The Money Did Not Expire. It Changed Owners. NSF Closed FY2026 With $1 Billion Unobligated and $1.7 Billion Heading Into an OSTP Portfolio.

Two weeks ago the story was a countdown. The National Science Foundation was running roughly $1 billion behind its prior-year obligation pace with eight days left in the fiscal year, and the operating assumption across the research community — ours included — was that a large share of that money would simply lapse on October 1 and return to the Treasury. We wrote that up in Eight Days Left to Obligate.

The fiscal year has now closed, and the final numbers are in. The forecast was right about the shortfall and wrong about the ending.

NSF obligated $6.4 billion in research funds across 6,218 awards in FY2026. In FY2025 the comparable figures were $8.3 billion across 8,647 awards. In FY2024: 10,937 awards. That is a 23% decline in dollars and a 28% decline in award count year over year, and a 43% decline in award count against the FY2024 baseline.

But the unobligated balance did not evaporate. An OMB apportionment approved in August 2026 estimated that NSF would carry $1.7 billion of unobligated research funding into fiscal year 2027. An NSF spokesperson confirmed the agency had "carried over funds to invest in agency priorities that are aligned with broader administration initiatives."

That single sentence is the most consequential thing said about U.S. research funding this quarter, and it deserves to be unpacked carefully, because it changes what a grantseeker should actually do.

What a carryover is, and why it is not the same as a lapse

The confusion here is structural, so it is worth being precise.

Most discretionary agency program accounts hold one-year money: budget authority that expires at the end of the fiscal year for which it was appropriated. Unobligated balances in such an account cannot be used for new awards after September 30. The account stays open five more years for adjustments to existing obligations, then closes.

NSF's Research and Related Activities account does not work that way. R&RA is a multi-year (two-year availability) appropriation. Funds appropriated for FY2026 remain available for obligation into FY2027. This is a longstanding feature of the account, designed precisely so NSF can fund awards that straddle fiscal years without arbitrary end-of-September scrambles.

So the $1.7 billion is not stranded, not rescinded, and not — in the technical sense — lost. It is still NSF money, still obligable, still available to make grants.

What changed is who decides where it goes.

The Grand Research Challenges question

Of the carryover, NSF has been withholding roughly $1 billion to fund the White House Office of Science and Technology Policy's Grand Research Challenges initiative.

The mechanics matter. Congress appropriates money to NSF, not to OSTP. OSTP is a policy office with a small staff and a correspondingly small appropriation; it does not run grant competitions and has no statutory grantmaking authority of its own. If OSTP priorities are being funded with NSF R&RA dollars, the funds are nominally still flowing through NSF awards — but the topic selection, and in practice the portfolio design, is happening at OSTP.

On August 6, 2026, House Science Committee Ranking Member Zoe Lofgren and Rep. George Whitesides sent a letter to OSTP Director Michael Kratsios demanding a briefing. Their accounting of the diversions:

• Nearly $1 billion clawed back from NSF directorates
• $300 million specifically pulled from the Mathematical and Physical Sciences and related directorates
• A $1.5 billion project they described as OSTP's "brainchild"
• A $5 billion figure announced by Kratsios himself

Their stated objections were three: that the funds were not appropriated to OSTP; that prior testimony to the Committee about where NSF's money was going may have been misleading; and that the arrangement bypassed the Committee's oversight jurisdiction. They set a response deadline of August 14. The letter also characterized NSF as running "30-40% behind where it was last year at this time on getting money out the door" — a figure the final FY2026 numbers have now essentially validated.

We take no position on the legal questions. What matters operationally is simpler and not in dispute: a very large block of NSF research money is being allocated through a channel that is not a published solicitation with a deadline, a panel, and a posted set of review criteria.

The directorate-level damage is not evenly distributed

If you are deciding where to submit in FY2027, aggregate numbers are the wrong altitude. The directorate-level picture is far more lopsided.

The Directorate for Mathematical and Physical Sciences had made roughly 550 awards in FY2026 against an FY2021–FY2024 average of about 1,350 at the same point in the year. For the physical sciences specifically, grant counts ran about 31% below the 2000–2024 average with less than a month left in the year. MPS also absorbed a disproportionate share of the $300 million diversion Lofgren and Whitesides flagged.

The Directorate for Engineering saw obligations of $20 million in the two months following January 20, 2025, against $133 million in the same window a year earlier. The biology directorate absorbed a roughly $200 million budget cut. Hundreds of individual program budgets were reduced by up to 30%.

The strategic read: the slowdown is not an across-the-board haircut you can plan around with a uniform discount. It is a reallocation. Some programs were cut 30% and some were zeroed in practice while the agency simultaneously held back $1.7 billion for a differently-governed portfolio.

Compare NIH, which did the opposite

NIH makes a useful control case, because it faced similar political conditions and produced a different outcome.

NIH spent essentially its entire $34 billion extramural research budget in FY2026 — roughly $35.2 billion in total obligations, about 1% above its five-year average and level with FY2025.

And yet NIH's award count fell too: 60,924 awards through September 24, against an FY2021–FY2025 average of 68,561 at the same date, and 63,303 for all of FY2025. That is roughly 7,600 fewer awards, about 11% below the five-year average, and the lowest count at this point in six fiscal years.

Scott Delaney of Grant Witness, who has tracked obligation rates all year, attributes the NIH gap primarily to a mechanism change rather than a money change: "NIH's move toward multiyear funding is a major reason" for the drop in grant numbers. Fully funding a five-year award up front consumes the same dollars in fewer transactions. Delaney's national tally across agencies: "We're down 15,000 grants."

These are two genuinely different failure modes, and they call for different responses.

At NIH, the money went out. The number of entry points shrank. If you are an early-career investigator seeking a first award, that is still bad news — forward-funding concentrates resources in existing multiyear commitments and reduces the number of new starts — but the budget is being spent on peer-reviewed science selected through published competitions.

At NSF, the money did not go out, and a large share of it is now being directed through a channel you cannot find in the solicitation list. Those require different planning assumptions.

DOE is the third data point, and it points the same direction

The Department of Energy announced its FY2027 Office of Science annual open call at up to $400 million, down from up to $500 million in the prior cycle — a 20% reduction in the broad, investigator-initiated lane that covers the seven Office of Science programs and work falling outside the office's narrower topical notices. The announcement frames the funding as delivering on the "Restoring Gold Standard Science" executive order.

We covered both cycles in detail — the current call in DOE DE-FOA-0003665 and the prior one in DE-FOA-0003600. The pattern across all three agencies is consistent: the open, investigator-initiated, bottom-up lane is contracting, and directed or priority-aligned portfolios are growing.

What this means for your FY2027 strategy

Six things follow from the above, in rough order of how much they should change your behavior.

1. Stop treating NSF's FY2027 budget as the number that governs your odds. The pool available to published FY2027 solicitations is the FY2027 appropriation plus whatever portion of the $1.7 billion carryover is actually released to those solicitations, minus the roughly $1 billion reserved for Grand Research Challenges. Those are three independent unknowns. A proposal plan built on "NSF has more money next year because of carryover" is making an assumption the evidence does not support.

2. Check the program's actual FY2026 award count before you write, not the solicitation's anticipated number. Anticipated awards in a solicitation are drafted months ahead and frequently did not survive contact with FY2026. NSF's award search lets you filter by program element and date. If a program made six awards in FY2026 against twenty the year before, that is your real denominator regardless of what the solicitation says. This single check is the highest-value hour in the whole process and almost nobody does it.

3. Read the Grand Research Challenges topic areas as a framing signal even though you cannot apply to them directly. You cannot submit to an OSTP portfolio. But an agency holding back $1 billion for administration-aligned priorities is an agency whose program officers know exactly what language the building rewards right now. Framing legitimate work in terms that map to stated national priorities is not pandering; it is the same translation exercise every successful proposal has always done. The difference is that the priorities are currently moving faster than the solicitations.

4. Call the program officer, and ask about carryover specifically. The question to ask is not "will my project be funded" — nobody can answer that. It is: "Is this program element expecting to run its FY2027 competition at FY2026 volume, FY2025 volume, or something else? And is any carryover expected to flow to it?" Program officers generally know the answer and are generally willing to say. One call can save you a 15-page proposal into a program that is not going to make awards.

5. Diversify across the mechanism, not just the agency. The structural shift favors (a) large directed programs with named priorities, (b) center and institute mechanisms, and (c) fast-turn, lower-dollar instruments like EAGER, RAPID, and supplements to existing awards. It disfavors the single-PI, moderate-dollar, bottom-up core-program award — which is exactly what most of the community is organized to write. If your entire FY2027 plan is three single-PI core submissions, you are concentrated in the one lane that is shrinking fastest.

6. Protect your non-federal diversification now, because the authorization environment is about to get noisier. The OMB uniform grants rule — which would require senior political appointee pre-issuance review of every discretionary grant and let agencies terminate awards "for convenience" when priorities shift — was frozen by continuing resolution until December 11, 2026. We walked through what to do with that window in The Reprieve Has a Date on It. If that rule takes effect roughly as proposed, the discretion demonstrated by the FY2026 carryover becomes codified default procedure rather than an exception requiring a letter from the House Science Committee.

The one-sentence version

NSF's FY2026 money did not disappear on October 1 — it was carried forward, and a large share of it is being allocated by a policy office rather than a solicitation. For FY2027 that means the open competitive pool is smaller than the appropriation implies, the directorate you target matters more than it used to, and the single most useful thing you can do before writing anything is look up how many awards your target program actually made last year.

Sources: AIP FYI, Science Policy This Week, Oct 5, 2026 · Lofgren and Whitesides letter to OSTP · The Scientist, Federal Research Grant Awards Fall to a Six-Year Low · DOE Office of Science FY2027 open call · AAU, NSF Is Making Fewer Grants · CRS R48783, NSF FY2026 Appropriations