Department of Commerce’s Bureau of Industry and Security (BIS) has unveiled a comprehensive framework aimed at regulating the export of advanced computing chips and artificial intelligence (AI) technology. This initiative [1] [6], known as the “AI Diffusion rule,” seeks to bolster US national security, address economic interests [2], and limit access to advanced technology for adversarial nations [5], particularly China [1] [2] [3], Russia [1] [2] [5] [7], Iran [6] [7], and North Korea [5] [6] [7]. The framework introduces a tiered system of access for different countries, with the goal of maintaining US leadership in AI technology while ensuring that benefits are shared with allied nations.
Description
The new regulations categorize countries into three tiers of access. First-tier countries face no restrictions [7], while second-tier countries can receive up to 50,000 advanced computing chips [7], potentially increasing to 100,000 with technology security agreements with the US [7]. Trusted nations [3] [6], including the UK [6], Canada [6], South Korea [5] [6] [7], Taiwan [5], Australia [5], and several European countries [5] [6], will have relatively free access to advanced AI silicon and algorithms [6]. The controls specifically target advanced computing chips and certain closed AI model weights, closing loopholes in existing export restrictions that already focus on adversaries like China and Russia [2]. Notably, no restrictions apply to chip sales to 18 key allies and partners [3], allowing jurisdictions with robust technology protection regimes to make large-scale purchases [3]. Under the License Exception Artificial Intelligence Authorization (AIA) [4], orders with collective computation power up to approximately 1,700 advanced GPUs do not require a license [3] [5], facilitating rapid shipments of US technology for benign purposes [3], particularly by universities [3], medical institutions [5], and research organizations [3] [5] [7].
Entities that meet high security and trust standards and are based in allied countries can obtain “Universal Verified End User” (UVEU) status [3], enabling them to deploy up to 7% of their global AI computational capacity in various countries [3], potentially amounting to hundreds of thousands of chips [3]. This status is granted globally and allows for flexible expansion while maintaining frontier training within the US [3]. Additionally, the “National Verified End User” (NVEU) status permits entities in non-concern countries that meet security requirements to purchase computational power equivalent to up to 320,000 advanced GPUs over two years, supporting trusted national entities in serving local and regional customers while mitigating diversion risks [3].
Non-VEU entities outside close allies can still purchase up to 50,000 advanced GPUs per country [3], ensuring US technology remains accessible to foreign governments and local businesses [3]. Government-to-government arrangements can double chip caps for nations that align their export control and technology security efforts with the US [3].
The framework also includes the License Exception Advanced Compute Manufacturing (ACM), which facilitates the export and transfer of advanced computing chips for development and production [4], with restrictions on arms-embargoed countries [4]. Advanced semiconductors sold abroad are restricted from being used by countries of concern for training advanced AI systems [3], while access for general-purpose applications remains [3]. The transfer of model weights for advanced closed-weight models is restricted to non-trusted actors [3], although open-weight models can still be published [3]. Security standards are established to protect the weights of advanced closed-weight AI models [3], ensuring secure storage and use while preventing unauthorized access [3].
US officials have expressed particular concern about data centers in the Middle East and Southeast Asia [2], where Chinese companies have leveraged technology that cannot be imported to China [2]. The rapid implementation of the new rules has raised alarms within the industry, with the Information Technology Industry Council cautioning that it could disrupt global supply chains and disadvantage US firms [2]. The Semiconductor Industry Association has voiced disappointment over the expedited policy [2], warning it could negatively impact the US economy and competitiveness in semiconductors and AI [2].
These regulations build on previous measures aimed at safeguarding US national security [3], reflecting extensive engagement with stakeholders [3], Congress members [3], industry representatives [3], and foreign allies [1] [3] [4]. A 120-day comment period will be established to gather input from experts [5], industry stakeholders [3] [5], and partner nations regarding the new regulations [5]. License applications for exports to specific destinations will be reviewed with a presumption of approval until a designated allocation is reached [4], after which a presumption of denial will apply [4]. Adequate physical and cybersecurity measures will be required for license approval [6], and security conditions will be enforced to protect advanced models and mitigate risks associated with the diversion of advanced computing chips [4].
The proposed Interim Final Rule (IFR) for an “Export Control Framework for Artificial Intelligence (AI) Diffusion” introduces country-specific caps and a licensing regime on US exports of semiconductors [1], particularly impacting GPU chips essential for AI applications like large language models (LLMs) [1]. While the initiative aims to restrict the use of advanced US technologies by geostrategic competitors [1], it raises significant economic [1], security [1] [2] [3] [4] [5] [6] [7], and foreign policy concerns [1] [4]. The framework imposes aggregate country caps on US advanced chip exports, potentially limiting market opportunities for US companies while benefiting foreign competitors [1], particularly in China [1] [2].
Compliance challenges are anticipated [1], as companies may struggle to determine if exports exceed country-specific compute caps [1], complicating an already unpredictable and costly licensing process [1]. The proposed regulations could distort the market [1], with the government inadvertently favoring certain countries while risking damage to the US semiconductor industry [1]. A more effective strategy would focus on controlling frontier AI system installations rather than the total number of GPUs [1], allowing for a more precise assessment of risks associated with advanced AI models [1].
Conclusion
The proposed global country cap system for AI compute presents significant challenges and potential drawbacks. While it aims to enhance US national security and maintain technological leadership, it also raises concerns about economic impacts, market distortions, and compliance difficulties. A more consultative and effective approach may be necessary to balance these objectives and ensure the continued competitiveness of US firms in the global AI and semiconductor markets.
References
[1] https://itif.org/publications/2025/01/07/export-controls-on-ai-chips-bidens-overreach-risks-us-leadership-in-tech/
[2] https://apnews.com/article/biden-ai-artificial-intelligence-chips-computer-trade-4495b5b4a48e856dc612e7abe3e47d20
[3] https://www.whitehouse.gov/briefing-room/statements-releases/2025/01/13/fact-sheet-ensuring-u-s-security-and-economic-strength-in-the-age-of-artificial-intelligence/
[4] https://www.bis.gov/press-release/biden-harris-administration-announces-regulatory-framework-responsible-diffusion
[5] https://www.usatoday.com/story/news/politics/2025/01/13/biden-ai-chip-sale-curbs/77663159007/
[6] https://www.wired.com/story/new-us-rule-aims-to-block-chinas-access-to-ai-chips-and-models-by-restricting-the-world/
[7] https://arstechnica.com/ai/2025/01/biden-administration-puts-quotas-on-global-ai-chip-sales/
