NSF Just Told You How It Will Decide What to Fund for the Next Decade. Four Frontier Initiatives, Golden Tickets, and a Target to Double Output Per Dollar by 2036.

Agencies issue vision documents constantly and almost none of them change anything. The memo Brian Stone — performing the duties of the NSF director — sent to the science and technology community on September 10, 2026 is the exception, because it does not describe priorities. It describes mechanisms. And mechanisms are what determine whether your proposal gets funded.

Titled NSF Leadership in a New Golden Age of Science, the memo announces four inaugural Frontier Initiatives, establishes an Office of Metascience inside the Office of the Director, and commits NSF to experimenting with review procedures that would have been unthinkable at the agency five years ago — including "Golden Tickets" that let a single reviewer push a proposal past a panel that would otherwise reject it.

It also states a target with a number attached: double the scientific productivity and impact generated by each federal research dollar by 2036.

If you write NSF proposals, this document is worth reading in full. If you only have ten minutes, here is what actually changes.

The context the memo does not mention

The memo opens on Vannevar Bush and a "period of extraordinary scientific possibility." The operating environment is considerably less lyrical.

NSF is on pace to issue roughly 6,100 new grants in FY2026 — approximately 46 percent fewer than its 2021–2024 average, and the lowest count in over four decades. As of August 19, 2026, about 5,684 grants had been issued with some 400 pending at the Office of Award Management. Within directorates the picture is sharper: Mathematical and Physical Sciences recorded its lowest award count since the 1980s, astronomical sciences fell roughly 50 percent, and materials research dropped more than 40 percent.

Part of the shortfall is arithmetic, not philosophy. Of NSF's roughly $8.8 billion budget, approximately $1 billion is held in a central account that core grant-issuing programs cannot reach, much of it earmarked for the White House's Grand Research Challenges initiative. Money held centrally is money not flowing through the panels that fund individual investigators.

Read the memo against that backdrop and its logic becomes legible. NSF is not proposing to fund more proposals. It is proposing to change which proposals get funded, through what mechanisms, and to measure the result. The productivity-doubling target is what you say when the grant count is falling and you need a metric that does not depend on it.

The four Frontier Initiatives

NSF has a real tradition here: the National Nanotechnology Initiative (2000), the Materials Genome Initiative (2011), and the BRAIN Initiative (2013) each spanned multiple directorates around a shared thesis. Frontier Initiatives formalize that pattern as standing capacity rather than one-off launches.

Each Frontier Initiative Lead sits in the Office of the Director — not in a directorate. That placement is the whole design. A lead with directorate-level authority can only coordinate; a lead in the Director's office can move resources across directorate boundaries. The Office of Metascience will periodically assess whether each initiative still addresses "a consequential and distinct frontier," and NSF explicitly reserves the right to retire initiatives.

The inaugural four:

1. Future of Intelligence. Fundamental understanding of intelligence — architectures, hardware, and theories of learning that define the next AI paradigm. The memo is unusually explicit that NSF expects that paradigm to come from outside computer science, drawing on physics, mathematics, biology, neuroscience, and engineering, in the way neuroscience seeded neural networks. If you are a physicist or a neuroscientist who has been told your work is not an AI proposal, the agency just told you otherwise.

2. Frontier of Quantum Phenomena. Foundational QIS — the physics, materials, and theory needed to compute, sense, and communicate beyond classical limits, plus the engineering and instrumentation to apply quantum capabilities to other frontiers.

3. Advanced Materials and Matter. Design, discovery, and manipulation of materials underpinning quantum, computing, semiconductors, robotics, fusion, and biomanufacturing — plus early-stage advanced manufacturing methods that close the loop between discovery, synthesis, and characterization.

4. Foundations of Biological Design. Measurement and intervention tools to read, write, and reliably control living systems. The memo's framing is tools-first: gene editing, cheap synthesis and sequencing, and advanced microscopy drove modern biology, and the next generation of tools will draw on engineering, mathematics, chemistry, and materials science.

The strategic read: all four are explicitly cross-disciplinary by construction. That is a genuine signal for a proposal type that has historically fared badly at NSF — the project that does not sit cleanly inside one directorate's review culture and gets penalized by both. The memo names that problem directly, citing the literature on peer review's difficulty with risky and unconventional work, and a 2012 National Science Board finding about panels struggling to evaluate larger-than-typical unsolicited proposals.

Golden Tickets and distributed review

This is the part that will generate the most argument, and the part with the most direct effect on your funding odds.

NSF says it will experiment with proposal review mechanisms, specifically:

• Golden Tickets — "empowering individual reviewers to champion exceptional proposals that may not survive panel consensus."
• Distributed review — to reduce administrative burden and shorten time to decision.
• New measures to identify and support novel research.

The stated goal is to "surface high-potential ideas that may be too divisive for committee consensus and to attract higher-quality reviewers by empowering them to exercise independent scientific judgment."

Understand what a Golden Ticket does to the scoring distribution. NSF panels reward proposals that no reviewer objects to. That selection pressure systematically favors the well-constructed incremental project over the divisive ambitious one — a proposal that three reviewers love and two consider unsound loses to one that five reviewers find adequate. A Golden Ticket inverts that for a small number of proposals per panel: it rewards maximum enthusiasm from one reviewer rather than minimum objection from all.

What that means for how you write. Under consensus scoring, the dominant strategy is defensive — anticipate every objection, hedge every claim, make the proposal impossible to dislike. Under a mechanism where one champion can carry you, the defensive strategy becomes actively costly, because hedging is exactly what prevents a reviewer from falling in love with your idea.

Two cautions before you rewrite anything. First, the memo says experiment, not adopt — these will appear in specific programs, not across the agency, and you will not always know which. Second, NSF explicitly preserves the requirement that "every funded proposal meets a level of scientific rigor appropriate for Gold Standard Science." A Golden Ticket is a route past consensus, not past rigor.

The pragmatic posture: keep writing rigorously defensible proposals, but stop sanding down the one claim that makes your project matter. Make sure a reader who is predisposed to your approach finds something worth championing on page two.

Time horizons: the five-year award and the few-page application

The memo splits research by time horizon and matches mechanism to each.

Long-horizon. NSF will expand opportunities that let exceptional researchers request award durations of five years or longer, on the reasoning that long awards lower administrative burden and give investigators the certainty to pursue directions without incremental outputs. This flexibility is already built into the $1.5 billion in core discovery research funding opportunities released August 17, 2026.

Check those solicitations before your next submission. A five-year request that the solicitation permits and you did not make is a self-inflicted wound — and the differential burden is enormous. Five years of runway instead of three is roughly one fewer full resubmission cycle per decade of a career.

Short-horizon. Building on EAGER (high-risk, high-reward exploratory) and RAPID (time-sensitive), NSF is expanding short-duration grant pathways featuring applications of only a few pages and rapid funding decisions.

This is the most under-appreciated line in the memo. A few-page application with a fast decision changes the economics of trying something. The cost of a conventional NSF proposal — measured in PI weeks — is high enough that researchers only submit ideas they are already fairly confident in. Drop the cost to a few pages and the rational behavior is to test speculative ideas cheaply and often. If these pathways materialize at scale, the researchers who benefit most will be the ones who build a pipeline of small bets rather than treating each submission as a major event.

Degrees of specification, and the coming Grand Research Challenges

NSF will balance investigator-driven inquiry against NSF-directed opportunities in areas of special emphasis. Directed opportunities will range from milestone-based funding tied to demonstrated technical progress to prizes and challenges where NSF specifies a goal in advance.

The memo commits to announcing a series of Grand Research Challenges "in the coming weeks and months" — thesis-driven bets meant to unlock new frontiers. This is where that centrally held roughly $1 billion is headed.

One line deserves quoting because it is the load-bearing constraint on the whole directed-funding turn: "When NSF provides direction, we will specify the objective, never the method scientists must use to achieve it." Whether the agency honors that in practice is the thing to watch as the first Grand Research Challenges are published.

Prizes, challenges, and milestone-based awards are also the mechanisms most likely to favor non-traditional applicants — small teams, companies, independent institutes — because they are scored on demonstrated progress rather than institutional track record. The memo also flags that these will be used "where substantial non-federal participation can be leveraged," which is a cost-share signal in polite language.

The Office of Metascience

NSF will stand up an Office of Metascience (OMS) in the Office of the Director, staffed by metascience experts and data scientists, to develop and rigorously evaluate new approaches to funding, merit review, program management, and tool adoption — scaling what works and discontinuing what does not.

OMS commits to operating with full transparency, publicly reporting the effectiveness of various funding mechanisms and publishing postmortems for both positive and negative results. It will track measures like grant turnaround time and identify duplication and gaps across the portfolio. The memo lists the experimental questions it intends to run prospectively: When do longer awards produce greater scientific ambition? Do rapid exploratory awards surface ideas conventional processes miss? Which approaches give early-career researchers the most independence to produce novel, disruptive work?

Researchers have reacted to OMS with visible skepticism, and the skepticism is reasonable — a self-evaluating office reporting to leadership that set the targets is not an independent evaluator. But treat the transparency commitment as an opportunity rather than a claim to be graded. If OMS publishes mechanism-level effectiveness data, that is the highest-value NSF document you will ever read as an applicant. No agency currently tells you which funding mechanisms actually work. If NSF starts, read every postmortem the day it appears.

AI, the Genesis Mission, and infrastructure

The memo positions NSF at the center of the Genesis Mission, the national effort to apply AI to scientific discovery. NSF has issued a Dear Colleague Letter and written language into core discovery solicitations encouraging proposals that leverage AI and other emerging technologies where scientifically justified.

Note the qualifier. "Where scientifically justified" is doing real work — it is an invitation that comes with a filter, and reviewers will apply it. Bolting AI onto a proposal that does not need it has been a reliable way to lose points at NSF and this language does not change that. What it does change is that AI-enabled methodology is now affirmatively welcome in domains where it previously read as off-topic.

On infrastructure, NSF commits to AI-ready datasets, next-generation scientific data infrastructure, and a national network of programmable cloud laboratories, plus investment in autonomous and self-driving laboratories. Access expands through the National AI Research Resource (NAIRR) and a set of State and Regional AI Infrastructure Hubs that NAIRR will federate. That is NSF 26-513 — the $100 million, up-to-10-award program with a November 4, 2026 deadline and a one-award-per-state-or-region cap, which we covered in NSF's $100M AI Infrastructure Hubs. The memo now places that program inside a much larger architecture, which strengthens the case for applying and raises the bar on what a credible proposal looks like.

Talent: portability and hands-on time

The talent section restates a commitment to merit-based identification of exceptional students and early-career researchers, with two concrete items.

NSF renewed the Graduate Research Fellowship Program on August 25, 2026, and the memo emphasizes its portability — the freedom to choose research questions, mentors, and institutions. The memo also proposes extending REU, RET, and INTERN mechanisms to support hands-on technical training in federal laboratories and industry, and to fund substantial student time on real equipment — cleanrooms, cryogenic systems, advanced microscopes, autonomous labs, HPC facilities, robotics testbeds and fabrication facilities — paired with portable, industry-recognized credentials.

There is also an explicit place-based commitment: NSF will "co-locate, sequence, or jointly award workforce and education investments alongside federal R&D funding," with non-federal partners expected to bring co-funding, paid internships, infrastructure access, curriculum, and mentorship. If you run a workforce program, the message is that a proposal arriving with committed industry and state partners will beat an equally good one that does not.

What to do differently

1. Map your work onto a Frontier Initiative by name. Four labels now sit in the Director's office. Using the agency's own framing is free, and cross-directorate framing just became an asset rather than a liability.
2. Check whether your target solicitation permits five-plus-year durations — and if it does, request it.
3. Watch for short-form pathways and build a pipeline. A few-page application with a fast decision rewards volume of ideas, not perfection of one.
4. Stop over-hedging your central claim. Consensus-proofing is a losing strategy under Golden Tickets.
5. Read the Grand Research Challenges announcements as they publish. That is where the centrally held money goes, and prizes and milestone awards favor applicants without long institutional track records.
6. Do not decorate proposals with AI. "Where scientifically justified" is a filter, not a formality.
7. Read every OMS postmortem. If NSF publishes which mechanisms work, that is the most actionable applicant intelligence any agency has ever released.

The memo's honest subtext is that NSF is trying to get more science out of a shrinking number of awards by changing the selection machinery rather than the budget. Whether that works is an empirical question — which, to the agency's credit, is precisely what it says the Office of Metascience exists to answer.