White House AI science plan makes the federal funding mix the real test

• The FY 2028 guidance asks agencies to shift a larger share of their R&D portfolios toward foundational research, AI-enabled discovery and shared infrastructure; it is not an appropriation.
• It explicitly casts new institutional models, including NSF X-Labs, as complements to conventional academic, industry and federal laboratories.
• Reporting on deep cuts to NSF core programs raises a real financing question, but it does not prove that X-Labs caused the cuts.

The White House has set a broad new direction for federal research: more foundational science, more AI built into scientific work, and more flexible ways to fund people and teams. The consequential question is not whether the memo abolishes university research—it does not—but whether agencies can add those ambitions without shrinking the investigator-led programs that supply much of the underlying research.

An initial report described the guidance as affecting roughly $200 billion in annual federal R&D spending. The more precise baseline is the National Center for Science and Engineering Statistics' count of $194.2 billion in actual FY 2024 federal R&D obligations—a measure of transactions such as contracts awarded and services received, rather than a new pot of appropriated money. That difference matters: the White House document gives agencies priorities for future budget formulation; it does not allocate dollars by itself.

A portfolio shift, not an instruction to abandon universities

The FY 2028 memorandum tells agencies to seek a greater share of foundational research relative to later-stage development, particularly in physical sciences, computer science, engineering, mathematics and foundational biological sciences. It also directs them to identify the programs through which they would make that shift in their submissions to the Office of Management and Budget.

That is a funding-mix instruction, not a claim that every kind of science should be redirected to AI. The memorandum says the government's comparative advantage is long-horizon work whose returns are broadly distributed and difficult for private industry to capture. At the same time, it asks agencies to connect discovery more closely to manufacturing, regional ecosystems and industry partners.

What AI-enabled science means here

The policy is more specific than support for generic AI research. It calls for domain-specific scientific foundation models; curated datasets and data infrastructure for AI training and inference; AI-ready instruments; robotics; and closed-loop experimental systems. For compute, agencies are told to offer capabilities differentiated from commercial services—such as secure environments, unique federal datasets or specialized hardware—while using commercial capacity when cost-effective.

The central vehicle is the Genesis Mission, a government-wide effort launched by Executive Order 14363 in November 2025. The order makes the Energy Department responsible for implementation within DOE and gives the Assistant to the President for Science and Technology a coordinating role across participating agencies through the National Science and Technology Council. Its proposed American Science and Security Platform is meant to combine national-laboratory supercomputers, secure cloud AI environments, datasets, models and experimental tools. The order also makes that work subject to existing law and available appropriations.

The memo's counterweight to its own disruption

The guidance encourages agencies to try long-duration awards, fast grants, prizes, regranting experiments and funding that follows students and researchers among qualifying institutions. It also favors cost sharing and public-private partnerships. Those choices could broaden access to facilities and capital, but they also put more weight on funder-selected milestones, commercial participation and organizational design.

Its language is notably less absolute than the claim that federal science is moving "away from colleges." It says new institutional models should complement conventional principal-investigator laboratories, industry laboratories and federal R&D facilities. It also calls for greater access to federal facilities for university and industry partners, and for fellowships and placements across academic, industry and federal settings. Whether the resulting mix is genuinely additive is an execution question, not something the memorandum resolves.

NSF makes the financing tension concrete

The U.S. National Science Foundation is a useful early test because it has traditionally funded curiosity-driven research and is now building a vehicle intended to move research toward deployment. Its Technology, Innovation and Partnerships directorate, created in 2022 to help translate discoveries into societal and commercial benefits, is managing X-Labs.

In its May announcement, NSF said it would direct up to $1.5 billion over the next decade to X-Labs: independent teams of researchers, engineers and entrepreneurs working with milestone-based federal funding. The first calls seek teams for next-generation sensing and imaging, including AI-driven computational imaging, and for quantum-system interconnects and integrated photonics. NSF says the teams are intended to have enough runway and independence to take technology from early concepts or prototypes to commercially viable platforms ready for private investment.

That design is not interchangeable with the core-grant system. NSF will use an Other Transactions Agreement Solutions Offering for X-Labs, a mechanism that permits multiple topic announcements and large multiyear awards to selected teams. It targets coordinated, infrastructure-heavy work that the memorandum says can fall between conventional institutional models; it also concentrates discretion over recipients, goals and funding conditions.

What the cuts reporting establishes—and what it does not

A report based on an internal NSF memo and unnamed sources said the agency was reducing hundreds of traditional basic-science programs by 20% to 30% or more while its overall budget was down 3%. The June 18 memo reportedly said a unit in the mathematical and physical sciences directorate had its FY 2026 budget cut 30% from roughly $260 million in FY 2025, directed managers to pause funding requests, and warned that core programs, centers, facilities and other initiatives would be affected. The account also reported 60% cuts in each of three core geosciences programs and said NSF declined comment.

The same account cited Grant Witness's calculation that, through June 15, NSF had made one-eighth as many new grants as at the same point in FY 2025. It also said the reductions appeared to conflict with a final FY 2026 appropriations directive that no NSF directorate receive a reduction greater than 5% from its FY 2024 enacted level. Both points are reported claims, not an NSF explanation of how it allocated the money.

The same reporting said sources inside and outside NSF suspected the savings were meant to help finance X-Labs. An access-limited report similarly said it had learned NSF was planning to take money from core programs for a White House initiative. Those accounts make the alleged trade-off consequential, especially for proposals that had already received strong reviews. They do not establish a documented transfer from a particular core program to X-Labs, nor do they show that every cut had that cause.

FY 2024 federal R&D obligations by agency, in current dollars. Source: NCSES, U.S. National Science Foundation. Source: National Center for Science and Engineering Statistics.

The national plan is larger than one NSF program

X-Labs is salient, but it is small beside the federal R&D system the memo seeks to influence. The NCSES survey records NSF at $7.0 billion, or 4% of FY 2024 federal R&D obligations. The Defense Department accounted for $97.9 billion, or 50%, and the five largest agencies together accounted for $181.4 billion, or 93%.

Those are FY 2024 obligations in current dollars, not the FY 2026 NSF budget figures in the cuts reporting or FY 2028 requests still to be made. They cannot show whether the new priorities will be financed by new appropriations, delayed development spending, or smaller core-grant portfolios. They do show why the Genesis Mission depends on agencies beyond NSF: its platform needs shared computing, data, facilities and coordinated funding across departments that control most federal R&D activity.

What would settle the question

Agencies must reflect budget-formulation choices in their FY 2028 submissions to OMB. Separately, the memorandum requires agencies with at least $3 billion in FY 2026 R&D budget authority to provide, within 90 days, action plans explaining how FY 2026 and FY 2027 program execution will support its priority practices. Those plans are not a substitute for budget submissions; the memo says budget formulation remains outside their scope.

The useful evidence will be program-level: which foundational-research lines gain or lose funding, the sources of X-Labs awards, the terms and recipients of its milestone agreements, and whether agencies expand access to the Genesis infrastructure without displacing work it is meant to build on. Until those decisions are public, the strongest conclusion is narrower than a university funding exodus: federal science policy is being asked to support new AI-and-infrastructure institutions while the NSF episode tests whether the base can be protected at the same time.