From Pilot to Permanence: NSF's $35M NAIRR Operations Center Puts SDSC and TACC in Charge of America's AI On-Ramp

The most consequential AI infrastructure announcement of the past week didn’t come from a frontier lab. It came from Alexandria, Virginia, in a press release with no keynote and no livestream: the U.S. National Science Foundation has established the NSF National Artificial Intelligence Research Resource Operations Center (NAIRR-OC), awarding a five-year, $35 million cooperative agreement to the San Diego Supercomputer Center (SDSC) at UC San Diego, in collaboration with the Texas Advanced Computing Center (TACC) at UT Austin.

If you don’t follow federal research infrastructure, that sentence can slide past you. But it marks the moment one of the U.S. government’s most-watched AI programs — the National AI Research Resource — began its transition from an experiment into a permanent institution.

What was announced

The NAIRR-OC, funded under NSF award number 2616251, will serve as the operational backbone of NAIRR, the public-private effort that connects the U.S. research and education community with advanced AI resources: computing, datasets, models, software, training, and technical expertise.

Concretely, the Operations Center will:

• Coordinate resource providers — the private-sector, federal, and academic organizations that contribute compute, data, and tools to the resource
• Integrate the stack — computing, distributed scientific data, AI models, software, and research tools into a coherent whole
• Build and operate a unified national NAIRR portal, giving researchers a single front door to an ecosystem of resources
• Provide user support, documentation, training, and outreach, plus a user forum and feedback channels so the resource evolves with community needs
• Lead the transition of appropriate functions from the NAIRR pilot into long-term national infrastructure
• Identify barriers that keep researchers, educators, and students from fully engaging in AI-enabled discovery

The leadership team is drawn from both institutions: Frank Würthwein, SDSC director, serves as principal investigator and director of the NAIRR-OC; Ashley Atkins, SDSC deputy director, is co-PI and deputy director; and Maytal Dahan, Subhashini Sivagnanam, and Mahidhar Tatineni round out the co-PI team from TACC and SDSC.

Why this matters: the pilot worked

The NAIRR Pilot launched in January 2024 with a mandate to prove demand. By NSF’s accounting, it succeeded: over its first two and a half years, the pilot connected roughly 900 research teams and educators across all 50 states with AI resources, supported more than 800 research projects, and enlisted over 25 private-sector partners — including major technology companies and nonprofits — in a public-private model that let academic researchers touch capabilities they could never have funded themselves.

That last point is the entire thesis. Frontier AI research increasingly requires resources — large-scale compute, curated datasets, model access — that exceed what an individual laboratory or university can muster. Without a shared on-ramp, ambitious questions either go unasked or get ceded to whoever can afford the GPU bill. NAIRR is the U.S. answer to that concentration problem, and the Operations Center is the piece that turns it from a demonstration into durable capability.

Why SDSC and TACC

NSF’s choice rewards operational track record over name recognition. UC San Diego’s SDSC operates national-scale cyberinfrastructure and leads the NSF-supported National Data Platform and National Research Platform. UT Austin’s TACC leads the U.S. NSF Leadership-Class Computing Facility — the program behind the nation’s next flagship academic supercomputers — and has spent two decades building the portal software the NAIRR will now run on.

The NAIRR Portal itself will build on TACC’s Core Portal and Tapis frameworks, according to Dahan, who directs advanced computing interfaces at TACC. That’s a pragmatic inheritance: rather than procuring a new front end, the country gets one already hardened by millions of users across existing national systems.

The pairing also reflects a deliberate strategy the NSF statement makes explicit: integrate leading existing capabilities rather than build isolated infrastructure. In a year where sovereign AI ambitions routinely arrive with nine-figure price tags, a $35 million operations layer that knits together existing regional hubs, data systems, and compute is a quietly contrarian bet on coordination over construction.

The connective tissue

The Operations Center won’t operate in isolation. NSF has wired it to three complementary programs that together form a national AI ecosystem:

1. NSF State and Regional AI Infrastructure Hubs — bringing together research institutions, government, industry, and philanthropy to expand regional compute access and workforce development
2. NSF Integrated Data Systems and Services — national-scale data infrastructure for AI-driven research, which the NAIRR-OC will specifically integrate
3. NSF Unlocking Dataset Value for AI-Enabled Scientific Discovery — squeezing new insight out of existing scientific datasets

The through-line is access. Atkins, the NAIRR-OC’s deputy director, tied the center’s mission to the White House’s Genesis Mission — the federal initiative to accelerate AI-driven scientific discovery — arguing that connecting researchers with infrastructure, training, and data builds “a workforce that will not only use AI, but also shape how it advances science, engineering and society.”

The honest caveats

Three qualifications are worth stating plainly.

First, $35 million over five years funds operations, not hardware. The center coordinates and integrates; it doesn’t buy the nation a new GPU fleet. The compute comes from partners — and the generosity of private-sector contributors is a variable, not a constant. If commercial providers’ priorities shift, the resource’s capacity shifts with them.

Second, the pilot’s 800+ projects demonstrate demand, but scale changes problems. A unified portal serving tens of thousands of researchers across 50 states — with identity management, allocation fairness, documentation, and training — is an order-of-magnitude harder operations problem than a pilot cohort. The NSF’s own framing (“lean and sustainable operations capability”) suggests it knows this.

Third, the measurement question remains open. The pilot counted projects and users; the long-term version will need to show that access translates into discoveries, startups, and workforce outcomes that wouldn’t have happened otherwise. The NAIRR-OC’s mandate to “identify challenges” is where that accountability will live.

The bigger picture

For all the attention lavished on model releases and nine-figure funding rounds, the infrastructure that decides who gets to do AI research is being assembled in documents like this one. The NAIRR-OC announcement won’t trend, but a decade from now, the list of researchers who got their first serious compute allocation through a NAIRR portal — and what they went on to build — is a fair candidate for the more consequential legacy.

The transition from pilot to permanence now has an operator, a budget, and a deadline-free five-year runway. The next milestone to watch: whether the unified portal ships on TACC’s frameworks ahead of schedule, and whether the pilot’s private-sector partners re-up for the long haul.