Figure AI
general-purpose humanoid robots + embodied-AI (VLA) · vertically-integrated robotics-as-a-service (RaaS)
Series A $500M valuation is per secondary aggregators (Sacra/TSG), not the primary release; Series B/C valuations are company-disclosed. Founder's ~$100M seed was self-funded with no disclosed valuation (omitted).
The thesis on this name
State of Physical AI
The strongest commercial-humanoid evidence in the market: the Helix VLA running Figure 02 on a live BMW Spartanburg line that built 30,000+ vehicles is the first production-scale humanoid deployment in auto manufacturing. $39B Series C (Sept-25) funds a vertically-integrated push (own VLA + own actuators + own BotQ manufacturing). If the humanoid TAM is real, Figure is a top-2 Western contender.
Earnings, margins, COGS & capex
No public 10-K/10-Q — Figure publishes no audited financials. Revenue model is RaaS: ~$1,000/robot/month (hardware + software updates + maintenance/support) per Sacra, with the BMW factory contract reported on a per-operating-hour basis (~$25/robot-hour). Cost base is R&D-heavy and capex-heavy: in-house actuators, batteries, hands/sensors, the Helix VLA model, and the BotQ factory. Production is the leading KPI, not revenue: BotQ went from ~1 robot/day to ~1/hour (24x in <120 days), delivered 350+ Figure 03 units by mid-2026; Apr-2026 shipments (~240) exceeded all of 2025 (~150). Cash runway funded by the Sep-2025 Series C; management's stated priority is revenue before any listing. All figures source-stamped to press/company posts (mid-2026).
Revenue trend
Margins
COGS structure
Not disclosed (private company).
Capex
Not disclosed.
Growth drivers
- Production is doubling — Apr-2026 (~240 units) beat all of 2025 (~150); BotQ proved a 24x throughput ramp to ~1 robot/hour with a path to 12k/yr then 100k over 4 years
- In-house Helix (post-OpenAI exit) removes dependence on a third-party model and lets Figure co-optimize AI + hardware the way Tesla did for autonomy
- BMW Spartanburg is hard proof, not a demo — 30,000+ vehicles, 90,000+ parts, 99%+ placement accuracy — a reference customer that de-risks the next enterprise sale
- Strategic-investor depth (NVIDIA compute, Microsoft/Salesforce enterprise, Brookfield's 100k residential units + $1T balance sheet) gives capital, distribution, and training data few rivals can match
Bull & bear
The best-capitalized Western humanoid with the only verified at-scale factory deployment and a real, fast-ramping manufacturing line — if Helix + BotQ convert into a multi-customer RaaS install base, Figure is the category's Model-T moment.
- Production is doubling: Apr-2026 (~240 units) beat all of 2025 (~150); BotQ proved a 24x throughput ramp to ~1 robot/hour with a path to 12k/yr then 100k over 4 years
- In-house Helix (post-OpenAI exit) removes dependence on a third-party model and lets Figure co-optimize AI + hardware the way Tesla did for autonomy
- BMW Spartanburg is hard proof, not a demo: 30,000+ vehicles, 90,000+ parts, 99%+ placement accuracy — a reference customer that de-risks the next enterprise sale
- Strategic-investor depth (NVIDIA compute, Microsoft/Salesforce enterprise, Brookfield's 100k residential units + $1T balance sheet) gives capital, distribution, and training data few rivals can match
A $39B mark on essentially no revenue, in an unsolved field, against 10x-cheaper Chinese fleets and Tesla — with a fresh safety-whistleblower overhang and a customer that just diversified away.
- $39B exceeds the projected 2035 humanoid TAM on some estimates — the price discounts ~a decade of flawless execution against negligible current revenue
- Nov-2025 ex-safety-chief allegations (skull-fracture-level force, 'gutted' safety roadmap timed to the Series C) are a legal + reputational tail risk on a product that works near humans
- BMW's choice of Hexagon AEON for Leipzig signals Figure is not a locked-in sole supplier; customer concentration + diversification both bite
- Chinese players at ~1/10th price and Tesla's manufacturing machine threaten to commoditize hardware before Figure's RaaS economics mature; home market is years out by the CEO's own words
What it is worth
Last-priced-round + reverse-sanity. Anchor = Series C $39B post-money (Sep 2025). No earnings to multiple; value is an option on production ramp × RaaS install base × Helix capability. Reverse read: $39B implies tens of thousands of profitable RaaS robots and multi-customer adoption within ~5 years — aggressive vs low-tens-of-$M FY26 ARR.
$10–20B
if the safety overhang escalates, China/Tesla commoditize hardware, customers diversify (à la BMW/Hexagon), and the next round/secondary prints flat-to-down on still-negligible revenue.
~$39B held
production ramps but revenue stays small and customer concentration persists; the mark is sustained by category momentum and strategic-backer support, not fundamentals.
$60–80B+
if BotQ hits its 100k/4yr cadence, RaaS diversifies past BMW, and Helix opens home + logistics; the category-leader premium compounds and a 2027-28 IPO marks up the round.
SWOT
Strengths
- Best-funded US pure-play humanoid — ~$1.9B raised, $39B mark, with strategic backers across compute (NVIDIA, Qualcomm, Intel), cloud/enterprise (Microsoft, Salesforce, T-Mobile), and capital (Brookfield, Bezos Expeditions)
- Real commercial deployment — Figure 02/03 on paying BMW Spartanburg lines — 30,000+ X3 vehicles, 90,000+ parts, ~1,250 operating hours, 99%+ placement accuracy over an 11-month pilot
- Full vertical integration — designs its own actuators, batteries, hands, sensors AND the Helix VLA model in-house (post-OpenAI exit Feb 2025), enabling fast iteration and cost control
- BotQ manufacturing is a genuine differentiator — proven 24x throughput ramp, ~12k/yr first-gen line capacity, 100k-over-4-years target — manufacturing as a moat, not just a robot demo
Weaknesses
- Valuation–revenue gap is extreme — $39B on low-single-digit-$M 2025 revenue; the mark exceeds Goldman's projected 2035 humanoid TAM (~$38B) on some reads
- No audited financials, thin customer base, and unproven unit economics on a capex-heavy vertically integrated stack
- Home-readiness is years out by the CEO's own admission ('would not let my robot run free… with my young kids'), capping the consumer optionality the valuation implies
- Nov-2025 product-safety whistleblower complaint (ex-Head of Product Safety) alleging Figure 02 force levels and a 'gutted' safety roadmap around the Series C close — a reputational/legal overhang
Opportunities
- Massive TAM: physical labor automation in manufacturing, logistics, then home — a multi-trillion-dollar addressable pool if reliability + cost curve bend down
- Brookfield's 100,000 residential units as a Helix training/deployment ground gives a data + go-to-market edge toward the home market
- RaaS recurring-revenue model scales with the install base — per-hour factory pricing (~$25/hr) undercuts loaded human labor in many tasks
- Helix 02 full-body autonomy (Jan 2026) and 3-gram fingertip force sensing open dexterous tasks (fragile/irregular objects) competitors can't yet do
Threats
- China price competition — Unitree (G1 from ~$16k, 5,500+ units shipped 2025), AgiBot, UBTech scaling at ~1/10th the price — commoditization risk on hardware
- Tesla Optimus (Fremont volume ramp targeted 2026, ~$20–30k target) brings Tesla's manufacturing + capital + autonomy stack to the same market
- Customer diversification cuts both ways — BMW chose Hexagon's AEON for its Leipzig pilot (Feb 2026), downgrading Figure from sole partner to one supplier of several
- Embodied-AI is unsolved at scale — generalization, reliability, and safety remain open research problems; a high-profile failure/injury would be category-defining
Moats, dependencies & bottlenecks
Moats
In-house Helix vision-language-action model co-designed with the robot's hardware (post-OpenAI vertical integration) — a tightly coupled AI+actuation stack rivals can't license
a proven 24x throughput ramp and ~12k/yr line is itself a moat in a field where most rivals can't mass-produce
Proprietary hands/sensors (3-gram fingertip force sensing) enabling dexterous manipulation of fragile/irregular objects
Reference deployment + proprietary real-world operating data from BMW (and Brookfield's 100k residential units) compounding the Helix training flywheel
Dependencies
NVIDIA GPUs/compute for Helix training + inference (also a strategic investor)
vertically integrated hardware + factory build-out is cash-intensive; depends on the Series C runway and future raises
Anchor enterprise customers (BMW today) and conversion of pilots into volume RaaS contracts
batteries, and semiconductors despite in-house design (raw cells, chips, rare-earth magnets sourced externally)
Advantages
- First-mover at-scale factory deployment among US humanoids (BMW Spartanburg)
- Capital + strategic-backer depth unmatched among Western pure-plays
- Vertical integration speed — owns AI, actuators, hands, batteries, and the factory
- RaaS recurring-revenue model that scales with the install base rather than one-time hardware sales
Weaknesses
- No audited financials and revenue orders of magnitude below the valuation
- Thin, concentrated customer base now diversifying to rivals
- Capex/cash intensity of full vertical integration
- Home/consumer market — the biggest TAM slice — is years from readiness by management's own admission
Bottlenecks
- Embodied-AI generalization + reliability — Helix must handle open-world variation safely before broad deployment
- Manufacturing yield + cost-down — hitting the ~$20k unit target and 100k-over-4-years cadence at acceptable quality
- Safety certification + liability for robots operating near humans (sharpened by the whistleblower complaint)
- Battery life / uptime and the duty-cycle economics that make per-hour RaaS pricing profitable
Top signals & trends
Top signals
Apr-2026 (~240/mo) already beat all of 2025 (~150); sustained doubling is the core bull KPI
Diversifying past a single anchor is the gate to justifying the mark; BMW's Hexagon AEON pick is a yellow flag
A regulatory or litigation escalation would hit a robot that operates near humans
No S-1 yet; a markup priced round would validate $39B, a flat/down round would not
Helix 02 full-body autonomy (Jan 2026) is the AI moat under test
Trends
Category moving from demos to deployments; Figure is among the few with verified industrial evidence
VLA models + cheap compute make general-purpose manipulation newly tractable
Hardware commoditization pressure; US firms must win on AI + reliability, not unit price
Validates the TAM but raises competitive intensity and valuation froth
Per-hour RaaS humanoids slot into hard-to-staff manufacturing/logistics roles
Ecosystem & competitor graph
Suppliers feed the company; customers pull from it. Line thickness shows the strength of each tie (supply-chain dependency, customer earnings contribution). Hover to isolate a tie.
GPUs/compute for Helix training + inference, and strategic investor
on-robot compute (Qualcomm Ventures investor)
silicon, Intel Capital investor
fab for the AI/compute silicon up the chain
battery cells; LG Technology Ventures is an investor
anchor factory customer, Spartanburg X3 lines, expanding to additional workstations
providing 100,000 residential units as a Helix training/deployment ground (investor + channel)
the RaaS funnel beyond BMW
Most-capitalized rival; Fremont volume ramp targeted 2026, ~$20–30k target, full Tesla manufacturing + autonomy stack — the direct US threat
Electric Atlas production version at CES 2026; enterprise-grade (~$150k+), Hyundai + Google DeepMind deployments committed
Closest US peer on verified logistics deployment; Amazon warehouse testing, volume production facility
US humanoid backed by Google/Mercedes ties; a direct Western pure-play competitor
OpenAI-backed; targets the harder home market rather than factories
China leader — G1 from ~$16k, 5,500+ units shipped 2025; the price-commoditization threat (analysis only, not a buy/own rec)
Won BMW's Leipzig pilot over Figure 03 (Feb 2026) — concrete competitive loss for Figure