The National Artificial Intelligence Research Resource (NAIRR) is a US shared research infrastructure, led by the National Science Foundation (NSF), intended to give researchers and educators outside the large frontier labs access to the compute, data, models, software, and training needed to do AI research. Its stated purpose is to widen participation in AI research beyond the well-resourced companies and universities, addressing the compute divide between industry and academia. It launched as a pilot in January 2024, and NSF announced funding for an operations center to scale it in September 2025.
Origin and trajectory
The National AI Initiative Act of 2020 created the NAIRR Task Force to study how to build a shared national AI research resource. In January 2023 the Task Force delivered its final report, Strengthening and Democratizing the U.S. Artificial Intelligence Innovation Ecosystem, recommending a full NAIRR.
In January 2024, NSF launched the NAIRR Pilot, a two-year demonstration run with partner federal agencies (DOE, NIH, NASA, NIST, and others) and non-governmental contributors (major AI companies and compute providers donating model access and cloud credits). The CREATE AI Act, proposed legislation to formally authorize and fund a full, permanent NAIRR, has been introduced but not enacted. In September 2025, NSF announced funding to establish a NAIRR Operations Center (NAIRR-OC) to manage and expand the resource beyond the pilot phase (Source: nairrpilot.org).
By NSF's account, the pilot has supported more than 600 research projects, and NSF describes NAIRR as a "proven national research infrastructure" rather than an experiment (Source: nsf.gov).
What NAIRR provides
The resource bundles several access types into one entry point for researchers:
- Compute — allocations on NSF-funded supercomputers and on cloud and accelerated-compute capacity contributed by partners.
- Data — curated government and partner datasets.
- Models and software — access to open and partner-provided AI models and tooling.
- Training and support — educational resources and user support to lower the barrier to entry.
A large share of NAIRR's compute and model access is donated by frontier labs, the same actors whose concentration of compute the resource is intended to counterbalance.
Role in US AI policy
NAIRR is the US government's primary policy answer to the concern that frontier AI research has concentrated inside a few companies because compute is concentrated there. Whether a pilot-scale resource can meaningfully shift that balance is contested.
It is research-enabling infrastructure rather than an evaluation or regulatory body, distinguishing it from NIST CAISI (Center for AI Standards and Innovation) and the AI Safety Institutes. Some analysts, including in Sociotechnical AI Risk Governance discussions, argue that the existing US AI infrastructure institutions, NAIRR included, have been shaped by private-sector technical priorities. NAIRR is referenced alongside NIST CAISI (Center for AI Standards and Innovation), the Frontier Model Forum, and International AI Safety Institute Network (INSAI / AISIN) as one of the existing US AI infrastructure institutions (see Dominance by Understanding (Frazier-Rozenshtein policy frame) and the May 17 briefing).
State-level public-compute proposals are modeled on the NAIRR concept, including California's CalCompute consortium under SB 53 and New York's Empire AI.
Relationships
- related: NIST CAISI (Center for AI Standards and Innovation), International AI Safety Institute Network (INSAI / AISIN), Frontier Model Forum — other US AI infrastructure / governance institutions.
- related: California SB 53 — CalCompute consortium modeled on the NAIRR concept.
- related: Government AI Procurement, Dominance by Understanding (Frazier-Rozenshtein policy frame).
Sources
- (Source: nsf.gov) — NSF, National Artificial Intelligence Research Resource overview (pilot history, 600+ projects).
- (Source: nairrpilot.org) — NAIRR Pilot news: September 2025 NSF announcement of funding for the NAIRR Operations Center.