On July 23, the Trump administration released its much anticipated AI Action Plan, a blueprint for achieving what it describes as “unquestioned and unchallenged global technological dominance” in artificial intelligence. The Plan is the administration’s most comprehensive statement to date for how it intends to win the AI race against China, spur innovation at home, and promote U.S. technologies overseas.
Despite the administration’s sharp rhetorical break from its predecessor, the Plan reflects notable continuity with several key Biden-era priorities. Both administrations have emphasized scaling domestic AI companies, increasing government adoption of AI tools, and reforming permitting and regulatory processes to meet the extraordinary compute and energy demands required to train and deploy advanced models.
Still, the new Plan diverges from earlier approaches in some significant ways. It is strikingly techno-optimist in tone, generally prioritizing AI innovation over the need for guardrails. Together with an executive order targeting “woke AI,” it emphasizes the risks of “ideological bias” in certain AI models and calls for limiting federal procurement to model developers “who ensure that their systems are objective and free from top-down ideological bias,” although how exactly that will be implemented remains unclear. And in the wake of Congress’s rejection of the 10-year moratorium on state-level AI regulation, the Plan now seeks to block AI-related federal funding from flowing towards “states with burdensome AI regulations.”
The Plan’s scope and ambition have largely resonated with the AI community. Major U.S. tech companies have welcomed the Plan’s emphasis on stripping red tape and promoting the U.S. tech stack overseas. Technooptimists have praised its emphasis on a “dynamic, ‘try-first’ culture of experimentation” and its “upholding [of] American values and free speech.” Advocates of open-source systems have celebrated the plan’s calls for increased investment in open models, although the executive branch has limited resources and influence in this domain. Perhaps more surprisingly, some AI safety advocates, along with analysts and companies generally more concerned by the risks of advanced AI systems, have also expressed qualified support for many of the Plan’s recommendations.
Ultimately, the Plan’s impact will depend on implementation. How the administration handles critical areas such as export controls, immigration, AI testing and evaluations, labor, and international engagement will determine whether this sweeping vision becomes a reality.
Export controls
Export controls on advanced AI chips and semiconductor manufacturing equipment have been core to U.S. AI policy since the first Trump administration. This Plan likewise recognizes that “Denying our adversaries access to [advanced AI compute]… is a matter of both geostrategic competition and national security.” Yet it comes a week after the administration approved the sale of Nvidia’s H20, a powerful AI chip designed for running AI models, to China. That decision seems to reflect the administration’s view that prioritizing the preservation of Nvidia’s market share in China—keeping U.S. adversaries “addicted to the American technology stack,” in Commerce Secretary Howard Lutnick’s words—overrides the benefits of denying China millions of advanced AI chips. Balancing these considerations is not straightforward, but the AI action Plan provides little guidance for how to think through this trade-off.
The Plan also calls on the United States to strengthen export control enforcement and impose restrictions on semiconductor manufacturing equipment. Yet the administration has so far done little on this front. Indeed, the day after the Plan’s release, the Financial Times reported that at least $1 billion worth of Nvidia’s advanced AI processors, including the banned B200 chip, were smuggled to China over the last three months. To respond to this problem, the administration could expand restrictions on equipment, control component sub-systems, target Huawei’s expanding chip fab network, or restrict sales to well-documented Chinese smuggling routes in southeast Asia. But for now, as Alasdair Phillips-Robins and I have argued elsewhere, the Trump administration appears to be avoiding actions Beijing might view negatively to deescalate the trade conflict. As the administration seeks access to both a presidential visit to Beijing and China’s rare earths, it may struggle to roll out the Plan’s call for robust semiconductor manufacturing controls.
Immigration
Notably, the AI Action Plan says nothing about immigration, even as the administration doubles down on policies that make it tougher for skilled immigrants to enter. This is a critical omission considering that around half of PhD-level AI researchers in the United States are foreign-born, almost two-thirds of top AI startups were founded by immigrants, and 72 percent of immigrant AI startup founders arrived on student visas. Indeed, three-quarters of Meta’s new Superintelligence team are first generation immigrants.
As a matter of politics, the choice not to wade into this debate is unsurprising: high skilled immigration is one of the most sensitive cleavages between technologists and the president’s MAGA base. But as a matter of policy, it is a serious gap in a strategy designed to accelerate American innovation and outcompete China. In 2023, a bipartisan congressional report found that “the PRC is gaining on the United States in the race for global talent,” and China is rapidly outpacing the United States in producing STEM and AI PhD graduates.
Misuse Risks
Consistent with the rhetoric of many senior administration officials, the Plan shies away from the word “safety.” But it does discuss the need to defend against the misuse of powerful AI models and prepare for future AI-related risks. For example, it calls for the build-out of an AI “evaluations ecosystem” run by the Center for AI Standards and Innovation (CAISI, formerly the AI Safety Institute) that would, among other things, evaluate models for potential national security risks related to chemical, biological, radiological, and nuclear weapons, as well as cyber risks.
The Plan also emphasizes the need to promote secure-by-design AI technologies and applications. It proposes support for research into AI interpretability and control, biosecurity investments to prevent malicious actors from synthesizing harmful pathogens, and a new AI Information Sharing and Analysis Center to enable prompt incident reporting between the AI labs and the U.S. government.
Much of this is welcome. But the federal government will need the personnel, expertise, and authorities to make many of these goals a reality. For CAISI to deliver on the Plan’s national security priorities, it will need adequate resources and congressional codification. The administration has so far done little to fix the talent pipelines that plague federal hiring; in fact, the Department of Government Efficiency’s aggressive attempts to reduce federal headcount have often targeted probationary employees, who are often younger, less entrenched, and more technologically sophisticated. The administration’s broader hostility to federal R&D, science funding, and universities sits awkwardly with the Plan’s emphasis on the urgency of AI research, although DARPA and the national labs may still be able to fund useful work.
Labor and Automation
The Trump administration has dismissed concerns that AI will lead to widespread job losses. The Action Plan reflects that stance. It offers only perfunctory support for worker retraining programs, instead casting the Republican Party as unapologetically pro-AI. The Plan blends China-threat rhetoric, anti-woke signaling, and deregulatory zeal to bind together two uneasy allies: small-government tech leaders and the MAGA movement’s economically populist base. But this posture may prove politically risky. Polls show Americans remain more anxious than optimistic about AI’s impact; experts are far more positive about AI than the general public. And cracks in the coalition are already visible, including over issues like high-skilled immigration and the federal moratorium. For now, the administration is aligned on accelerating AI innovation and deployment. Whether that unity survives if AI-driven job displacement begins to emerge is a different question.
Export Promotion
The Plan—and an accompanying executive order—calls for the diffusion of the U.S. tech stack as a cornerstone of American AI leadership. Many of its suggestions here are welcome: the United States needs a forward-looking international engagement strategy. But to pay off, it will need sustained diplomatic and commercial engagement, backed up by sources of strategic capital. The Plan says relatively little, however, about reforms to U.S. export promotion institutions, such as reauthorizing, expanding, and streamlining the Export-Import Bank and the Development Finance Corporation.
There is also a deeper contradiction at play. The Plan calls for exporting U.S. AI systems to “ensure our allies are building on American technology,” yet it sits uneasily alongside the administration’s broader trade and alliance posture. Foreign governments are already voicing concerns about becoming too dependent on U.S. cloud providers. And Washington’s growing reliance on coercive trade tools—including sweeping tariffs on swing states and close allies alike—risks undermining the very trust and alignment that an export-led AI strategy demands.
* * *
The AI Action Plan represents the Trump administration’s most ambitious statement yet on AI leadership. Whether it becomes anything more than another aspirational policy document remains to be seen.
