First Frame Foundation Grants

ONR's Long Range BAA is the standing front door to Navy and Marine Corps science and technology funding, and it is the vehicle most AI and autonomy researchers should be using rather than waiting for a topic-specific call. It accepts proposals on a rolling basis - the FY25 announcement, N00014-25-S-B001, runs through 30 September 2026 at 2:00 PM EST - and covers the full ONR portfolio, with AI-relevant priorities concentrated in autonomous maritime systems, human-machine teaming, and machine learning for sensor fusion and decision support. ONR's 2026 science and technology strategy tied programs to explicit transition paths, which changes what a competitive proposal looks like: the naval-unique problem underneath the AI method now has to be legible, and a general-purpose ML advance with a thin naval framing reads poorly. Firms with commercially mature AI, autonomy or sensing products are often better served by acquisition and rapid-fielding vehicles, reserving LRBAA proposals for genuine research questions. Practically, the process runs white paper first then full proposal on invitation, and identifying the right ONR program officer before writing anything is the single highest-leverage step - the BAA lists technical points of contact by division and proposals that arrive without prior program officer contact rarely advance. Seven amendments to the FY25 BAA were issued, so working from the current amended document matters.

MFAI is the program to target if the research question is why modern AI works rather than what it can be made to do. NSF funds mathematical and theoretical foundations of artificial intelligence here, explicitly including the capabilities and limitations of foundation models, generative models, deep learning, statistical learning and federated learning, alongside the development of mathematically grounded design principles for current and next-generation AI systems. That framing has become considerably more consequential than it was when the solicitation first appeared: the empirical success of large language and diffusion models has outrun theoretical understanding of them, and a program funding rigorous accounts of foundation model behavior - expressivity, generalization, scaling, sample complexity, the mechanics of in-context learning - sits directly on that gap. Awards are $500,000 to $1,500,000 over 36 months with up to 15 awards per competition against roughly $8,500,000 in annual new-award funding, which puts the effective success rate in ordinary NSF core-program territory rather than the long odds of a center competition. This is a joint MPS and CISE program, and proposals that are genuinely bilingual - mathematically serious and connected to real machine learning practice - do better than those legible to only one of the two communities. A practical eligibility constraint catches some applicants off guard: PIs, co-PIs and senior/key personnel must hold a tenured or tenure-track position, or a primary full-time paid appointment in a research or teaching position, at an eligible U.S. organization, which excludes many postdoctoral and soft-money researchers from leading proposals. The October 9, 2026 deadline continues an established annual cycle on approximately October 10, so groups that miss it have a predictable next opportunity, and the recurring schedule makes MFAI worth building a multi-year strategy around rather than treating as a one-off.

SBIR Ignite is NASA's commercially oriented SBIR track and it inverts the usual federal calculus: instead of asking a small business to bend its roadmap toward a NASA mission need, Ignite explicitly targets high-growth, product-oriented startups that have never worked with NASA and asks them to keep their own commercial strategy while NASA de-risks the early, high-risk technology development that private investors will not touch. That framing is the reason it is worth a separate look from the main NASA SBIR Appendix A solicitation. Phase I runs up to USD 225,000 and Phase II up to USD 1,275,000, all non-dilutive as part of America's Seed Fund. The subtopic set is small and unusually AI-heavy for a space agency: AI-Powered Design for Multidisciplinary Space Hardware, Advanced Manufacturing and Robotics for Space Applications, and Low-Cost Radar for Planetary Navigation and Autonomy. The first of those is the clearest fit for machine learning teams working on generative or surrogate-model design tools, and the third rewards teams doing perception and autonomous navigation in GPS-denied environments. Because Ignite scores commercial potential heavily, a proposal that reads like a research plan tends to underperform one that reads like a product plan with a credible non-NASA customer base. Note that NASA is moving its overall SBIR/STTR programme from a traditional annual solicitation cycle to a Broad Agency Announcement structure, so the mechanics of the surrounding programme are in flux even though Ignite retains its own window.