In the final week of the Biden administration, the U.S. Department of Commerce released a new interim final export control rule that aims to “regulate the global diffusion” of advanced artificial intelligence chips and models. The rule, if not reversed or revised by the Trump administration during or shortly after the 120-day comment period that ends on May 15, will significantly undermine U.S. global AI leadership.
While the AI diffusion rule addresses both advanced computing chips and AI model weights, the present post focuses on the restrictions on chip exports and the associated consequences.
3 tiers of countries
The AI diffusion rule partitions the world into three categories of countries. The top tier consists of the United States and 18 other countries: Australia, Belgium, Canada, Denmark, Finland, France, Germany, Ireland, Italy, Japan, the Netherlands, New Zealand, Norway, the Republic of Korea, Spain, Sweden, Taiwan, and the United Kingdom. Shipments from the United States of advanced AI chips to entities in these countries are generally unrestricted, though there are some important exceptions. (For example, unrestricted shipments cannot be made to entities in a top-tier country with a parent company or headquarters located in a non-top-tier country.)
The bottom tier is U.S. arms embargoed countries (formally designated Country Group D:5), as well as Macau, which is an administrative region of China. This list contains approximately two dozen countries including China, Cuba, Iran, North Korea, and Russia. The AI diffusion rule leaves in place the existing extremely strict regulations blocking export of advanced AI chips to these destinations.
The world’s approximately 150 other countries are in a middle tier, to which exports of advanced computing chips will be subject to a byzantine and problematic set of limits. These limits will create a cascade of unintended consequences.
US government-mandated global compute scarcity
To address countries in the middle tier, the AI diffusion rule introduces something called a “Universal Validated End User” or “UVEU.” Companies “headquartered in, or whose ultimate parent is headquartered in” a top-tier country can apply to be deemed a UVEU, though the criteria for granting such an application are unclear. One of the most problematic portions of the rule is as follows:
“A UVEU headquartered in a country [in the top tier] cannot transfer or install more than 25% of its total AI computing power —i.e., the AI computing power owned by the entity [sic] all its subsidiary and parent entities—to or in locations outside of [top-tier countries], and cannot transfer or install more than 7% of its total AI computing power to or in any single country outside of” top-tier countries.
In plain English, here’s what this means: The U.S. government will subject the approximately 150 middle-tier countries to an artificial scarcity of U.S.-supplied advanced computing chips, and then will anoint a small set of U.S. companies to dole a limited number of these chips out to entities concentrated in a few of those countries.
The rule also introduces “National Verified End Users” (NVEUs). Companies outside of the bottom-tier countries can apply to be an NVEU, and if authorization is granted, can receive advanced computing chip exports “in a specific country as specified in their NVEU authorization,” subject to a strict per-company, per-country quarterly cap on processing power. It is unclear which company/country combinations regulators will privilege when making NVEU decisions.
Market dynamics
Regardless of market demand, companies with UVEU Authorization will be limited to shipping no more than 25% of their “total AI computing power” to entities in middle-tier countries. That will dramatically shrink the available global market for American advanced computing chips.
Additionally, companies with UVEU Authorization will be limited to shipping no more than 7% of their total AI computing power to any one middle-tier country. This will push each UVEU to identify the three countries where they can most effectively focus their sales efforts to hit their 7% cap, thus reaching 21% across the three countries; i.e., most of their total allocation of 25%.
The middle-tier countries most likely to be favored by UVEUs will be those with the largest markets (e.g., places such as India and Brazil). But being “favored” may not be much of a victory, because middle-tier countries with large markets will almost certainly find that the American sales caps mean that their computing needs won’t consistently be met. In addition, if only a handful of U.S. companies receive UVEU authorization, most middle-tier countries may find themselves out in the cold, largely unable to obtain significant quantities advanced computing chips.
Advanced computing chips: Much more than AI
While AI makes use of advanced computing chips, those same chips have a broad range of applications, many of which don’t necessarily involve AI. Advanced computing chips are designed to perform mathematical operations very quickly, and often in a highly parallel manner. This makes them useful for analyzing large sets of data, generating graphics, solving optimization problems, and many other tasks that are commonly performed using non-AI-based methods.
Impeding access to these chips by middle-tier countries will not only stifle their progress in AI, but will also obstruct their broader efforts to advance their computing infrastructure more generally, with impacts that will cascade across nearly every sector of society.
A centrally planned global computing economy
The AI diffusion rule imposes what amounts to a centrally planned global computing economy, giving U.S. regulators an unprecedented level of control over the global distribution of computing resources. Under this framework, the cartel of privileged companies selected as UVEUs by the government will build close relationships with their government overseers.
The inevitable regulatory capture will mean that UVEUs will be well positioned to push the government to interpret and update the rules in ways that preserve their privileged market access.
It may be extraordinarily difficult for newly emerging companies, regardless of the quality of their products, to get a seat at the table. In addition, the artificially induced supply scarcity will drive up prices for advanced computing chips and systems, to the benefit of the UVEUs and to the detriment of middle-tier countries and non-UVEU companies.
Quixotic limits on computing
A decade from now, we will look back and recognize how quixotic it was for the U.S. government of the mid-2020s to attempt to limit the ability of people in 150 countries to perform fast multiplications. Furthermore, these disfavored countries are not simply going to sit back and accept U.S. government attempts to make them computing backwaters.
Instead, they will work on developing U.S.-independent supply chains in ways that avoid triggering U.S. jurisdiction under the Foreign Direct Product Rule. Countries including Brazil, India, Israel, and the UAE are eminently capable of ramping up investments aimed at securing new ways to access increased computing capacity. Preventing companies in middle-tier countries from relying on the U.S. to supply computing chips is a surefire way to push them into building non-U.S. alliances that include stronger technology ties with China. Newly ascendant non-U.S. makers of advanced computing chips will be more than happy to grow their global market share at the expense of U.S. companies that are prevented from meeting market demand.
A better way to maintain US AI leadership
The current U.S. preeminence in computing is an integral part of a feedback cycle that maintains U.S. AI leadership. Strong global sales of computing technology by U.S. companies translate directly into high levels of domestic investment. That investment creates jobs, leads to the development of the next generation of even faster chips and systems, and provides the capital to invest in and acquire promising U.S. computing startups. The resulting environment is why the U.S. is the destination of choice for the next generation of young entrepreneurs and computer science graduate students with a vision to advance the frontiers of computing.
While the language of the AI diffusion rule leaves much about its implementation and interpretation uncertain, the shortcomings of the framework laid out in the rule are clear. Instead of promoting national security, it will undermine U.S. computing technology presence in the global marketplace, spurring the development of a global AI ecosystem anchored outside the U.S. A far better approach would be to shelve the AI diffusion rule, and let the new administration take a holistic, clean slate approach to identifying ways to help the U.S. retain and build its lead in AI while also protecting national security.
The Brookings Institution is committed to quality, independence, and impact.
We are supported by a diverse array of funders. In line with our values and policies, each Brookings publication represents the sole views of its author(s).
Commentary
The new AI diffusion export control rule will undermine US AI leadership
January 23, 2025