The Defense Held and the Building Emptied Anyway

Ardem Patapoutian won the Nobel Prize in Physiology or Medicine in 2021 for working out how human cells sense temperature and touch. He did it at Scripps Research, where he has worked since 2000. In early 2025, after the administration's cuts to NIH left Scripps facing a $38 million shortfall, his federal grant was halted. Then China called with an offer: any city, any university he wanted, with funding guaranteed for twenty years.

He said no. He stayed, because he loves the country he adopted, and he said so plainly. What he also said is the part worth sitting with: the scientists coming up behind him will not all have the standing, the security, or the sentimental attachment to turn that down.

That's the shape of this thing. It is not a story about famous people being poached, because the famous people are mostly staying. It's a story about thousands of quieter decisions being made by researchers with an established lab, a functioning career, and no particular desire to relocate, all doing arithmetic about whether the funding environment they built a life inside is still something they can plan against.

Applications from US-based researchers to the European Research Council's Starting Grant program climbed sharply for the 2026 cycle. A Nature poll found three quarters of US-based scientists surveyed were considering leaving the country. Canada committed roughly C$1.7 billion to a dedicated talent recruitment scheme. France, Germany, and Austria stood up their own.

Here is the part that should be strange: on paper, the funding fight was won.

What the Scoreboard Says

Congress mostly held. The courts mostly held. The topline number survived.

The administration's FY2026 budget proposed cutting NIH by roughly 40 percent, from $48 billion down to $27.5 billion, alongside a cut of more than half to NSF and a flat 15 percent cap on indirect cost reimbursement across all federal research grants. Indirect costs are the unglamorous infrastructure of research: facilities, utilities, compliance staff, animal care, the IT that keeps data secure. Negotiated rates at major research universities typically run somewhere between 50 and 70 percent. A flat 15 percent cap would have functioned as a deep cut disguised as an accounting reform.

Congress rejected essentially all of it. The enacted FY2026 bill kept NIH near level funding at roughly $47.2 billion, preserved the existing institute structure, and blocked the indirect cost cap outright.

The courts finished the job, and the sequence matters. A federal district court in Massachusetts permanently enjoined the cap in April 2025. The First Circuit unanimously affirmed on January 5, 2026, holding that NIH had violated both statutory law and its own rulemaking procedures. The Justice Department then let the Supreme Court filing deadline pass on April 6, 2026 without petitioning, making the injunction permanent. Underneath all of it sat a provision most people have never heard of: since fiscal 2018, appropriations law has barred NIH and HHS from developing or implementing a modified approach to indirect cost rates, and Congress carried that language forward in the Consolidated Appropriations Act, 2026. The cap didn't lose on a technicality. It lost because Congress had already prohibited it, repeatedly, in writing, for seven years running.

There is also a legitimate version of the critique underneath all this, and it deserves acknowledgment before being set aside. NIH peer review has real flaws. It has been criticized from the left as well as the right for conservatism, for favoring established investigators over risky work, for a payline structure that rewards incrementalism. Reasonable people have argued the system needs restructuring. None of that is what happened here, but it exists, and pretending the institution was perfect before 2025 would be its own kind of dishonesty.

So: the deepest cuts failed, the illegal mechanism got enjoined, and the money stayed roughly flat. That reads like a system working.

Why the Building Emptied Anyway

Because the number was never the mechanism.

Money can be restored with a signature. The things it was supposed to buy cannot.

Consider what actually happened during the year the topline was being defended. NIH spent much of 2025 with funding frozen or front-loaded, with agencies directed to award half of remaining annual funds up front. That sounds procedural. What it does is compress the number of new grants an institute can fund even when the annual total never changes, which pushes the payline down and rejects proposals that would have been funded in any normal year. Not rejected on merit. Rejected because the money was already spoken for.

Now consider what that looks like from inside a career. A postdoc deciding where to take a first faculty position is not evaluating this year's appropriation. They're evaluating whether the appropriation is knowable five years out, because that's the length of the project they're proposing and the timeline of the students they'd be training. A principal investigator deciding whether to keep a lab open is making the same calculation with more people depending on the answer.

And the answer available to them is that the fight is annual. The FY2027 request proposes cutting NSF by 55 percent, from $8.75 billion to $3.96 billion, landing barely three months after the FY2026 version had been beaten back. NSF's own budget documents spell out what that would mean in units of actual scientific work: roughly 2,100 research grants funded in FY2027 against 5,800 in FY2025, a funding rate falling from 19 percent to 7 percent, and total people supported by agency activities dropping from 305,400 to 94,100. That last number includes senior researchers, postdocs, graduate students, and undergraduates. It is, in other words, a proposal to remove about two thirds of the people from the pipeline that produces American scientists.

The lesson available to anyone paying attention is not "the cuts failed." It's "the cuts will be proposed again next year, and the year after, and the outcome each time depends on whether a handful of appropriators feel like fighting."

Uncertainty does not require winning. The FY2026 proposal lost, and the damage landed anyway.

That's the mechanism, and it's more efficient than defunding would have been. An actual 40 percent cut would have produced a visible catastrophe, congressional hearings, and a clear target for reversal. Recurring instability produces a slow, distributed, individually rational exodus that never generates a single news event and cannot be reversed by restoring a number.

The Part That Persists

Annual budget fights end. Rules do not.

The FY2026 and FY2027 appropriations fights are noisy and temporary. There's a quieter thing running underneath them with a much longer half-life.

In May 2026, OMB published a proposed rewrite of the Uniform Guidance, the government-wide rulebook covering somewhere between $1 trillion and $1.2 trillion a year in federal grants across roughly forty agencies. Two provisions matter here. First, discretionary awards would require pre-issuance review by senior political appointees, who would have authority to override the outcome of scientific peer review. One physician-researcher commenting on the docket singled out the specific section, §200.205, that does this. Second, agencies could terminate an award at any point if it no longer aligns with agency priorities or the national interest, a "termination for convenience" standard imported from federal procurement contracts, with sharply limited ability to appeal or object.

Read those together from the perspective of someone planning a research career. The first converts "will my proposal be judged on its merits" into "will a political appointee approve it." The second converts "I have a five-year grant" into "I have a five-year grant that can be cancelled in year three for reasons I cannot anticipate and cannot contest."

That is a categorically different planning environment than a tight payline. And unlike an appropriations number, it does not reset when the next budget passes. Rules take years to unwind, and this one is designed to be harder to unwind than the guidance it replaces.

The Accounting

Nobody made an irrational decision. That is the problem.

Scientific training is not information transfer. It's apprenticeship in things that don't exist in any protocol document: the intuition for why an experiment isn't working, which of six plausible variations produces clean results, which questions in a field are genuinely open versus quietly settled. That knowledge transfers by proximity, over years, from people who have it to people who don't. Building a researcher who can train other researchers takes roughly twenty years from undergraduate to independent principal investigator, and the ones leaving now are the ones who would have been training the 2040s cohort.

So the loss isn't a stock of people. It's a compounding loss of training capacity, and it lands on a schedule that guarantees nobody currently in office will be present when the bill comes due.

None of the people leaving are panicking. That's what makes this hard to reverse. A researcher who moves their lab to Munich because they can't plan against a funding environment that regenerates a crisis every twelve months has read the situation accurately and responded proportionately. The appropriations wins were real, and they were insufficient, and the people making twenty-year decisions understood that distinction faster than the people writing the headlines did.

They are also going somewhere specific. The recruitment infrastructure catching them was built deliberately, by governments that identified an opportunity and moved on it, and the largest of those governments has spent the last decade preparing for exactly this. That's the next piece.

Like what you've read, but not able to switch to paid? Why not buy me a coffee instead! All money goes to help defray the expense of running the site.