NSF SBIR Restart: Project Pitch Tips for AI and Deep Tech Startups

Most federal agencies make you write a 25-page proposal before anyone tells you whether your idea fits. NSF flips that model. Through its Project Pitch system, SBIR Phase I applicants submit a two-page concept — problem, solution, team, market — and get a decision within three weeks. If invited, you write the full proposal. If not, you've lost days instead of months.

With SBIR reauthorization signed and NSF expected to reopen solicitations in April or May 2026, AI and deep tech startups have a narrow window to prepare. The Project Pitch is deceptively simple. Two pages sounds easy until you realize that reviewers spend roughly six minutes on each one, and a vague pitch lands in the reject pile before anyone reaches page two.

How the Project Pitch Actually Works

NSF's SBIR program operates on a rolling submission basis — no fixed deadlines for the initial pitch. You submit through Research.gov, a program director reviews it, and you receive an invite or decline typically within three weeks. Only invited companies submit full Phase I proposals, which are worth up to $275,000 over 12 months.

This two-step structure exists because NSF receives thousands of pitches annually and wants to filter early. The pitch is not a miniature proposal. It is a screening tool. Program directors are looking for three things: a real market problem, a defensible technical approach, and a team that can execute. If all three are present, you get the invite. If any one is missing, you don't.

Unlike DoD SBIR topics, which prescribe specific technical problems down to the subsystem level, NSF topics are intentionally broad. Categories like "Information and Communication Technologies" or "Chemical and Environmental Technologies" leave room for interpretation. That breadth is an advantage for AI startups — your natural language processing tool for clinical trials and your computer vision system for agricultural monitoring both fit under the same umbrella.

Lead With Market Size, Not Technical Novelty

The single biggest mistake deep tech founders make on the Project Pitch is writing it like a conference abstract. NSF SBIR is not a research grant. The program exists to commercialize technology, and reviewers evaluate commercial potential alongside technical merit.

Your pitch should open with the market problem and its dollar value. A statement like "The U.S. predictive maintenance market for industrial equipment exceeds $6.3 billion and is projected to reach $15.9 billion by 2030" immediately tells the reviewer there is a business here. Follow that with why current solutions fail and what your technology does differently.

Customer validation is the strongest signal you can send. Even at the pitch stage, a letter of intent from a potential customer, pilot data from a beta deployment, or revenue from an adjacent product tells the program director that someone outside your lab cares about this technology. NSF explicitly weights "competitive advantage and value proposition" in its review — vague claims about disruption do not satisfy that criterion.

For AI-specific pitches, avoid the trap of describing your model architecture in detail. Reviewers care about what the model does for customers, not how many transformer layers it has. Frame the AI as the mechanism, not the product.

NSF's Two-Phase Review and What It Means for Your Proposal

If your Project Pitch is invited, the full Phase I proposal goes through a rigorous two-phase review. First, a panel of technical experts evaluates the intellectual merit and broader impacts. Second, NSF conducts a commercialization review that scrutinizes your business model, market analysis, and go-to-market plan.

This dual review is where many first-time applicants stumble. A proposal can score well on technical innovation and still be declined if the commercialization plan is weak. NSF publishes its evaluation criteria clearly: Intellectual Merit and Broader Impacts carry equal weight, and within Broader Impacts, commercial potential is the dominant factor for SBIR.

Phase I awards are meant to demonstrate feasibility — not build a finished product. Scope your work plan accordingly. If you propose Phase II deliverables in a Phase I budget, reviewers will flag the disconnect. Phase I should answer the question: "Can this work?" Phase II, which provides up to $1 million, answers: "Can this scale?"

For a deeper look at how the SBIR and STTR programs differ, including which is better suited for university-partnered deep tech teams, see our comparison guide.

The I-Corps Connection and Why It Matters

NSF runs the Innovation Corps (I-Corps) program, a seven-week customer discovery boot camp that has become a feeder for SBIR success. Teams that complete I-Corps demonstrate exactly what NSF wants to see: systematic evidence that customers want what you're building.

You don't need to complete I-Corps before submitting a Project Pitch, but referencing customer discovery interviews — even informal ones — strengthens your case. If your pitch is invited and you haven't done I-Corps, consider applying to a regional I-Corps hub before submitting the full proposal. NSF program directors notice when teams have done the work.

For AI and biotech startups especially, I-Corps forces you to move past the "build it and they will come" mindset. A team that has conducted 30 customer interviews and identified specific pain points writes a fundamentally different proposal than one operating on assumptions.

Preparing Before Solicitations Reopen

The 2026 SBIR reauthorization extended the program through September 2029 and introduced new provisions around foreign risk screening and proposal caps. NSF's specific implementation details — including any changes to the Project Pitch format or topic areas — will be published when solicitations reopen.

In the meantime, the preparation work is concrete. Register in SAM.gov if you haven't already (allow 4-6 weeks for processing). Create accounts on Research.gov and SBA's Company Registry. Draft your two-page pitch using NSF's published format guidelines from the most recent solicitation. Line up at least two letters of support from potential customers or partners.

Review the complete SBIR application guide to understand the full proposal lifecycle, and browse current SBIR opportunities across all federal agencies to see how topic descriptions are structured.

Related SBIR reading:

• Your First SBIR Application in 2026
• NSF SBIR/STTR for AI Startups
• Tips for Writing a Successful SBIR Proposal

The companies that win Phase I are rarely the ones with the most sophisticated technology. They are the ones who clearly articulate why that technology matters to a paying customer — and NSF's Project Pitch system is designed to surface exactly that signal. Tools like Granted can help you identify the right topics and move from a rough pitch to a polished submission before the window closes.