Two Renewed, One Left to Die: Inside NSF's Quiet Restructuring of America's AI Institutes

Three of the original seven National Science Foundation AI Research Institutes reached the end of their five-year awards this year. Two of them are still running. One is spending down a no-cost extension and preparing to close.

That single sentence describes the state of America’s flagship federal AI research program in September 2026 — and it is a very different program than the one announced with fanfare in 2020, when the National AI Research Institutes initiative promised to knit universities across the country into a durable AI research backbone. The network today spans roughly 29 institutes, more than 500 collaborating institutions, and over $500 million in NSF investment. But the pipeline that kept it growing has quietly gone dark, and the institutes that survive do so by winning a competition most of the public has never heard of.

What just happened

In September 2026, NSF renewed the Artificial Intelligence Institute for Advances in Optimization (AI4OPT) with $20 million for a second five-year phase running through 2031, beginning October 1, 2026. The Georgia Tech-led institute, whose first phase produced more than 250 publications and 12 technology transfers, will reorganize around new thrusts covering AI for optimization, optimization for AI, solver development, multiagent learning, and responsible AI for decision-making — with a new emphasis on real-time and interactive optimization aimed at energy systems, supply chains, and manufacturing. Lead PI Pascal Van Hentenryck remains at Georgia Tech, with co-PIs Bistra Dilkina at USC and Alper Atamtürk at UC Berkeley. USC’s share alone rose to $5.6 million, more than double its phase-one funding.

In June 2026, NSF had already renewed the MIT-led Institute for Artificial Intelligence and Fundamental Interactions (IAIFI), raising annual support from $4 million to $4.98 million — roughly $24.9 million across phase two. IAIFI’s renewal marks a sharp intellectual pivot: phase one demonstrated that machine learning can accelerate physics discovery; phase two inverts the arrow toward what the institute calls the “physics of AI,” using physical reasoning and physical tools to understand and improve AI systems themselves. Its collaboration widened to include MIT, Harvard, Northeastern, Tufts, and Boston University.

And in 2025, NSF declined to renew AI2ES — the University of Oklahoma-led institute for trustworthy AI in weather, climate, and coastal oceanography — after roughly $20 million over five years. The institute has one year to spend remaining funds under a no-cost extension, after which it effectively ceases to exist.

The door for new entrants is shut

The most consequential fact is not any single renewal decision. It is that the National AI Research Institutes program page currently lists no upcoming due dates, and its status reads “waiting for new publication.” The most recent solicitation, NSF 23-610, covered the FY2024 and FY2025 competitions and has not been reissued.

For teams outside the network hoping to build a new institute, there is no pending deadline to prepare against and no announced reissuance timeline. For the institutes already funded, renewal is not one path among several. It is the path.

NSF’s own guidance makes clear that renewal is a competition, not a continuation with paperwork. The PI is expected to open the conversation with the cognizant program officer during the fourth funding year. The renewal proposal is submitted with the fourth annual project review, and no later than the end of October of the fifth funding year, following the Traditional Renewals procedures in Chapter V of the Proposal & Award Policies & Procedures Guide. It goes to external peer review — and reviewers may be handed the results of the institute’s annual evaluation reports. Decisions normally arrive within six months.

That clause about annual evaluation reports deserves emphasis. Those reports are typically treated as compliance artifacts — filed, acknowledged, forgotten. NSF is telling institutes they may become the evidentiary record on which a $20 million decision gets made. Four years of routine reporting quietly becomes the résumé.

Notably, the original solicitations did not promise renewal at all. NSF 20-503 and its successors created five-year awards; the renewal opportunity was offered afterward, competitively. Nothing in the architecture guarantees a phase two — as AI2ES learned.

What the survivors did differently

Read the two renewal announcements side by side and a pattern emerges that has little to do with which science is more exciting.

Neither institute proposed a continuation. AI4OPT proposed a reorganization — new thrusts, a new emphasis on real-time optimization, named application domains in grid complexity under renewable integration and logistics resilience. IAIFI proposed an inversion — from “AI for physics” to “the physics of AI.” Both argued, in effect, “the field moved, and here is our next problem,” backed by a countable phase-one record and a widened institutional coalition. IAIFI’s evidence was overwhelmingly about people: eight postdoctoral fellows trained, three of whom secured faculty positions; an interdisciplinary PhD program that has awarded 20 doctorates since 2021; nearly 600 applications to its 2026 summer school.

Both also expanded their education footprints rather than defending them. AI4OPT is extending its Seth Bonder computational and data science camps into the Los Angeles area alongside K-12 programming, teacher training, and faculty development.

The contrast with AI2ES is stark precisely because that institute’s science was not the issue — its work on trustworthy AI for weather prediction overlapped with some of the most consequential applications of the technology. Director Amy McGovern described what ends when the money does: “We had seven universities involved, and people working together that wouldn’t normally work together producing amazing new results, and that’s going to stop.” Co-PI Philippe Tissot framed the stakes differently: “When you see the rest of the world, our competitors in other countries are not slowing down.” An NSF spokesperson offered only that the agency is “deeply grateful for the groundbreaking work of all AI Institutes and remain committed to building on their success,” noting that “additional award actions remain possible, subject to appropriations.”

Where the money went instead

The federal AI research dollars did not evaporate; they redistributed into instruments with different shapes and eligibility. NSF has been building out infrastructure and access programs — the NAIRR operations center, which received a five-year $35 million cooperative agreement to become permanent; the AI datasets program under NSF 26-512; and the TechAccess coordination hubs under NSF 26-508 — while the $1.5 billion foundational research reset reshaped how core CISE and MPS proposals get submitted at all.

A team that would have written an institute preliminary proposal is now better served assembling a portfolio across those instruments than waiting for a solicitation with no announced return.

Why it matters

Two dynamics make this story bigger than grant accounting. First, it is unfolding against a global race: while the US lets a weather-AI institute expire, competitors — including China, which now trains and deploys frontier models at scale — are not slowing down, as Tissot put it. Weather and climate AI in particular has become a strategic field, with AI forecasting models now beating traditional numerical weather prediction on key metrics.

Second, the pattern is a preview of how federal science funding contracts in an age of fiscal pressure. The program does not announce that it is closing; it just stops posting due dates on one door while keeping peer review running on the other. For the remaining institutes in the network — those funded in 2021 and 2023 whose renewal clocks are already running — the lesson from September 2026 is unambiguous: a countable phase-one record and a phase-two proposal that names a new problem, rather than extending the old one, are what separate a renewal from a shutdown.