The Information Technology Industry Council, a technology industry group representing Amazon, Microsoft, Meta and other companies, urged the Biden administration not to issue regulations at the “last minute” of the presidential term that would restrict global access to artificial intelligence (AI) chips, warning that these Restrictions would jeopardize U.S. leadership in AI. The Information Technology Industry Council said the rule would impose arbitrary restrictions on U.S. companies’ ability to export computer systems and cede global markets to competitors.
Endangering U.S. Technological Leadership
As the Biden administration nears the end of its term, one of its final acts is to propose an over-designed but under-informed interim final rule (IFR) to establish a new “Artificial Intelligence (AI) Proliferation Export Control Framework.” The rule would impose restrictions on certain countries and impose a licensing system on U.S. semiconductor exports, which would most notably affect GPU chips that underpin critical artificial intelligence applications. While the Biden administration is using export controls to restrict geostrategic rivals like China and Russia from using advanced U.S.-developed technologies to strengthen themselves. This certainly makes sense logically, but the proposed framework fails to meaningfully address core challenges and could have disastrous consequences for U.S. digital industry leadership. Furthermore, the Biden administration developed the proposed rules with almost no input from industry and ignored better ways to address core challenges.
The Problems with The Proposed Framework are Manifold
Limiting and controlling the global sales of U.S. advanced chips by limiting the total amount of each country will raise major economic competition, security and even foreign policy issues. The bottom line is that placing restrictions on U.S. AI GPU exports would limit market opportunities for U.S. companies while opening the door for foreign AI chip suppliers to enter and capture market share. By artificially restricting the export of AI chips to most countries around the world, the U.S. government is actually opening the door to foreign competition around the world, diverting revenue that would otherwise go to U.S. chip manufacturers to invest in next-generation chips with overseas competitors.
The proposed regulations misunderstand a key element of the AI chips used to develop LLMs and solve other computing challenges: GPUs are both scale-up and scale-out, meaning their power comes from operating multiple GPUs simultaneously to solve computing challenges. Therefore, there is no point in controlling GPUs when competitors can achieve parity simply by adding more (even less powerful) GPUs to solve computing challenges. In other words, even if competitor-made GPUs are less powerful than U.S.-made GPUs in the short term, rival companies will be happy to provide the computing power needed to meet customer challenges, which will undermine the government’s efforts to limit overall AI computing. goals while undermining U.S. companies’ leadership in global artificial intelligence computing technology and market share.
Beyond that, the policy could create significant market distortions, particularly by putting the U.S. government in the difficult position of picking winners and losers, both in specifying country-specific caps and in specifying which companies can hit those caps. The U.S. government would actually have to develop a system to oversee and regulate the sales of GPUs by U.S. companies to non-exempt foreign countries. If the U.S. government’s estimates are wildly wrong—for example, a friendly ally like Singapore could easily consume twice as much GPU computing power—then market demand won’t go unmet. Instead, it will simply be filled by foreign competitors, who will then get a share of future sales. This could severely damage high-value industries pioneered by American innovators who lead the world in technology and markets.
U.S. companies currently account for 70% to 90% of the global market share of generative AI solutions. U.S. chipmaker NVIDIA itself accounts for 70% to 95% of the global AI accelerator market. U.S. competitiveness in the AI/cloud/hyperscale market extends well beyond established companies like AWS, Google, Oracle, and NVIDIA to include other leaders like Cerebras, Tenstorrent, Groq, and D-Matrix, as well as dozens of other startups. These companies employ tens of thousands of high-paying jobs and generate significant exports for the U.S. economy. The Biden administration has devoted so much energy and resources to revitalizing the U.S. semiconductor industry (notably spearheading passage of the $52.7 billion CHIPS Act) that it is now handcuffing companies as they seek to sell their world-leading chips to global markets It makes no sense to live with them and risk damaging their leadership status.
Proposed Framework Upends U.S. Export Control Policy
As U.S. GPU solutions for AI remain the most sought-after globally, this raises concerns that countries may compete and fight for higher caps, bringing foreign policy considerations into the mix. The United States may risk damaging diplomatic relations with various countries.
The framework is designed to take a very blunt approach rather than address what should be the core goal: restricting core competitors from engaging in AI computing capabilities that are inconsistent with core U.S. national security interests. U.S. export control policy has long been associated with targeting specific end users and end uses of U.S.-developed technologies, which the U.S. hopes to prevent adversaries from using.
The proposed framework directly upends decades of U.S. export control policy, moving it from a narrow, tailored approach to a blanket licensing system for all countries (even if some are exempt). This represents a significant expansion in the conceptualization and operation of the U.S. export control regime and therefore demonstrates how excessive these proposed regulations really are.
Summarize
As ITIF has previously argued to the U.S. Bureau of Industry and Security (BIS), regulators should shift from focusing on compute-based assessments of AI systems to performance-based assessments. In fact, computing power is not a good measure of AI model risk. Some high-compute-power models may pose minimal threats, while superior-performing low-compute power models may pose a greater risk. Moving to performance-based thresholds will provide more accurate capability and risk assessments, better identifying state-of-the-art AI models. The next administration should focus on improving U.S. competitiveness in the field of artificial intelligence, expanding market access to U.S. chips and artificial intelligence technology, and limiting the influence of its geostrategic competitors in the field. The vast majority of uses for these chips will be legal, and the goal should be to capture and retain as much of this market as possible.
Many of the concerns relate to the Biden administration’s “artificial intelligence (AI) spread export control framework.” Perhaps most concerning, however, is that although these rules were promulgated with little industry input, they have been submitted as interim final rules. Furthermore, the rules will take effect only 60 days after being published in the Federal Register, barely enough time to set up such a massive regulatory system, let alone give industry a chance to adapt.
