NSF's $100 Million AI Infrastructure Hubs: A New Consortium Model for Compute Access, and Why November 4 Is the Date That Matters

On August 4, 2026, the National Science Foundation announced a program with an unusually clear theory of the problem. Modern science increasingly runs on artificial intelligence, and AI runs on compute — GPUs, high-bandwidth interconnect, curated data, and the specialized staff who know how to wire all of it into a working research workflow. But that infrastructure has concentrated in a small number of already-elite institutions and the largest technology companies. A materials scientist at a regional public university, a genomics lab at a primarily-undergraduate institution, a community-college data program — these sit outside the frontier not because the science is weaker but because the compute is out of reach. The NSF State and Regional Artificial Intelligence Infrastructure Hubs program, solicitation NSF 26-513, is a $100 million bet on closing that gap through a structure NSF has not leaned on this heavily before: state and regional consortia that pool public and private resources to build and operate shared scientific compute.

This is the deep dive on how the program is built, who can compete, and how to assemble a proposal that survives contact with the review panel — because the consortium model rewards a very different kind of preparation than a standard single-PI grant.

The numbers, precisely

The program's structure is specific enough to plan against:

• Total funding: approximately $100 million
• Anticipated awards: up to 10
• Award size: $4 million to $12 million per hub
• Duration: five years
• Instrument: cooperative agreement (not a standard grant)
• Full proposal deadline: November 4, 2026, then the first Wednesday in November annually
• Letters of intent / preliminary proposals: not required
• Cost sharing: voluntary committed cost sharing is prohibited

That last point deserves immediate attention because it is easy to misread. "No voluntary committed cost sharing" does not mean the partners bring nothing. It is the opposite of a lean single-investigator award: the entire design assumes that state and local governments, universities, philanthropies, and industry supply the actual hardware and operations, while NSF funds the coordination and human layer. The distinction NSF is drawing is that you should not formally pledge matching dollars in a way the agency then has to track and enforce — but the partner contributions are the whole point of the model. A proposal that shows up without serious, documented partner commitments has missed the thesis.

The catalytic funding model — read it carefully

NSF has been explicit that its role here is catalytic, not comprehensive. The agency is not buying everyone a data center. It is funding "the people and coordination layers" — the AI infrastructure professionals with the technical expertise to help researchers actually apply compute to their science — while the consortium and its partners supply and operate the machines. White House OSTP Director Michael Kratsios framed the logic around scale that no single actor could reach: pooling resources across research institutions, governments, philanthropy, and the private sector to accomplish what "no individual stakeholder, and no federal program, could achieve alone."

For applicants, this reframes what a winning proposal is. You are not primarily proposing to buy infrastructure. You are proposing to organize a durable regional coalition, staff it with the right expertise, and demonstrate that the combined contributions add up to real, sustained compute access for a defined research and education community. The federal dollars are the seed and the connective tissue; the hardware story has to come from your partners.

Who can lead, and who should be at the table

Eligible lead organizations are broad by design:

• U.S. institutions of higher education — including two-year and four-year colleges and, notably, community colleges
• Non-profit, non-academic organizations — museums, observatories, research labs, professional societies

Each hub is organized as "a flexible state or regional consortium of state and local governments, research institutions, philanthropies, and the private sector." There is a one-proposal-per-institution limit, and individuals are capped at one PI/co-PI role per deadline — which means institutions cannot hedge by fielding multiple competing bids, and prominent researchers cannot spread themselves across several consortia. That scarcity forces early coordination: within a state or region, the serious players need to converge on a single, strongest coalition rather than fragment into competing proposals that split the field.

The community-college eligibility is a genuine signal, not boilerplate. NSF explicitly ties the program to workforce development and regional job markets, and StateScoop's reporting emphasized that each hub is expected to partner with regional industry to align infrastructure investment with local workforce needs. A consortium that pairs a research university's compute demand with a community-college pipeline for training AI infrastructure technicians is telling exactly the story the solicitation is written to reward.

Three goals, and how the review will weigh them

The solicitation names three program goals, and a strong proposal has to speak to all three rather than optimizing one:

1. Expand access to AI computing infrastructure across regions. The equity-of-access core: who currently cannot get compute, and how does your hub change that concretely?
2. Accelerate AI-enabled scientific discovery. The science payoff: which research communities, which disciplines, which specific problems become tractable once the compute exists?
3. Develop a skilled technical workforce trained in AI infrastructure and science applications. The durability layer: who operates and sustains this, and how does the region build the human capital to keep it running past year five?

The through-line is sustainability. A five-year cooperative agreement that leaves nothing standing in year six is a poor investment, and the catalytic framing makes clear NSF wants coalitions that outlast the federal money. Proposals that treat workforce and governance as afterthoughts to a hardware wish-list will read as fragile.

The industry angle

The private-sector role is structural, not decorative. NVIDIA has publicly joined the program, and the consortium design explicitly invites industry contributions of hardware, cloud credits, and expertise. This is where applicants should be thoughtful rather than opportunistic: a vendor relationship that amounts to a logo on a slide is weak, while a partnership that brings committed compute capacity, real technical staff, and an aligned workforce-training component is exactly the kind of contribution that makes the "no single actor could do this alone" argument credible. Regional industry partners also anchor the workforce goal — the jobs the trained technicians move into should be real jobs at real regional employers.

How to compete — a build plan before November 4

With no letter of intent and no preliminary proposal, there is no soft on-ramp. The full proposal on November 4 is the only shot this cycle, and the annual first-Wednesday-in-November cadence means missing it costs a full year. A realistic plan:

1. Convene the consortium now, not in October. The binding constraint is not writing — it is assembling and aligning partners across government, academia, philanthropy, and industry, and getting documented commitments. That is months of relationship work, and the one-proposal-per-institution limit means you must also win the intra-region competition to be the proposal before you can win the national one.
2. Document partner contributions without pledging formal cost share. Letters of collaboration, in-kind hardware commitments, cloud-credit allocations, and staffing pledges tell the resource story. Structure them as genuine contributions, not as the prohibited voluntary committed cost sharing.
3. Name your research communities and your compute deficit specifically. Vague "broaden access" language loses to a proposal that quantifies who is currently locked out, what they would compute, and what the hub changes in measurable terms.
4. Make workforce and sustainability first-class. Build the community-college pipeline, the technician-training track, and the post-award operating model into the core narrative — and connect the trained workforce to named regional employers.
5. Treat it as a cooperative agreement. These carry substantial NSF involvement during performance — reporting, milestones, and ongoing coordination. Budget the administrative capacity to manage that relationship, not just to run the science.

The strategic context

NSF 26-513 lands amid a wave of federal AI-for-science investment — from the $380M programmable cloud laboratories under the Genesis Mission to the DARPA/NSF AI Forge — and against a broader federal-grants backdrop where posted opportunities have contracted and windows have compressed. What sets the Infrastructure Hubs apart is the deliberate distribution of access. Rather than concentrating compute at the frontier, it pushes resources into states and regions and asks local coalitions to build durable capacity. For a regional university, a community college with an ambitious data program, or a state economic-development office trying to anchor an AI ecosystem, this is one of the clearest on-ramps NSF has offered — provided the consortium starts forming today, because November 4 is closer than it looks.

Granted tracks federal AI and research infrastructure funding across NSF, DOE, and DARPA. Browse active grant opportunities to monitor how compute-access and AI-for-science programs are evolving.