NSF Just Spent $83 Million Building the Plumbing for AI-Driven Science. The Six IDSS Winners, the $10M–$30M Award Ceiling, and Why the Next Deadline Is Already the One That Matters

Every headline about AI for science this year has been about the compute and the models — the $5 billion Genesis Mission, the NAIRR pilot, the GPU allocations. Almost none of it has been about the least glamorous and most decisive layer of the whole enterprise: the data. An AI model for science is worthless if researchers cannot find, access, move, and combine the datasets it needs to learn from. That plumbing — national-scale, interoperable, production-quality data infrastructure — is exactly what the National Science Foundation just put $83 million behind.

On July 22, 2026, NSF announced the first cohort of awards under its Integrated Data Systems & Services (IDSS) program: $83 million to build and operate the data backbone for "data-intensive and AI-driven science, engineering research, innovation and education." It is deliberately positioned as the complement to NAIRR — where NAIRR provides the compute, IDSS provides the data systems that make the compute useful. And because the program runs on an annual cycle, the FY26 winners are already set, the July 28 full-proposal deadline just closed days ago, and the institutions that want in are now competing for the window that reopens next July.

This is the deep dive on what IDSS funds, who won the first round, how the two-category structure and the strict one-proposal rule shape strategy, and what a competitive submission looks like for the cycle ahead.

What IDSS actually funds — and why it's different from a normal NSF grant

IDSS is not a research grant in the ordinary sense. You are not proposing to discover something; you are proposing to build and operate a national-scale service that other researchers will depend on. That framing changes everything about how the program is judged. NSF is buying infrastructure, so it evaluates proposals the way you'd evaluate a utility: Is it reliable? Does it scale? Will it still be running — and staffed, and maintained — in year five? Who depends on it, and what happens to them if it fails?

The program is organized into distinct categories with very different risk profiles:

• Category I — national-scale systems. These are the development, deployment, and operation of novel national-scale integrated data systems and services, which may interface with or leverage existing capabilities. Awards run $10 million to $30 million for up to five years. This is the flagship tier: you are building something new at national scale, and NSF is committing eight figures over half a decade to see it operate.
• Category II — transition to operations. These take an established smaller-scale, regional, pilot, or prototype data system and harden it into national-scale, production-quality, operational-grade service. Category II is the "graduation" path — you've proven the concept regionally, and now you're being funded to make it robust, reliable, and available to the whole country.

The distinction matters enormously for positioning. Category I rewards ambition and novelty; Category II rewards a track record and a working prototype. Choosing the wrong lane — pitching an unproven idea as a Category II transition, or dressing up a mature regional platform as a Category I novelty — is one of the fastest ways to lose.

The six winners tell you exactly what NSF wants

The FY26 cohort is instructive precisely because it is small and legible. NSF funded two national-scale flagship systems and four transition projects:

Category I — National-Scale Systems:

• Morgridge Institute for Research (Madison, Wisconsin) — Fabric for AI-Driven Science (FabAID)
• UC San Diego — National Data Platform (NDP)

Category II — Transition to Operations:

• UCLA — Interactive Discovery Laboratory (iDLab)
• UC Irvine — BRIDGE National Center
• University of Tennessee, Knoxville — National Science Data Fabric (NSDF)
• University of Arizona — Multidisciplinary Environment for Scientific Advancement (MESA)

NSF also issued planning grants to seed future proposals — a signal worth reading carefully, because a planning grant is often the on-ramp to a full Category I or II award in a subsequent cycle.

Look at the pattern. Every winner is building something that makes scientific data discoverable, movable, and combinable at national scale — data fabrics, data platforms, discovery environments. NSF explicitly frames all of it as advancing "America's leadership in artificial intelligence" by giving researchers the data infrastructure to use AI and advanced computing effectively, complementing NAIRR. If your proposal doesn't clearly connect to that mission — enabling AI-driven science across disciplines, not just serving one lab or one field — it is swimming against the program's stated purpose.

The winners also skew toward major research institutions with existing cyberinfrastructure operations. That is not an accident. Running a national-scale data service is an operational commitment, and NSF is understandably wary of funding a great idea attached to an organization that has never run production infrastructure. This is a program where operational credibility is a first-class evaluation criterion, not an afterthought.

The one-proposal rule that quietly decides your institutional strategy

IDSS carries submission restrictions that most PIs discover too late, and they reshape how an entire university has to approach the program.

First, at the individual level: a person may participate as PI, co-PI, or other Senior Personnel on at most one proposal across Categories I and II per deadline. If you're on a Category I proposal, you cannot also be on a Category II proposal. Your name is a scarce resource, and you must spend it on exactly one bet.

Second, at the institutional level: an organization may submit only one proposal as lead for each of Category I and Category II per deadline — though it may be a subawardee on other proposals. This makes IDSS a limited-submission program at most universities, which means there is an internal competition before the NSF competition. Your research office likely runs a pre-proposal down-select to decide which single Category I and single Category II bid the institution will put forward. If you wait for the NSF deadline to start organizing, you will have already lost the internal round.

The strategic consequence is that IDSS planning begins months before the NSF deadline, inside your own institution. The teams that win are the ones that lock in their internal down-select early, assemble the operational partners, and treat the university's limited-submission process as the real first deadline. The July 28, 2026 full-proposal date that just passed was the end of that process for this cycle's winners, not the beginning.

The annual cadence — and why next July is already in play

IDSS runs on a recurring schedule: the full-proposal deadline is the fourth Tuesday in July, annually. The FY26 cycle closed July 28, 2026. That means the next window is July 2027, and — given the multi-month internal limited-submission timelines described above — the competition for it effectively starts now.

Here is how to use the runway.

Pick your lane honestly. Before anything else, decide whether you have a genuinely novel national-scale system (Category I) or a proven regional platform ready to harden into national operations (Category II). Misclassification is fatal. If you have a promising but unproven concept, the smarter move may be to pursue a planning grant first and build the operational evidence that a full award requires.

Win the internal round first. Identify your institution's limited-submission process and its internal deadline — usually several months ahead of July. Get your team, your operational partners, and your letters of institutional commitment lined up to survive the down-select. At most universities, only one Category I and one Category II bid go forward; make sure it's yours.

Lead with operations, not just vision. Because NSF is buying a service, your strongest evidence is a credible operations plan: staffing, uptime, sustainability past the award period, user community, and governance. A proposal that reads like a research idea will lose to one that reads like a business plan for a national utility. Show who depends on the service and what your plan is to keep it running in year six, when the money is gone.

Anchor to AI-for-science explicitly. The program exists to complement NAIRR and advance AI-driven science. Draw the line directly: which AI and data-intensive research communities your system serves, how it interoperates with existing national capabilities, and why national scale — not a regional deployment — is necessary. Cross-disciplinary reach is a feature NSF is actively rewarding.

Consider the subawardee path. If your institution's lead slots are already spoken for, or your capability is a component rather than a whole system, remember that organizations can participate as subawardees on other proposals. Being a strong component partner on a winning national platform is a legitimate — and often faster — way into the IDSS ecosystem than leading your own bid.

The bottom line

IDSS is the quiet, load-bearing layer beneath the entire AI-for-science push. While the attention goes to compute and models, NSF is spending $83 million a cycle to make sure the data those models need can actually be found, moved, and combined at national scale. Category I awards of $10 million to $30 million over five years are among the largest infrastructure commitments NSF makes to a single team — and the six FY26 winners show a clear preference for operationally credible institutions building genuinely national, AI-enabling services.

The FY26 cohort is set and the deadline has closed, but the annual cadence means the real work for the next award starts immediately — inside your own institution, in the limited-submission down-select that determines who even gets to submit. The teams that treat next July as this summer's problem, pick the right category, and lead with an operations plan rather than a research pitch will be the ones NSF trusts with the plumbing of American science.