"Sagebrush" opens AI inference compute to researchers nationwide — NSF AI grant $5M (UT Austin, NAIRR)
NSF awarded about $5 million to "Sagebrush - An AI Inference Resource for ACCESS, NAIRR, and the Nation" at the University of Texas at Austin. Made under the National AI Research Resource (NAIRR), it provides compute for AI inference to researchers across the country.
Grant overview (primary data)
- Award amount$4,999,956
- RecipientUniversity of Texas at Austin (Texas)
- ProgramNAIRR-Nat AI Research Resource
- Period2026-09-15 〜 2028-08-31
- FunderU.S. National Science Foundation (NSF) / NSF
Key points
- NSF Award 2537075, "Sagebrush - An AI Inference Resource for ACCESS, NAIRR, and the Nation," to UT Austin (TX).
- Program: NAIRR (National AI Research Resource) — opening compute and data access to researchers nationwide.
- The subject is compute for AI inference, a different demand from model training.
- Obligated $4,999,956, equal to the estimated total. Period: September 15, 2026 to August 31, 2028.
- No abstract is included in this dataset; see the official NSF page for detail.
- AI computing divides into training, one enormous calculation, and inference, many small ones; this award addresses the second.
1Inference, not training
Demand for AI compute divides roughly in two: training, which builds a model, and inference, which uses a finished model to produce answers. Training needs enormous computation at once; inference runs many small computations repeatedly. The name of this award points squarely at the second. Even a researcher who never trains a model needs inference compute to apply an existing model to their own data.
2Two frameworks named: ACCESS and NAIRR
The title names ACCESS and NAIRR. NAIRR, the National AI Research Resource, is the effort to open access to compute and data for researchers across the country; this site also covers the NAIRR legislation by the same name. Seeing the debate that builds the framework and the award that actually provisions the resource side by side in public data is part of the value of reading these sources together.
3A two-year term
The period runs two years, September 2026 to August 2028, and the obligated amount equals the estimated total ($4,999,956). Unlike center-type awards obligated in stages across many years, the full plan is already funded. The specific plan is not in this dataset, which carries no abstract, so consult the official NSF page (Award 2537075).
4Training and inference need different resources
Demand for AI computing divides in two, and what this award addresses is the second.
A researcher who never builds a model still needs inference to apply an existing one to their own data. Public money going there reflects an intent to widen the base of people who use AI, not only those who build it.
Why it matters
Public funding directed at inference compute points toward a widening base of AI users. Because what resources an institution can reach shapes the terms of collaboration, the build-out of national frameworks such as NAIRR is useful background when considering industry-academic partnerships.
FAQ
How do training and inference differ?
What is NAIRR?
Why are obligated and estimated amounts the same?
Sources (primary)
Source: NSF Award Search (U.S. National Science Foundation, public domain). Amounts are the obligated amount. For privacy, we do not handle principal investigator names.
- NSF Award (original, official)
- NSF Award ID: 2537075