Trump’s 'Big, Beautiful' Blunder on AI: Starving Science

Kat Duffy is Senior Fellow for Digital and Cyberspace Policy and Sebastian Elbaum is Technologist in Residence at the Council on Foreign Relations.

Most of the headlines about the budget bill passed by House Republicans last week concern a proposed federal moratorium on the enforcement of state AI laws. But the “Big, Beautiful Bill” to advance President Donald Trump’s agenda also makes a bold wager on protecting federal AI research—at least in theory. The budget correctly recognizes AI as a "critical [technology] for maintaining our economic and national security" and protects federal AI research. Yet, it guts funding for essential and broader scientific research supported by the National Science Foundation (NSF)—research that will power, and be powered by, AI’s advancement.

The administration's strategy is both politically expedient and strategically myopic. It demonstrates a fundamental misunderstanding of how AI development and innovation have been fueled in the United States. It is now up to the Senate to see the bigger picture and protect America’s future as a global leader in technology. Investing in AI research as an isolated field ignores the essential computing ecosystem that underpins AI's development. Data management and cybersecurity, scalable networking and distributed computing, programming languages and software engineering—these are not “AI” research; they are the result of broader federal investment in the scientific ecosystem that will both transform and be transformed by AI. Crippling investment in these foundational pillars will stunt AI's transformative potential at precisely the moment China and the US appear to be nearly tied in the race for geopolitical primacy in AI.

Perhaps more dangerously, such cuts undermine the value AI can bring to everyday Americans and the issues that affect them. AI is a general-purpose accelerator. Imagine a budget that prioritizes federal dollars for building advanced energy sources, but then slashes funding to support the use of that energy. What, precisely, would be the return on such an investment?

AI's true power lies in addressing real-world challenges, informed by deep domain expertise. Cutting-edge research in chemistry, physics, biology, materials science, geoscience, or engineering—all products of broader scientific exploration—can be turbocharged by AI, producing outcomes that help all Americans to benefit from AI and position the US as a leader in its application. Focusing overwhelmingly on AI research while defunding these essential research areas is akin to investing in a world-class supercomputer and then failing to give it meaningful problems to solve.

The ramifications are not solely domestic. Undermining America’s scientific foundations creates a strategic vulnerability its adversaries will eagerly exploit. China, in particular, continues to scale up its comprehensive AI infrastructure, and invest in its ability to apply AI broadly across multiple sectors in China and abroad—with clear intentions to surpass the US. This is a moment for America to seize every advantage it has, not to throw them away.

Congress and President Trump should urgently double down on funding initiatives that help level the AI playing field for America’s researchers and build a foundation for widespread innovation. Programs like the National AI Research Resource (NAIRR) demonstrate the impact of targeted federal investments that prioritize scientific research as well as AI. For just $10 million annually, the NAIRR-supported National Center for Supercomputing Applications gives thousands of researchers across 200 universities access to computational resources that would otherwise be monopolized by tech giants, including high-performance computing clusters, curated datasets, and specialized software libraries. That access enables researchers to tackle core AI problems, such as continuously training foundational models, as well as applied research in areas like material sciences or agriculture. Such an investment isn’t just sound public policy: it is sound fiscal policy. Compare the efficiency of that $10 million investment to the $170 million Meta is estimated to have spent on training a single AI model, or the $191 million Google spent on Gemini Ultra.

Congress has already acknowledged as much. For example, the bipartisan CREATE AI Act would formalize NAIRR, but it fatally relies on voluntary contributions from individual federal agencies and the private sector, rather than guaranteeing federal resources. Without mandated support, America's long-term research capabilities and geopolitical future remain vulnerable to political whims and market fluctuations.

A free market alone will not save us. For all its capability, speed, and scale, the private sector—be it a start-up or a vast multinational company—chases near-term returns, not the foundational scientific breakthroughs that have historically transformed America’s economy. That was particularly true in AI. For decades, researchers in universities pursued the possibilities offered by deep neural networks; it was only a decade ago that such research, which had been dismissed by many, catapulted AI forward and emerged as the underpinning of today’s increasingly massive AI models, infrastructure, and investments. To secure America's technological lead, including next-generation AI, the nation must leverage its full research capacity. The next paradigm shift could come from anywhere—including a university researcher whose funding is now on the chopping block. Underfunded academic and federal research labs risk losing talent and the potential to seed unforeseen breakthroughs that drive entire industries.

If President Trump truly aims to make America "the unrivaled world leader in critical technologies," the federal budget needs to reflect a fundamental truth: AI leadership requires a thriving scientific ecosystem, not just isolated investment. The proposed strategy in the “Big, Beautiful Bill” won’t power beautiful AI innovation: just big strategic loss. Thankfully, Congress still has the power to course correct in service of America’s global technological leadership. In 2025, as AI races forward, America should be doubling down on science—not starving it.