SBIR as a non-dilutive path for an AI-enabled RMA venture — feasibility & eligibility

BLUF. SBIR is real, sizeable non-dilutive R&D money — roughly $4B+/yr across 11 agencies, with per-award ceilings near $323K (Phase I) and $2.15M (Phase II) as of April 2026 [Public: sbir.gov, 2026]. An AI-enabled reverse-logistics/RMA-automation pitch has genuine precedent and topic pull, but almost all of it sits inside DoD/DLA/Air Force sustainment and depot-repair topics — the same defense/compliance direction TBD killed for the commercial market in 2026-05 compliance_wedge_killed. Three things gate us before any of this matters: we are pre-incorporation [Interview: Grant/Cooley calls, 2026-07-01 & 07-12], SBIR requires a registered US small business plus a 6-8+ week SAM.gov/UEI/SBA registration onramp [Public: sbir.gov / NIH SEED, 2026], and our only internal SBIR seed (Dasher, 2026-04) predates the RMA pivot and is framed around the now-dead compliance/national-security angle 2026-xx-richard-dasher-all. The anchor is our NVIDIA reverse-logistics/RMA engagement — a multi-billion-dollar trapped-inventory problem we’re documenting [Interview, NDA — figure withheld]. This brief surfaces the evidence; it does not draw a go/no-go. Synthesis is the founders’ call.

Timeliness flag (read first). SBIR/STTR lapsed on 2025-09-30, freezing >$4B for ~5 months; it was reauthorized through FY2031 by the Ernst-Markey deal, passing the Senate 2026-03-03 [Public: CSIS, 2026; Granted AI, 2026]. Solicitation cadence is still normalizing in mid-2026. Any timeline below assumes agencies are back to steady-state, which is not yet fully confirmed.


1. What SBIR is

SBIR (Small Business Innovation Research) is a federal set-aside program that channels a fixed slice of large agencies’ external R&D budgets to US small businesses as non-dilutive awards — grants or contracts, no equity taken [Public: sbir.gov/about, 2026].

  • 11 agencies run SBIR (each with extramural R&D budgets over $100M): DoD, HHS/NIH, DOE, NASA, NSF, USDA, Commerce (incl. NIST/NOAA), DHS, DOT, EPA, Education [Public: sbir.gov/about, 2026; CRS R43695].
  • Set-aside %: agencies over the $100M threshold must allocate 3.2% of extramural R&D to SBIR (rising toward ~3.45% for FY26 per reauthorization) and 0.45% to STTR [Public: CRS R43695; SSTI, 2026]. Total program ≈ $4B+/yr [Public: CSIS, 2026].
  • SBIR vs STTR: STTR is the sibling program requiring a formal partnership with a research institution (university/FFRDC), which must perform ≥30% of the work; the small business does ≥40%. Only 5 agencies run STTR (DoD, DOE, HHS, NASA, NSF) [Public: sbir.gov, 2026]. STTR is relevant to us only if we deliberately partner with a university (e.g., a Stanford tie via Dasher) — otherwise SBIR is the path.
  • Two solicitation models:
    • Topic-driven (DoD, DARPA, AFWERX, DLA, NASA): the agency publishes specific problem topics; you must map to one. This is where reverse-logistics/depot-repair money lives.
    • Open-topic (NSF “America’s Seed Fund”): you pitch any technology area; NSF is agnostic to the specific application and evaluates technical risk + commercial potential [Public: NSF/sbir.gov, 2026].

Three-phase structure [Public: sbir.gov, 2026; getcada.com, 2026]:

PhasePurposeTypical durationFunded by SBIR?
Phase IFeasibility / proof-of-concept6-12 monthsYes
Phase IIFull R&D / prototype development~24 monthsYes
Phase IIICommercialization / deployment / govt procurementn/aNo — must be private or non-SBIR federal $

Phase III is the payoff for defense-track players: it can be a sole-source (no re-competition required) government contract derived from prior Phase I/II work [Public: Army SBIR / business.defense.gov, 2026]. But Phase III carries no SBIR funding — SBIR pays only for the R&D of Phases I-II.


2. Dollar amounts & what the money is for

Current guideline ceilings (SBA, as of April 2026) — the amount an agency can award without seeking SBA approval [Public: sbir.gov/about, 2026]:

Phase I ceilingPhase II ceiling
SBA guideline (no waiver)$323,090$2,153,927
Statutory inflation-adjusted base~$314,363 (15 U.S.C. 638(j)(2))
Typical actual range$150K-$323K$750K-$2.15M

Agency variation [Public: sbir.gov, grantsights.com, sledai.com, 2026]:

AgencyPhase I typicalPhase II typicalNotes
NSF~$305K (up to 12 mo)up to ~$1M+Open-topic; Project Pitch gate first
NIHnear ceiling; waivers to $400K+can exceed with SBA topic listDirect-to-Phase-II exists
DoD (services)~$250-295K band~$1.25M-$2MTopic-driven; monthly cadence
AFWERX (Air Force)~$1.25M+Direct-to-Phase-II common
DARPAwaivers to $400K+high on priority topicsDirect-to-Phase-II exists

Waivers/exceptions: NIH and DARPA waive caps most often; SBA maintains an approved “topics list” that lets specific projects exceed the hard caps [Public: sbir.gov, matter-labs.com, 2026]. Direct-to-Phase-II (D2P2) — skipping Phase I — is offered at DoD/AFWERX/Army/NIH, useful if you already have feasibility evidence (we arguably have early feasibility signal from the NVIDIA engagement, though that’s commercial, not a prior SBIR) [Public: AFWERX D2P2 FAQ, 2025].

What SBIR funds: R&D — feasibility studies, algorithm/prototype development, technical de-risking [Public: sbir.gov tutorial-1, 2026]. What SBIR will NOT fund: sales, marketing, market-scaling, pure deployment/rollout of an already-working product, or general working capital [Public: sbir.gov, getcada.com, 2026]. This matters for framing: an RMA pitch has to be posed as unsolved technical R&D (e.g., novel AI for repair-vs-replace triage under uncertainty), not “we built RMA software, fund our go-to-market.”


3. What it takes to win

Eligibility checklist (13 CFR Part 121, Subpart A) [Public: eCFR 13 CFR 121; NIH/NIA, 2026]

RequirementRuleTBD status
Legal entityRegistered US small business concern (for-profit)pre-incorporation [Interview, 2026-07-12]
Size500 employees incl. affiliates✅ (2 founders)
Ownership>50% (≥51%) owned/controlled by US citizens or permanent residents✅ Bliss + Dustin both US-based [Interview, 2026-07-12]
Principal InvestigatorPI’s primary employment (>50%) must be with the small business during the award⚠️ depends — if founders keep day jobs/consulting, PI rule bites
Place of performanceWork performed in the US✅ likely

The ownership rule is why entity structure matters: a standard LLC or C-corp owned by two US individuals qualifies cleanly. Complications arise only with foreign ownership or majority VC/PE/hedge-fund ownership (see §4).

The registration onramp — the hidden lead time

Before you can submit anything, you must complete a registration chain, in order [Public: NIH SEED / sbir.gov / NSF Seed Fund, 2026]:

  1. SAM.gov → issues a 12-char UEI (Unique Entity ID). This is the bottleneck: 4-6+ weeks, sometimes longer if data is incomplete.
  2. SBA Company Registry → issues an SBC Control ID (1-2 business days).
  3. Login.gov + Grants.gov (and agency portals like DoD’s DSIP or NSF’s system).

You cannot start any of this without a legal entity and an EIN. Net: budget 6-8+ weeks of pure administrative onramp after incorporation, before a first proposal is even possible.

Proposal components & win rates

Proposals generally require: technical volume (the R&D plan), commercialization plan, budget/cost proposal, and PI/team qualifications; NSF adds a Project Pitch pre-screen [Public: nsfproposal.com, sbir.gov, 2026].

Win rates (2025-2026) [Public: Granted AI / SBIRLand / NIH databook, 2026]:

Agency / trackPhase I acceptancePhase II conversion (from successful Phase I)
NIH~15-18%40-55%
NSF~20-25%40-55%
DoD (general)~15-25%40-55%
AFWERX D2P2~8-10% (>2,500 subs)n/a

Phase I is a coin-flip-to-long-shot; Phase II conversion is materially better if Phase I is won. AFWERX D2P2 is the most competitive of the common tracks.


4. Strings attached

Data rights / IP (the good news). Under FAR 52.227-20, the company keeps ownership of IP and technical data generated under SBIR. The government gets a limited, non-exclusive license for government purposes only and is barred from disclosing your SBIR data during a protection period. The protection period is 20 years from award (the 2019 revision; note the FAR clause text still shows an outdated 4-year figure not yet updated by the FAR Council, so check the actual funding agreement language before signing) [Public: acquisition.gov FAR 52.227-20; amadeolaw.com, 2026].

Reporting / accounting. Awards carry reporting, milestone, and — especially for DoD contract-type awards — cost-accounting and potential DCAA audit obligations. This is real administrative overhead a 2-person pre-co is not set up for [Public: team-80.com DoD accounting guide, 2026].

Valley-of-death gating. There is a gap between Phase I completion and Phase II award (re-competition, review lag, funding availability). Many awardees stall here; Phase II is not guaranteed even after a strong Phase I [Public: sledai.com, mjpsbirconsulting.com, 2026].

Foreign ownership. >50% foreign individual ownership disqualifies you (post-reauthorization scrutiny of foreign ties, esp. China, has increased — a key Ernst reform driver) [Public: Crowell & Moring SBIR 101, 2026; CSIS, 2026].

Majority-VC/PE/hedge-fund exception. Normally an SBC must be >50% owned by individuals. A statutory exception (15 U.S.C. §638(dd), from the 2011 reauth) lets NIH, DOE, NSF award up to 25% of SBIR funds — and other agencies up to 15% — to firms majority-owned by multiple VCOCs/hedge funds/PE, provided no single such fund holds >50% [Public: sbir.gov/vc-ownership-authority; GAO-24-107004, 2024]. Directly relevant to TBD’s undecided LLC-vs-C-corp / consulting-vs-venture-backed fork: if we later take majority VC money, SBIR eligibility narrows to specific agencies and requires the VCOC certification.


5. Timeline

End-to-end, assuming a clean run and steady-state agencies:

StageDurationNotes
Incorporate + EINdays-weeksmust happen first
SAM.gov / UEI / SBA registry / Login.gov6-8+ weeksthe real long pole; do before any solicitation opens
Wait for solicitation windowvariesDoD: monthly topic releases (new topics ~1st Wed, close ~last Wed of following month, ~6-8 wk write runway). NSF: rolling Project Pitch windows, 3-4/yr, each ~3 wks open [Public: sbir.gov/NSF, sciencedocs.com, 2026]
Proposal writing4-8 weeks
Agency review~2-6 monthsNSF Project Pitch decision ~3-4 wks (up to 2 mo); full review longer
Award → funds disbursedweeks-monthsdisbursement is notoriously slow; DoD contract awards can lag [Public: mjpsbirconsulting.com, 2026]

Realistic first-dollar horizon from a standing start (today, pre-incorporation): ~6-12 months, and that’s if a well-fit topic opens soon. Faster paths: NSF’s rolling Project Pitch (no waiting for a topic) and DoD/AFWERX D2P2 (skip Phase I) compress the front end. Slowest: anything gated on a specific DoD topic that isn’t currently open, compounded by the post-lapse cadence still normalizing in 2026.

Contrast: DoD = topic/rolling, monthly cadence, most reverse-logistics money, but slowest disbursement. NSF = fixed-ish rolling windows, open-topic (you don’t wait for the right topic), civilian/dual-use, generally cleaner for a commercial company.


6. Fit to an AI-enabled RMA / reverse-logistics venture — with defense-track kill-flag

The honest finding: the money that fits reverse-logistics/RMA best is overwhelmingly in DoD sustainment/depot topics — i.e., the defense direction we killed for commercial. Each mapping below carries an explicit kill-flag. This is evidence for the founders, not a recommendation.

Real precedent (named awards — this is not speculative)

  • Vibronyx, Inc. — AFWERX SBIR Phase II (~$1.25M) for an AI-powered Digital Supply Chain Twin of the C-130 platform (predictive maintenance, contingency planning) [Public: vibronyx.com, 2025]. Closest public analog to a supply-chain digital twin SBIR — but it’s Air Force sustainment. 🚩 defense.
  • DLA “Enterprise Digital Thread with Mission-Aware Decision Intelligence” — Phase II AI-enabled analytics/real-time processing for logistics readiness, integrating with enterprise systems [Public: DLA SBIR / sweetspotgov, 2026]. 🚩 defense.
  • Army SBIR — field-level maintenance & repair of weapon-systems electronics — explicitly aims to shorten supply-chain latency for electronic component repairs and improve turn-around-time [Public: armysbir.army.mil, 2026]. This is reverse-logistics/depot-repair R&D almost verbatim — and it’s a weapons-system topic. 🚩 defense.
  • Army “Contested Logistics and Sustainment” topic family; Army AI/ML decision-support (~$250K) and ~$6M AI/ML award tranches [Public: armysbir.army.mil, 2026]. 🚩 defense.

Agency/topic mapping

Agency / topic areaFit to AI-RMAKill-flag vs 2026-05 compliance/defense wedge
DoD depot maintenance / sustainment (Army, Navy, AF)Strongest — depot repair = reverse logistics🚩 DEFENSE. Direct conflict with killed wedge. Founder call.
DLA (supply-chain resilience, digital thread)Strong🚩 DEFENSE. Same.
DARPA (supply-chain resilience)Medium-strong🚩 DEFENSE. Same; also the exact “national-security/supply-chain-resilience” framing Dasher used pre-pivot 2026-xx-richard-dasher-all.
AFWERX (digital supply-chain twin — see Vibronyx)Strong, proven🚩 DEFENSE. Same.
NSF (open-topic; AI, autonomous logistics, digital twins)Medium — the only clean civilian doorNo defense conflict. Would require reframing RMA as a general commercial-manufacturing/logistics AI R&D problem, not defense sustainment.
Commerce / NIST (advanced manufacturing, supply-chain, digital twins for civilian industry)Medium✅ Civilian. Smaller SBIR footprint; worth a targeted search.

Read of the landscape: the reverse-logistics/depot-repair topic gravity is defense. The only SBIR doors that don’t reactivate the killed wedge are NSF (open-topic) and possibly Commerce/NIST (civilian manufacturing/supply-chain) — and both would require framing the RMA pitch as commercial-industrial AI R&D, decoupled from any national-security/compliance narrative. Per compliance_wedge_killed and CLAUDE.md: a govt/defense angle is flagged for the founders, not reframed by the agent.

Additional friction with our thesis: SBIR funds R&D, not deployment. Our internal RMA work is explicitly labeled EXPLORATORY and a departure from the sanctioned financialization thesis — the RMA data pipeline is positioned as a feeder for the financialization wedge (see the internal RMA horizontal scan (NDA), scrap-pile-rma-target-plan, and financialization-primer-2026-05-29). An SBIR award would tie 12-24 months of R&D effort to the exploratory branch, not the documented wedge. That’s a resource-allocation question for the founders, not a fit question.


7. Are we eligible? — internal reality check

Eligibility (mechanical): we would qualify once incorporated — 2 US-based founders, well under 500 employees, US-individual-owned [Interview: Grant/Cooley, 2026-07-01 & 07-12; Ilana Stern/Peterson, 2026-07-01]. Nothing structural disqualifies us. But several gaps are live:

  1. Entity gap. TBD is pre-incorporation as of 2026-07-12. The NDA was signed by the founders individually, not as an entity [Interview, 2026-07-12]. SBIR requires a registered small business + EIN + the full SAM/UEI/SBA chain before applying. Nothing can start until incorporation. [Synthesis]

  2. LLC-vs-C-corp timing. The plan is to form an LLC (“Dustin and Bliss Enterprises,” a partnership/investment vehicle) first and an operating company later; LLC-vs-C-corp is undecided, hinging on consulting-engagement vs venture-backed [Interview: Grant/Cooley, 2026-07-01 & 07-12]. SBIR can be won by either an LLC or a C-corp, but the applicant must be the operating small business, not an upstream investment vehicle. If the entity that does the R&D is the LLC-holding-vehicle rather than an operating company, that needs a counsel check. And the VC-ownership fork (§4) directly interacts with the venture-backed path. [Synthesis]

  3. PI primary-employment risk. SBIR requires the Principal Investigator’s primary employment (>50%) be with the awardee during the award. If founders remain in day jobs or run parallel consulting engagements, this rule bites and must be planned around. [Public: NIH/DOE, 2026] + [Synthesis]

  4. Exploratory-thesis risk. RMA is internally labeled a departure from the sanctioned financialization wedge (see the internal RMA horizontal scan (NDA) and scrap-pile-rma-target-plan). Committing to an SBIR (12-24 mo of R&D) deepens the bet on the exploratory branch. [Synthesis]

  5. NDA constraint. Our best feasibility evidence — the anchor NVIDIA reverse-logistics engagement — is NDA-bound and signed individually (NVIDIA in-person debrief, 2026-06-25 — NDA). An SBIR proposal cannot disclose NVIDIA-confidential specifics, and using them even in aggregate risks the customer-provenance rule customer-provenance. We’d have to build the technical/feasibility case from public evidence, not the anchor’s protected numbers. [Synthesis]

  6. Defense-buyer pull. The strongest topic fit re-opens the defense/compliance direction killed 2026-05 compliance_wedge_killed. Taking DoD/DLA/AFWERX SBIR money means the government becomes an R&D customer and a de-facto product direction — the exact pull we chose to walk away from for the commercial market. The clean-door alternative (NSF/Commerce) is narrower and requires civilian reframing. [Synthesis]

  7. Stale internal precedent. Our only internal SBIR mention (Dasher, 2026-04-08) predates the RMA pivot and frames SBIR around national-security/supply-chain-resilience — the compliance angle we’ve since killed 2026-xx-richard-dasher-all. That framing is no longer our direction and should not be reused. [Interview: Dasher, 2026-04-08]


Confidence summary

TopicEpistemic state
SBIR structure, agencies, phasesHigh — primary sources (sbir.gov, CRS, eCFR)
Award amounts / caps (2026)High — sbir.gov/about April-2026 figures
Eligibility rulesHigh — 13 CFR 121, NIH/agency pages
Data rights / 20-yr protectionMedium-High — FAR clause text lags the 2019 20-yr revision; verify per-award
Program lapse & reauth statusMedium-High — well-reported (CSIS, SSTI, trade press); post-lapse cadence still normalizing
Win ratesMedium — mostly consultancy blogs; NIH databook corroborates the band
RMA/reverse-logistics precedentMedium-High — named awards (Vibronyx, DLA, Army) are concrete; NSF/Commerce civilian precedent for RMA specifically is thin and under-searched
TBD eligibility / entity fitMedium — clear on mechanics; entity structure undecided, needs counsel

Open questions (tiered, with named people)

Tier 1 — blocking, resolve before any SBIR move

  • Which entity applies, and when do we incorporate it? Does the operating company (not the LLC holding vehicle) need to exist first? → startup counsel: Grant/Cooley (prior 2026-07-01 & 07-12 calls), or a dedicated SBIR-experienced firm (Gunderson, KBA).
  • Does the LLC-vs-C-corp / consulting-vs-venture-backed fork create the majority-VC-ownership problem (§4), and does that constrain us to NIH/DOE/NSF? → Grant/Cooley or Gunderson/KBA.
  • Can either founder satisfy the PI >50%-primary-employment rule given day jobs/consulting? → founders + counsel.

Tier 2 — direction / thesis (founder synthesis, per RDI)

  • Does pursuing SBIR — where the money is mostly defense — count as reactivating the killed compliance/defense wedge, or is an NSF/Commerce civilian-only path a genuine exception? → founders’ call; flag per compliance_wedge_killed.
  • Does an SBIR R&D commitment on the exploratory RMA branch pull resources from the sanctioned financialization wedge? → founders.
  • How do we build a fundable technical/feasibility case without NVIDIA NDA specifics? → founders + counsel on NDA scope.

Tier 3 — research gaps to close before drafting a proposal

  • What NSF and Commerce/NIST civilian SBIR topics/awards exist for reverse logistics, returns/warranty AI, or supply-chain digital twins (as opposed to defense sustainment)? Our search found strong defense precedent (Vibronyx/DLA/Army) but thin civilian precedent — this is the highest-value next search. → Dasher for non-defense/academic tracks and NSF navigation; he’s our only internal SBIR-literate contact, though his prior framing is stale 2026-xx-richard-dasher-all.
  • Post-lapse: which agencies have actually reopened normal solicitation cadence in mid-2026? → check agency portals directly before committing to a timeline.

This brief surfaces evidence and open questions. It does not recommend a go/no-go. Synthesis is a human activity — see rdi-methodology.


Sources — external

Sources — internal

Anchor (NDA — figures withheld, cited generically, not wikilinked to preserve public classification): NVIDIA in-person debrief 2026-06-25; NV task-force RMA pain-points session 2026-07-12; internal RMA horizontal scan 2026-07-13. Public/internal cites: 2026-06-23-nvidia-repair-flow-economic-model; Lonny Orona May 2026; Richard Dasher Apr 2026; scrap-pile-rma-target-plan; financialization-primer-2026-05-29; compliance_wedge_killed; customer-provenance; rdi-methodology