NAIRR Pilot
US federal pilot providing shared AI compute and data to researchers
The National Science Foundation launched this two-year pilot in January 2024 to give US academics access to government-funded AI infrastructure. It pools computing power, datasets and tools from federal agencies and industry partners to support safe and trustworthy artificial intelligence research.
- Launched by NSF in January 2024
- Capacity equals about 5,000 H100 GPUs
- Total compute is roughly 3.77 exaFLOPS
- Supported by 10 federal agencies
- Operations Centre announced in 2026
What it is
This initiative creates a shared national infrastructure for responsible artificial intelligence discovery. The National Science Foundation leads the effort with support from ten other federal bodies and non-governmental partners. It provides selected scholars with high-performance computing, curated data and software tools. The goal is to bridge the gap between well-funded private firms and academic institutions that lack such resources.
How it came about
The Office of Science and Technology Policy and the NSF started the programme in early 2024. It runs as a two-year proof of concept to test how best to distribute federal AI assets. In September 2026, the NSF announced an Operations Centre to manage the system long term. This new hub builds on the initial pilot success and prepares for a full-scale national resource.
The compute
Initial estimates place total processing capacity at 3.77 exaFLOPS. This equals roughly 5,000 Nvidia H100 graphics units when using optimal tensor cores. Major systems include the Summit, Frontera and Delta supercomputers. Availability varies because many machines serve other scientific needs too. The allocated time for AI tasks totals approximately 3.2 yottaFLOPs across the network.
Where the GPUs are
Resources come from existing federal laboratories and university-based advanced computing centres. These locations are spread across the United States. The Department of Energy and National Institutes of Health help develop secure elements of the infrastructure. Partners contribute hardware and data from their own facilities to expand the pool of available tools.
Customers and deals
The primary users are researchers, educators and students in the US public and private sectors. Projects often focus on safe, secure and trustworthy artificial intelligence. Other supported areas include healthcare, environmental sustainability and AI education. Teams from universities and smaller organisations use these tools to tackle problems they could not address alone.
Money and ownership
The programme relies on contributions from federal agencies and industry partners rather than direct sales. Hewlett Packard Enterprise is a named non-governmental partner. The NSF manages the allocation of these donated or shared resources. There is no public listing or revenue model as it serves a public research mission.
Its place in AI
It aims to diversify the field by giving under-resourced groups access to powerful models. This helps ensure a broader range of ideas influences AI development. The pilot addresses concerns that only wealthy companies can afford the compute needed for modern foundation models. It supports work on copyright challenges and privacy-preserving machine learning techniques.
Questions and criticism
Most resources are already near capacity due to non-AI scientific work. The largest system, Summit, faces decommissioning which limits its long-term utility. Access is not year-round for many components. The total compute remains a fraction of what large tech firms possess, so it cannot support every large-scale training request.
Questions people ask about NAIRR Pilot
Who can use the NAIRR Pilot resources?
US-based researchers, educators, students and post-docs in both public and private sectors may apply. The programme prioritises projects focused on safe and trustworthy AI, though it also supports healthcare and environmental studies.
How much computing power does the pilot provide?
The initial six resources offer a total capacity of roughly 3.77 exaFLOPS. This is equivalent to approximately 5,000 Nvidia H100 GPUs when measured using tensor cores optimised for artificial intelligence workloads.
Which agencies are involved in running the programme?
The National Science Foundation leads the effort. Ten other federal partners participate, including the Department of Energy and the National Institutes of Health. Non-governmental partners like Hewlett Packard Enterprise also contribute resources.
Good reading and sources
There is no English Wikipedia article on NAIRR Pilot yet, so the facts on this page come from the sources listed above: nairrpilot.org/, nsf.gov/geo/updates/national-artificial-intelligence-research-resource, cset.georgetown.edu/article/the-nairr-pilot-estimating-compute/. Some are the subject's own pages, so its claims about itself are reported as its claims. EveryNeocloud wrote the text in its own words, with AI assistance, and compares it with the sources to make sure no passage is copied.
It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-14.