(Editor’s Note: This is the next installment of our series, “Tech Policy under Trump 2.0.” Read the first article in the series here).
On Jan. 13, in one of its final acts, the Biden administration launched its most ambitious attempt yet to shape the future of AI. In its “Framework for Artificial Intelligence Diffusion,” the first of a flurry of AI rules and executive orders released last week, the Biden administration tried to set the global terms for the market in advanced AI chips — the critical technology needed to build and run powerful AI systems.
The rule is the administration’s answer to what will be a defining question for U.S. foreign policy and economic strategy in the coming years: how widely should the United States share its AI technologies? The regulation is an extraordinary assertion of U.S. power over information technology, an attempt to increase the baseline of security standards at AI data centers worldwide, and another salvo in the United States’ escalating AI competition with China.
With President Donald Trump’s return to office, many of these eleventh-hour regulations will face significant scrutiny and potential rollback, especially in the face of a concerted lobbying campaign by U.S. industry. But as the new administration reviews its AI policy inheritance, it may find that the fundamental challenges that drove the design of the diffusion framework — balancing the protection of America’s technological edge with the promotion of U.S. systems and governance standards worldwide — remain as pressing as ever. In the coming days or weeks, the Trump administration will almost certainly walk back or adjust elements of the rule. But if it jettisons the framework in its entirety, it will need to design an alternative approach to plugging the gaps in its export controls on China without hindering the global rollout of U.S. AI infrastructure — and soon.
What’s in the Rule?
The rule is an attempt to balance two competing interests. On the one hand, there is enormous commercial pressure to export U.S. technologies widely. U.S. chip makers and cloud providers see massive opportunities in markets from Southeast Asia to the Middle East. Spreading its AI technology can help maintain the United States’ technological leadership and prevent countries from drifting further into China’s economic orbit. And the diffusion of AI technology worldwide could bring enormous benefits, including in educational outcomes, access to healthcare, and agricultural productivity, to billions of people in the Global South and elsewhere.
But U.S. policymakers also want to control the proliferation of a technology that may have critical national security implications. If AI proves to be as transformative as many predict — a hypothesis that has gained more credibility with the results of OpenAI’s o3 model series released in December, which showed massive improvements in several longstanding intelligence benchmarks — it could help U.S. adversaries develop sophisticated military systems, reduce barriers to developing WMD, enhance offensive (and defensive) cyber capabilities, and turbocharge human rights violations through mass surveillance technologies. The United States has good reason to maintain its lead in the race to develop these highly capable systems — and to preserve its control over the data centers hosting these systems.
The diffusion rule’s architects were clearly motivated by both of these considerations. The rule is designed to limit chip diversion to arms-embargoed countries like China and Russia, prevent the exfiltration and theft of AI model weights, and slow the development of frontier capabilities by U.S. competitors. But it is also designed to allow the controlled diffusion of AI technology worldwide in trusted and secure environments.
To do so, the rule divides the world into three tiers. A select group of eighteen states — including close NATO allies and key partners in the semiconductor supply chain, such as Japan, South Korea, and Taiwan — will maintain essentially unrestricted access to chips. Most other countries will fall into the middle tier and face limits on the total computing power they can import, with the ability to apply for more. A small group of adversaries, including China, remains effectively blocked from obtaining these technologies altogether.
The real innovation lies in how the framework turns AI exports into diplomatic and technological leverage. Countries in the middle tier — including Brazil, India, Saudi Arabia, and the United Arab Emirates — can dramatically increase their access to U.S. chips through several pathways. Most of these involve making verifiable government-to-government assurances or company-level reforms to align export control regimes with the United States; accelerate decoupling from Chinese AI efforts; and develop comprehensive safeguards, including cybersecurity standards, personnel vetting, and physical security protocols, to protect AI chips and model weights. In this, the rule builds on an approach that the U.S. government developed in recent negotiations with the UAE: the country’s leading tech company, G42, gained access to U.S. chips only after divesting from Chinese firms and stripping out Huawei technology.
Companies headquartered in Tier 1, meanwhile, such as Amazon, Google, and Microsoft, can deploy their chips globally — except to U.S. adversaries — so long as they keep at least half of their total computing power on U.S. soil and deploy no more than seven percent of their capacity in any one country. This approach is designed to allow U.S. tech companies to compete worldwide and perhaps even to entrench the long-term dominance of American hyperscalers. It also aims to prevent U.S. tech companies from offshoring their largest data centers to places such as the Gulf states, where abundant capital and cheap energy might be commercially attractive, but where national interests do not always align with those of the United States. In short, the order represents a preemptive move designed to foreclose the development of new dependencies on nondemocratic states.
How the World has Reacted
The AI diffusion framework has drawn intense criticism from some quarters. The European Union has complained that numerous member states were excluded from Tier One; Israel is similarly unhappy. Companies like Nvidia and Oracle, which lobbied aggressively against the rule, have described it as “the most destructive to ever hit the U.S. technology industry.” The rule imposes new compliance costs, threatens Nvidia’s ambitious international expansion plans, and may hit Oracle particularly hard given its significant planned investments in Malaysian datacenters.
At the heart of Nvidia and Oracle’s objections is the claim that the rule will drive tier two countries — comprising most of the world — towards Chinese alternatives. (Oracle has suggested renaming the rule the “Framework for the Advancement of Alibaba, Huawei, Tencent, and SMIC.”) The rule creates powerful short-term incentives to build data centers in the United States (facilitated by the administration’s executive order directing federal agencies to make sites available for AI infrastructure), work with U.S. hyperscalers, and adopt U.S. security standards. But it may also intensify long-term incentives for countries to find other sources of computing power. As one analyst put it, “Countries want chips and they don’t want to be told what to do with them.”
For now, however, there is no real Chinese alternative for countries looking for advanced AI chips. Thanks in part to existing U.S. export controls on advanced chips and semiconductor manufacturing equipment, China still struggles to manufacture enough advanced AI chips for its own purposes, let alone to export them. As a result, China continues to rely on smuggling and stockpiling of older western chips. That may change in the future, but as long as it remains true, the United States has significant leverage.
What Comes Next
As President Donald Trump takes office, the rule’s future is uncertain. Republicans may look askance at any policy introduced in the dying days of a Democratic administration, and Trump has already rescinded Biden’s October 2023 Executive Order on AI. But as the Trump administration studies the rule in more detail, it might find itself growing more attached to it. The rule’s broad approach to AI exports gives the new administration a powerful framework for using U.S. technological advantage as leverage in international negotiations — a framework entirely in keeping with an “America first” and transactional approach to foreign policy.
Its framework is also easily adaptable. Nearly two dozen NATO states currently find themselves in the restricted second tier, as do Israel, the Gulf states, and India — a crucial partner in America’s competition with China. The EU, meanwhile, is divided between tier one and tier two countries, with Poland a notable loser. The Trump administration can play with these tiering decisions, dangling them as carrots in bilateral negotiations. It can also take advantage of the validated end user status — the program that unlocks additional chips for specific, approved tier two companies — and tweak the requirements to achieve that status (such as by scrapping the human rights-related restrictions, should it so choose). In other words, this framework gives Trump a ready-made new bargaining chip.
If the Trump administration does scrap the Biden-era rule, it will have to come up with its own method of balancing the promotion of U.S. technologies and governance standards with preventing Chinese circumvention of U.S. export controls. The regulatory approach outlined in this rule is not the only answer to that question, but it is a comprehensive and adaptable framework. The next administration should seek to build on its foundation.