In the final weeks of the Biden administration, the U.S. Department of Commerce announced new interim final export control regulations aimed at “regulating the global proliferation” of advanced artificial intelligence chips and models. This rule would significantly undermine America’s global AI leadership if it is not rescinded or amended by the Trump administration during or shortly after the 120-day comment period that ends on May 15th.
While the AI diffusion rules address both advanced computing chips and AI model weighting, this post will focus on chip export restrictions and their associated implications.
3 country hierarchy
The AI penetration rule divides 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, Netherlands, New Zealand, Norway, South Korea, Spain, Sweden, Taiwan and the United Kingdom. . Shipments of advanced AI chips from the United States to entities in these countries are generally not restricted, but there are some important exceptions. (For example, you cannot ship unrestricted to a first-tier country entity whose parent company or headquarters is in a non-first-tier country.)
At the bottom are the US arms embargoed countries (officially designated country group D:5) and the Chinese administrative region of Macau. The list includes about 24 countries, including China, Cuba, Iran, North Korea, and Russia. The AI Proliferation Rule leaves in place very strict existing regulations that prevent the export of advanced AI chips to these destinations.
About 150 other countries around the world are in the middle tier, subject to a complex and problematic set of restrictions on exports of advanced computing chips. These limitations create a cascade of unintended consequences.
Global computing shortage mandated by the US government
To accommodate middle-tier countries, the AI Dissemination Rules introduce something called “Universal Validated End User” or “UVEU.” Companies “headquartered or whose ultimate parent is headquartered” in the top country can apply to be considered UVEU, but the criteria for granting such an application are unclear. One of the most problematic parts of the rule is:
“A UVEU headquartered in a country (top tier) may have more than 25% of its total AI computing capacity, i.e. the AI computing capacity owned by all its entities (sic) subsidiaries and parent company; or 7% of total AI computing power outside of “(Tier tier countries)” may not be transferred or installed in its subsidiaries and parent companies. may not be transferred or installed in a single country other than a “Tier tier Country” or into a single country other than a “Tier tier Country”.
What this means, in plain English, is that the U.S. government has forced approximately 150 mid-sized countries into an artificial shortage of advanced computing chips that the U.S. supplies, and then a small number of U.S. companies We plan to make these chips available on a limited basis. It distributes tips to companies that are concentrated in some of those countries.
The regulation also introduces “National Verified End Users” (NVEU). Companies outside of the lowest-tier countries can also apply to join the NVEU and, if granted, advanced computing chips “in specific countries specified in the NVEU authorization,” subject to strict company-by-company and company-by-company rules. can be exported. Quarterly capacity limits for each country. It is unclear which combinations of companies and countries the regulator will privilege when making NVEU decisions.
market trends
Regardless of market demand, UVEU-authorized companies will be restricted from shipping more than 25% of their “total AI computing capacity” to companies in mid-sized countries. If that happens, the global market available for America’s advanced computing chips will shrink significantly.
Additionally, UVEU-certified companies will be limited to shipping no more than 7% of their total AI computing capacity to mid-sized countries. This will result in each UVEU identifying three countries where it can most effectively focus its sales efforts to achieve the 7% cap, reaching 21% across the three countries. That is, a large portion of 25% of the total allocation.
The middle-tier countries most likely to be favored by UVEU are those with the largest markets (such as India and Brazil). However, receiving “preferential treatment” may not be such a victory. That’s because middle-class countries with large markets will almost certainly find that America’s revenue cap means they won’t always be able to meet their computing needs. Moreover, if only a handful of U.S. companies receive UVEU approval, most mid-sized countries could be left with little access to large quantities of advanced computing chips, leaving them in a bind.
Advanced computing chips: Much more than AI
Although AI makes use of sophisticated computing chips, those same chips have a wide range of applications, many of which are not necessarily related to AI. Advanced computing chips are designed to perform mathematical operations very quickly and often in a highly parallel manner. This makes it useful for analyzing large datasets, generating graphics, solving optimization problems, and many other tasks commonly performed using non-AI-based methods.
Preventing middle-tier countries from accessing these chips will not only hinder advances in AI, but also broader efforts to advance computing infrastructure more generally, with implications that will affect nearly every part of society. It will have a ripple effect on the field.
A centrally planned global computing economy
The AI Pervasive Rule imposes the equivalent of a centrally planned global computing economy and gives U.S. regulators an unprecedented level of control over the global distribution of computing resources. Under this framework, a cartel of privileged companies selected by the government as UVEU would develop close ties with government supervisors.
The inevitable regulatory capture means that the UVEU is in a good position to lobby governments to interpret and update regulations in a way that preserves privileged market access.
Regardless of the quality of the product, it can be very difficult for startups to get a seat at the table. Moreover, artificially induced supply shortages will drive up prices for advanced computing chips and systems, benefiting UVEU but hurting mid-market countries and non-UVEU companies.
Strange limitations in computing
Ten years from now, we will look back and realize how far-fetched it was for the U.S. government in the mid-2020s to try to limit the ability of people in 150 countries to perform fast multiplication. Sho. Moreover, these disadvantaged countries will not sit back and accept the US government’s attempts to turn them into computing backwaters.
Instead, it will work to develop a supply chain that is independent of the United States in a way that avoids triggering U.S. jurisdiction under the Foreign Direct Product Rule. Countries such as Brazil, India, Israel, and the UAE are well-positioned to ramp up investments aimed at securing new ways to access increased computing power. Preventing mid-sized companies from relying on the United States for computing chips is a surefire way to encourage alliances outside the United States, including stronger technology ties with China. Newly emerging non-U.S. manufacturers of advanced computing chips will be willing to expand their global market share at the expense of U.S. companies, which are hindered from meeting market demand.
A better way to maintain U.S. AI leadership
Current U.S. excellence in computing is an integral part of the feedback cycle that maintains U.S. AI leadership. Strong global sales of computing technology by US companies translate directly into high levels of domestic investment. This investment will create jobs, lead to the development of the next generation of even faster chips and systems, and provide funding for investments in and acquisitions of promising U.S. computing startups. The resulting environment makes the United States the choice for the next generation of young entrepreneurs and computer science graduate students with a vision of advancing the frontiers of computing.
While the wording of the AI Proliferation Rule remains uncertain in many respects regarding its implementation and interpretation, the shortcomings of the framework set forth in the rule are clear. Instead of promoting national security, it would undermine the presence of U.S. computing technology in global markets and encourage the development of a global AI ecosystem based outside the United States. A far better approach would be to shelve the AI diffusion rules and let the new administration set the policy. A holistic, blank-slate approach to identifying ways to help the United States maintain and build on its lead in AI while protecting national security.