Bottoms-up research sprint complete (July 2026). Figures are working hypotheses for diligence — not booked ARR and not an external rate card. ASP assumptions stay internal until design-partner validation (Pricing). Primary sources: Sources. Cross-check context (not the model): Market analysis · Sources · ICP.
Headline (base case)
| Layer | Definition | Accounts | Working ASP / ACV | Annual $ |
|---|---|---|---|---|
| TAM | Global spend opportunity for third-party verification of robot policies and physics QA of generative / synthetic physical worlds | ~1,080 | $36k mature recurring | ~$39M |
| SAM | Reachable ICP for Haga’s current product (sim-first, private eval, NA/EU-primary GTM) | ~430 | $18k near-term blended | ~$8M |
| SOM | Year-3 obtainable revenue at five-person team capacity | ~25–35 won accounts | $25k blended | ~$0.8M |
How to say this on a slide: sell into the $27.6B physical-AI capital ecosystem as a thin verification layer — bottoms-up verification spend is tens of millions today, not billions. Do not quote robot-software CAGRs as Haga’s TAM.
Sensitivity: TAM low/high ~$27M–$54M (ACV $25k–$50k). SAM near-term ~$6M–$11M. SOM year-3 ~$0.5M–$1.5M.
Method (bottoms-up, not report CAGR)
- Build a buyer universe from countable proxies (funded companies, producers, pure-play world-model / synthetic-data sets).
- Apply ICP and product filters (sim-first private eval; exclude free institutional leaderboards as substitutes, not as buyers).
- Attach working ASP / ACV anchored to adjacent WTP (Sim2Real pilots) and Haga’s packaging hypotheses — labeled as assumptions.
- Keep SOM capacity-constrained (founder-led delivery), not “% of TAM.”
What we explicitly do not do: take MarketIntelo’s $5.8B world-models number (or robot-software $24–35B) and claim a percent as Haga TAM. Those stay directional ecosystem context in Sources.
Step 1 — Buyer universe (TAM accounts)
Counted vs judgment. Rows tagged Counted are third-party tallies. Rows tagged Judgment are ICP filters on those tallies — the main diligence risk until a named account list exists.
A. Policy / manipulation verification buyers (Pillar 1)
| Proxy | Kind | Figure | How we use it |
|---|---|---|---|
| PitchBook physical AI / robotics VC (2025) | Counted | 1,009 deals / $27.6B | Deal count ≠ unique companies; capital is concentrating (mega-rounds). Stock of relevant funded orgs after multi-round noise: ~750–900 → midpoint ~800. Crunchbase’s narrower robotics set (~$13.8B / 2025) warns PitchBook’s category is broad — do not equate deals to buyers |
| RAISE Observatory robotics sample | Counted | 1,279 startups (of 5,056 companies) | Order-of-magnitude check for venture-stage robotics (through mid-2024) |
| Crunchbase robotics hub | Counted | ~9,765 orgs | Breadth warning only (OEMs, integrators, dormant). Not a TAM denominator |
| IFR service-robot producers | Counted | 944 | Enterprise / compliance adjacent later; most outside near-term ICP |
Filter to Pillar 1 relevance (teams shipping or evaluating learned policies / controllers that could buy a sim stress report):
| Segment | Kind | Estimate | Rationale |
|---|---|---|---|
| VC-backed physical-AI orgs with ML policies | Judgment | 380 | ~45–50% of ~800 funded orgs; exclude pure hardware / commodity AMR without policy ML |
| University + corporate robot-learning labs that can procure | Judgment | 150 | Conservative global set of labs with manipulation / foundation-model programs |
| Enterprise robotics groups needing documented validation (later) | Judgment | 400 | Thin slice of IFR producers + industrial OEMs with ML teams — in TAM, mostly out of near-term SAM |
| Pillar 1 subtotal | ~930 |
B. World-model / synthetic physics-QA buyers (Pillar 2)
| Proxy | Kind | Figure | How we use it |
|---|---|---|---|
| New Market Pitch pure-play WM financings (Aug 2024–Jul 2026) | Counted | 23 unique companies / 30 deals / $5.88B | High-intent physics-QA buyers; narrow but capital-rich |
| Tracxn synthetic-data generation | Counted | ~60–63 companies (~59 active Apr 2026) | Broader synth-data set; only a subset is physical/embodied |
| Qbit / 36Kr China embodied-data survey | Counted | 97 domestic players | Real density; mostly outside v0 English-primary GTM → light TAM weight, out of SAM |
| Adjacent generators (no “WM” label) | Judgment | +80 | Labs / robotics teams shipping synthetic video — avoids under-counting non-pure-plays |
Pillar 2 unique buyers (deduped): ~180 (23 WM + ~60 synth + partial China/adjacent, minus double-counts).
C. Dedup and TAM account total
Some frontier labs are both policy and world-model buyers (e.g. robotics foundation-model companies). Deduct ~30 overlap (Judgment).
| Accounts | |
|---|---|
| Pillar 1 + Pillar 2 | 930 + 180 = 1,110 |
| Less overlap | −30 |
| TAM accounts | ~1,080 |
Step 2 — Working ASP / ACV (internal)
Rule: these are modeling inputs for the dataroom, not prices to quote externally (Pricing).
| Anchor | Signal |
|---|---|
| Sim2Real public pilots | $499–$2,500/mo (~$6k–$30k ARR) for training-loop SaaS — proves WTP for sim-to-real tooling, not an audit comps set |
| Haga packaging hypothesis | Price the report + turnaround, not GPU hours; pilot → package → continuous scoring |
| Motion | Working range | Model point |
|---|---|---|
| Single scoped pilot | $10k–$18k one-time | $12k |
| Near-term blended annual (1–2 pilots / light retain) | $15k–$22k | $18k (SAM) |
| Mature recurring (multi-eval or continuous) | $30k–$50k ACV | $36k (TAM) |
| Continuous scoring (Phase 5 upside) | $48k–$96k | Not in base SOM |
Step 3 — TAM
TAM ≈ 1,080 × $36,000 ≈ $39M
Interpretation: at mature category adoption, independent verification of policies + generative physical worlds is a tens-of-millions annual software/services market — a thin layer on a much larger physical-AI capital and robot-software stack. That thinness is the point: category is early; capital intensity around Haga is high; verification spend lags.
Cross-checks (not substitutes for bottoms-up):
- Interactive physics simulators ≈ 31.5% of $5.8B AI world-models (2025) ≈ $1.8B — Haga is a QA slice of that stack, not the simulator market.
- Robot software reports cluster ~$24–35B (2025–2026) depending on vendor — verification is a point solution inside that, not a share we invent.
Step 4 — SAM
Start from TAM accounts; keep only what matches current ICP + product + GTM (Judgment filters on Counted proxies):
| Filter | Effect |
|---|---|
| Sim-first private eval (pre-Robocurve / pre-RoboArena) | Drops pure real-hardware-only seekers |
| Seed → growth startups + buyable labs | Drops mega-labs that only use free institutional leaderboards and pre-product hobbyists |
| NA / EU / UK / Canada / ANZ primary outreach | Drops most China-only embodied-data players; consistent with NA/EU capturing the majority of disclosed pure-play robotics capital (e.g. New Market Pitch regional trackers) |
| Exclude enterprise continuous-compliance until Phase 5 | Moves ~400 enterprise TAM accounts out of SAM |
| SAM segment | Accounts |
|---|---|
| Policy startups (ML, reachable, sim stress need) | 280 |
| World-model / synthetic builders (physics QA need) | 90 |
| Buyable academic / corporate labs | 60 |
| SAM accounts | ~430 |
SAM ≈ 430 × $18,000 ≈ $7.7M (cite as ~$8M)
Upside SAM if continuous scoring lands for a material share: 430 × $36k ≈ $15.5M.
Step 5 — SOM (36 months, capacity-bound)
SOM is not a percent of SAM. It is what a small team can sell and deliver after Phase 4 design partners.
| Year | Won accounts (cum.) | Blended $/account | Year revenue |
|---|---|---|---|
| 1 | 5–8 pilots | $12k | ~$60k–$100k |
| 2 | 15–20 | $18k | ~$270k–$360k |
| 3 | 25–35 | $25k | ~$0.6M–$0.9M |
Base SOM (year-3 revenue): ~$0.8M
Conservative ~$0.5M · Aggressive ~$1.5M (earlier continuous scoring + higher ASP).
Design-partner proof path: Pipeline (target list live; Phase 3 outreach unlocked) · Evaluation offering.
What would change these numbers
| Update | Impact |
|---|---|
| Design-partner WTP locked | Replace $12k / $18k / $36k with measured ASP; refresh SAM/SOM first |
| Multi-task + generative-failure evidence published | Raises conversion assumptions inside SOM, not TAM account counts |
| Phase 5 continuous scoring live | Moves SAM toward $36k ACV; expands enterprise SAM |
| China GTM or partner channel | Adds SAM accounts from the 97-player embodied-data set |
| Robocurve / NVIDIA ships cheap narrow sim checks | Compresses WTP; keep SOM conservative |
Deck one-liner
TAM ~$40M · SAM ~$8M · SOM ~$0.8M (Y3) — bottoms-up verification spend; ecosystem capital ($27.6B physical AI, 2025) is the demand backdrop, not the TAM.
Source index for this page
| Claim | Source |
|---|---|
| $27.6B / 1,009 physical AI deals (2025) | PitchBook via Mean CEO / Embodied Global — see Sources |
| Crunchbase narrower robotics ~$13.8B (2025) | Mean CEO physical-AI stats — category-breadth cross-check only |
| 1,279 robotics startups / 5,056 companies | RAISE Observatory, Chronicles of Robotics Investments (sample through mid-2024) |
| ~9,765 Crunchbase robotics orgs | Crunchbase robotics hub — breadth warning only |
| 944 service-robot producers | IFR World Robotics 2025 Service Robots executive summary |
| 23 pure-play WM companies / $5.88B (24 mo) | New Market Pitch — World Model Startup Funding (2026) |
| ~60–63 synthetic-data companies | Tracxn — Synthetic Data Generation |
| ~97 China embodied-data players | Qbit survey via 36Kr (Jul 2025–Jul 2026 window) |
| Sim2Real $499–$2,500/mo | Public pilot materials — Pricing |
| AI world models $5.8B; interactive physics ~31.5% | MarketIntelo — Sources |
Sprint date: July 2026. Re-run after first paid design partners. Named ICP account list: Design-partner pipeline — still hypotheses until outreach and WTP land.