NSF's Biggest AI Funding Door Closes September 10 — and CISE Has Made 474 Grants This Year Against an 838 Average

The second Thursday of September is the single most important date on the calendar for American computer science researchers, and almost nobody outside the field notices it. On September 10, 2026, the fall panel cycle closes for Future CoRe — NSF 25-543, the Computer and Information Science and Engineering directorate's consolidated core research solicitation. It is the highest-volume pathway for foundational AI and machine learning funding in the United States: an anticipated $280 million supporting an estimated 400 to 600 awards across 11 programs.

That is the number in the solicitation. Here is the number that is not in the solicitation: as of August 19, 2026, the CISE directorate had made 474 new grants this fiscal year, against an average of 838 at the same point in FY2021 through FY2024. That is a 43 percent shortfall in a directorate whose submission volume has been climbing for a decade on the back of the AI boom.

Both numbers are true. Understanding why is the difference between writing a Future CoRe proposal that has a chance and writing one that was doomed by a budget decision made three levels above the program officer who reads it.

What Future CoRe actually is

Future CoRe is a consolidation, and a fairly aggressive one. It absorbed the prior CISE Core Research solicitation (NSF 24-589), the standalone Cyber-Physical Systems solicitation (NSF 24-581), and Smart and Connected Communities (NSF 25-527) into one document covering 11 program areas across all three CISE divisions:

Three structural changes came with the merger, and each one has a strategic consequence.

The Small and Medium project classes were collapsed into a single class. Where CISE Core previously offered a Small track (roughly $600,000 over three years) and a Medium track (roughly $1.2 million over four years) with separate deadlines and separate competitive pools, there is now one class: up to $1,000,000 over up to four years. NSF's own guidance inside the solicitation is far more modest than the ceiling — typical projects run $150,000 to $250,000 per year for three to four years, and proposals are explicitly discouraged from exceeding $300,000 in any single year.

Read that carefully, because it is the most commonly misread sentence in the document. The $1 million ceiling is not a target. A four-year project at $250,000 per year is exactly $1 million and sits squarely inside the guidance. A three-year project at $333,000 per year hits the same total and violates the annual guidance. The ceiling is a product of duration, not ambition. Reviewers who have internalized the $250,000-per-year norm read a $400,000 first year as a signal that the PI did not read the solicitation.

The OAC Core program was eliminated entirely. Researchers whose work sat in the Office of Advanced Cyberinfrastructure's core program no longer have a core home in this solicitation. That work has to find a fit inside CSR, SHF, or one of the cyberinfrastructure-specific programs elsewhere in NSF — and "elsewhere in NSF" in 2026 is a crowded, contracting place.

Mandatory collaboration plans and broadening participation plans were removed. This is a real reduction in proposal-preparation burden, and it is also a trap. Broader Impacts remains one of the two NSF merit review criteria and has not softened. What changed is that you are no longer told where to put it. Proposals that previously offloaded broader impacts into a required standalone document now have to integrate it into the Project Description, and the ones that simply deleted the section are the ones that get panel comments about a thin broader impacts case.

The two-proposal cap has no exceptions, and it is measured backwards

An individual may not serve as PI, co-PI, or Senior/Key Personnel on more than two Future CoRe proposals submitted within any consecutive 12-month period, across all 11 programs. Exceeding the limit results in a proposal being returned without review. The solicitation's language is unusually blunt: no exceptions will be made.

Two features of this rule cause avoidable damage every cycle.

First, it counts Senior/Key Personnel, not just PIs. A senior faculty member who agrees to a 5 percent effort role on a junior colleague's proposal has spent one of their two slots. In a large department where a productive systems researcher might be named on four or five proposals as a matter of collegiality, this is a live risk, and the person who gets hurt is not always the person who agreed carelessly — NSF returns proposals in reverse order of submission, so the last proposal in is the one that dies.

Second, the window is rolling 12 months, not per-cycle or per-fiscal-year. A PI who submitted two proposals to the February 2026 target date is not clear to submit again on September 10, 2026 — they are clear on February 5, 2027, the first Thursday in February. Proposals withdrawn before review, or returned without review, do not count against the limit, which gives you exactly one recovery move if you catch the problem in time.

Do this before you submit: build a departmental spreadsheet of every Future CoRe submission in the last 12 months and every person named on it in any role. This takes an hour and it is the single highest-return hour in the entire preparation cycle.

"Target date" is not "deadline," and the difference is not what most people assume

Future CoRe accepts proposals at any time. September 10 is a target date, not a deadline — submit by it and your proposal is considered in the corresponding panel cycle; submit after it and your proposal goes into the queue for the next one.

This is part of a wider NSF shift we covered when the agency released 12 solicitations on August 17 that replaced hard deadlines with soft target dates, and again when BIO Core dissolved three divisions into a single directorate-wide opportunity. The standard advice — "no deadline means less pressure" — is exactly backwards.

A hard deadline creates a discrete competitive pool. Everyone submits on the same day, the panel reviews a fixed set, and the funding rate is a clean fraction. A rolling window with target dates moves the competition from the calendar into the queue. Program officers now hold discretion over which cycle your proposal lands in and how a late-arriving strong proposal is weighed against an on-time adequate one. In a year when the directorate is 43 percent below its award pace, that discretion is being exercised under scarcity, and scarcity plus discretion means the proposals that get funded are increasingly the ones a program officer already recognizes.

The practical translation: contacting a program officer before submission has moved from optional courtesy to structural necessity. Not to lobby — to make sure your proposal is routed to the right one of the 11 programs, and to find out whether that program has money this cycle.

The number the solicitation does not tell you

Now the uncomfortable part. NSF's FY2026 budget is roughly $8.8 billion. Approximately $1 billion of it is being held in a central account that core grant-issuing programs cannot reach, with staff reporting that much of it is allocated to a White House initiative called Grand Research Challenges targeting advanced materials and artificial intelligence. In July 2026, NSF clawed back roughly $300 million from two directorates without stated explanation, forcing program officers to withdraw more than 150 proposals that had already been reviewed and recommended for funding.

As of August 19, the agency had issued 5,684 new grants — the lowest total in four decades.

So Future CoRe's "400 to 600 awards" is not a promise. It is a solicitation-standard estimate written against a budget that no longer describes reality. Historical CISE funding rates ran around 22 percent directorate-wide in FY2024, with individual programs dipping to 18 to 20 percent. If submission volume holds steady and the award count runs 40 percent below the four-year average, the effective funding rate for this cycle lands somewhere in the low teens.

There is a second-order effect that matters more than the headline rate. When an agency funds fewer, larger grants — NSF's FY2025 pattern showed average new award size up 15 percent while award count fell 20 percent — the projects that get cut are disproportionately the modest, single-PI, exploratory ones. Future CoRe's guidance range of $150,000 to $250,000 per year is precisely that profile. The solicitation's typical project is the type of project a contracting agency stops funding first.

What to actually do with six days

If you are submitting on September 10, three things are worth the remaining time.

Get the program acronym in the title right. Future CoRe requires the proposal title to begin with the program acronym, a colon, then the project title — RI: Learning Structured Representations for.... Collaborative proposals prepend Collaborative Research: before the acronym. This is not cosmetic. It is the routing instruction that determines which of 11 program budgets your proposal competes against, and picking the program with money is a larger effect on your odds this year than most of what is in your Project Description. If your work sits between RI and III, or between CSR and SHF, call both program officers this week and ask which has funds.

Prepare the Project Personnel and Partner Organizations supplementary document properly. Future CoRe requires a numbered list of full names, organizations, and roles for every PI, co-PI, senior personnel member, consultant, subawardee, postdoc, and advisory committee member. NSF uses it for reviewer selection and conflict management. An incomplete list does not just risk a compliance return — it degrades the panel you get assigned, because NSF cannot exclude conflicts it does not know about.

Decide honestly whether to submit now or in February. If your two-proposal window is tight, or your preliminary results are thin, February 4, 2027 is a real option, and it arrives after the FY2027 appropriations picture resolves. Under the continuing resolution running through December 11, NSF spends the first quarter of FY2027 at a restricted rate of operations with new starts constrained. Proposals reviewed in the fall cycle get funded into that environment. Proposals reviewed in the spring cycle get funded into whatever Congress actually enacts.

That is not a reason to skip September. It is a reason to stop treating the two target dates as interchangeable. The fall cycle is competing for FY2026 leftovers and constrained FY2027 openings; the spring cycle competes for a budget that exists. For a PI with one strong proposal and one adequate one, sending the strong one in February and holding the adequate one entirely is a defensible read of this particular year.

The larger pattern

Future CoRe is what agency consolidation looks like from inside a funding contraction. Twelve programs became one document. Two budget classes became one. Two required plans disappeared. Deadlines became target dates. Each change, taken alone, reads as simplification — and NSF has been consistent that simplification is the goal.

Taken together, they move discretion from the applicant to the agency. The PI used to choose a budget class, choose a deadline, and target a program. Now NSF chooses the cycle, the program routing is a title string the agency can override, and the budget class is a single band with informal norms enforced in review. In a year with a normal budget, that is administratively neutral. In a year when a directorate is running 43 percent below its own average, it means the funding decision has quietly moved upstream of the merit review that the whole system is nominally built around.

Write the proposal anyway. But write it knowing that the most consequential question is not whether it is good — it is whether the program you routed it to has money on September 10.

If you are trying to work out which of the 11 Future CoRe programs your work actually belongs in, or which federal AI and computing deadlines are worth building a team around this fall, Granted can turn a crowded funding landscape into a short list you can act on.