AI Grant Success Rates by Agency: Where Your Odds Are Best

Four out of five AI grant proposals submitted to federal agencies end in rejection. But that four-out-of-five figure obscures enormous variation depending on where you send your proposal. At NSF, a computer vision researcher submitting to the Information and Intelligent Systems division faces a 15% funding rate. That same researcher, reframing the work as cyberinfrastructure, could land in a division funding one in three proposals. The agency you target and the program you choose within it can matter as much as the science itself.

Browse our AI Grants page for current opportunities across all federal agencies.

NSF: The Best Odds in Federal AI Funding, If You Pick the Right Division

NSF remains the most transparent federal agency when it comes to publishing funding rates, and the FY 2024 numbers for the Directorate for Computer and Information Science and Engineering (CISE) reveal a striking spread. The overall CISE funding rate in FY 2024 was 22% — competitive but workable. Break that number apart by division, though, and the picture changes.

The Division of Information and Intelligent Systems (IIS), which houses core AI programs including machine learning, natural language processing, robotics, and human-AI interaction, funded just 317 of 2,122 reviewed proposals — a 15% success rate. Meanwhile, the Division of Advanced Cyberinfrastructure (OAC) funded 229 of 701 proposals at a 33% success rate, more than double the IIS figure. Computing and Communication Foundations (CCF) landed at 28%, and Computer and Network Systems (CNS) at 22%.

For AI researchers, the lesson is structural. If your work involves building AI-powered infrastructure, tools, or platforms rather than advancing core algorithms, OAC and CCF may offer better odds than the most obvious IIS home. Program officers can help you identify where your proposal fits best — a pre-submission conversation is worth the phone call.

The bigger uncertainty is what happens next. The administration's FY 2026 budget request would slash NSF's overall success rate to roughly 7%, down from 26% in FY 2024, cutting funded research proposals from 9,600 to 2,300. CISE's own AI research budget would take a 16.8% cut under that plan, even as the administration has increased AI investment elsewhere across the agency. Congress has not approved those cuts, and prior-year appropriations suggest lawmakers will preserve more funding than proposed. But the trajectory is worth watching.

NIH: Falling Rates and a Growing AI Appetite

NIH publishes success rates through its RePORT database, and the recent trend is not encouraging. The overall R01-equivalent success rate for established investigators dropped from approximately 27% in FY 2024 to about 20% in FY 2025. Early-stage investigators fared worse: their R01-equivalent success rate fell to 18.5% in FY 2025, down from 29.8% just two years earlier — an 11 percentage point collapse.

None of that is AI-specific, but it is the funding environment that AI-in-health proposals now enter. NIH does not publish a separate success rate for AI-tagged applications, but its AI investment is growing. The agency allocated $309 million in core AI funding for FY 2025, with $3.05 billion going toward the broader IT and data infrastructure that AI research depends on. The Bridge2AI program committed $130 million over four years to generate AI-ready datasets, and AIM-AHEAD continues building AI/ML capacity at under-resourced institutions.

The practical implication: NIH is spending more on AI but funding a smaller percentage of the proposals it receives. If your AI work has a clear biomedical application, the money is there, but reviewers are more selective than at any point in the past decade. Resubmissions have become essentially mandatory for R01-equivalent awards — first-submission success rates are significantly lower than the headline figures.

One policy note that trips up applicants: NIH will not consider applications substantially developed by AI. Using language models to draft your proposal about language models is a disqualification risk.

DARPA: High Rewards, Opaque Odds

DARPA does not publish success rates. It does not publish the number of proposals received per BAA. It publishes Broad Agency Announcements, evaluates white papers and full proposals through an internal review process, and makes awards. The rest is inference.

What we know: DARPA's FY 2025 budget is $1.41 billion, with $314 million directed specifically at core AI research. Active programs include AI Forward, the $2 billion AI Next campaign, the Air Intelligence Reinforcements (AIR) program at $41 million, and the ASIMOV autonomous weapons ethics program at $22 million. The Information Innovation Office (I2O) released its FY 2026 Office-Wide BAA covering AI, cybersecurity, and complex software systems.

What experienced DARPA proposers will tell you is that the two-stage process — abstract or white paper first, then invited full proposal — functions as an aggressive filter. Anecdotal estimates from defense grant consultants put the white-paper-to-award conversion rate somewhere between 5% and 15%, depending on the program. But DARPA's model is fundamentally different from NSF or NIH. You are not competing against a pool of peer-reviewed proposals on a scoring rubric. You are pitching a program manager who has a specific technical vision. If your approach aligns with that vision, you can win with a team of five. If it does not, no amount of preliminary data will help.

DOE: Big Checks, Narrow Windows

The Department of Energy holds $1.54 billion in IT and AI R&D funding, with $187 million in core AI investment for FY 2025. The Office of Science announced $68 million for 11 multi-institution AI-for-science projects, while the Advanced Scientific Computing Research (ASCR) program issued a separate $36 million FOA for foundational AI and applied mathematics research. More recently, DOE committed over $320 million toward AI capabilities through its Genesis Mission initiative.

DOE does not publish competition-wide funding rates comparable to NSF or NIH. But the structure of its awards tells you something about selectivity. The $68 million AI-for-science round funded just 43 awards across 11 projects. The ASCR ALCC program awarded 38 million node hours to 56 projects. These are small, selective pools where institutional partnerships with national labs significantly improve your chances.

For AI researchers in energy, materials science, or climate modeling, DOE's advantage is award size — individual project budgets routinely exceed $1 million. The disadvantage is that FOA windows are narrow, the applicant pool is dominated by national lab affiliates, and there is no standing program equivalent to NSF's CISE Core Programs where you can submit any cycle.

Where to Aim in a Tightening Market

The numbers point toward a few actionable conclusions. NSF's CISE remains the most accessible entry point for core AI research, but the 15% IIS rate means most proposals will be declined — look carefully at whether OAC (33%) or CCF (28%) might be legitimate homes for your work. NIH's falling success rates make AI-in-health proposals harder to fund, but the agency's $309 million AI commitment means the money is real if you can navigate increasingly competitive study sections. DARPA offers transformative funding but requires a fundamentally different approach: relationship-building with program managers rather than peer review optimization. DOE favors large collaborative teams with national lab connections.

Across every agency, the trend is the same: more researchers are submitting AI proposals while funding rates either hold steady or decline. The researchers who win are the ones who treat agency selection as a strategic decision, not an afterthought — matching their work to the program where it fits best rather than defaulting to the most obvious solicitation.

For tracking open solicitations across all of these agencies in one place, Granted can surface AI funding opportunities matched to your specific research profile before deadlines close.