Figure AI — humanoid robotics company targeting warehouse and manufacturing labor
September 14 at 04:20 · $0.124 total
Investment Memo: Figure AI
Stage framing: Hypothetical Series A evaluation. Figure has in reality raised far beyond Series A (reportedly a $675M round in early 2024 at ~$2.6B valuation, with a much larger round discussed in 2025 at valuations reported in the tens of billions). For this exercise, we evaluate as if entering early.
---
Thesis
The bull case is simple and enormous: labor is the largest expense category in the global economy. Warehouse and manufacturing labor alone represents hundreds of billions of dollars annually in the US, with chronic shortages, ~40%+ annual turnover in warehousing, and injury/insurance costs that make employers actively hungry for automation. A general-purpose humanoid that works in human-designed environments avoids the fatal flaw of traditional automation — retooling the facility. If any company achieves a humanoid that does even 2–3 economically useful tasks at <$100K unit cost with reasonable uptime, the TAM is not "robotics" (~$50B), it's a meaningful slice of labor itself. That is fund-returner math even at aggressive entry prices. Figure's specific claim to the prize: vertical integration (hardware + AI), speed of iteration (Figure 01 to 02 to 03 in roughly three years), and a founder (Brett Adcock, previously Archer Aviation and Vettery) who has shown he can recruit elite talent and raise capital at scale — which matters, because this is a capital war.
Product & Wedge
Figure builds full-size electric humanoids (Figure 02, now Figure 03) with dexterous hands, onboard compute, and vision-language-action models controlling behavior. After ending its OpenAI collaboration in early 2025, Figure built its own end-to-end neural model, "Helix," which maps vision + language directly to actions and has been demonstrated on multi-robot collaborative tasks and logistics package handling.
The wedge is deliberately narrow: repetitive, structured material handling — parts sequencing, tote handling, machine tending — inside partner facilities (BMW's Spartanburg plant is the flagship pilot). This is smart: constrained tasks, high tolerance for slow cycle times early, clear labor-cost comparables, and a customer motivated to co-develop. Success in one BMW workflow becomes the reference sale for every automotive OEM and 3PL.
Market & Competition
- Tesla Optimus — the scariest competitor: manufacturing scale, in-house AI, captive first customer (Tesla factories), and effectively unlimited capital. Execution and timelines remain unproven.
- Agility Robotics (Digit) — arguably ahead commercially: deployed with GXO Logistics under a robots-as-a-service model; purpose-built for logistics; a real factory (RoboFab).
- Boston Dynamics (electric Atlas) — best-in-class hardware pedigree, Hyundai backing and a captive automotive deployment path.
- 1X Technologies, Apptronik (Apollo, with Mercedes and GXO pilots), Sanctuary AI, plus a wave of well-funded Chinese entrants (Unitree, UBTech, Fourier) who will compress hardware pricing brutally.
No one has crossed from pilot to profitable fleet deployment. The race is genuinely open, but hardware cost curves favor China and AI talent/capital favor Tesla and Figure.
Traction & Business Signal
Publicly known:
- BMW commercial agreement (Jan 2024); robots performing sheet-metal parts placement tasks at Spartanburg — scale and economics unknown, and reporting has suggested the deployment was smaller than marketing implied.
- Helix demos: logistics parcel handling, household tasks, multi-robot collaboration (demos, not audited deployments).
- Announced a second commercial customer (unnamed publicly at announcement) and stated ambitions to ship thousands of robots.
- BotQ manufacturing facility announced with a stated capacity target of 12,000 humanoids/year.
Unknown: revenue (likely negligible), pricing, unit economics, robot uptime/MTBF, tasks-per-hour vs. human baseline, contract structure with BMW, churn/renewal on pilots. Essentially all commercial fundamentals are unverified. Signal today = capital raised + talent density + demo velocity, not business results.
Risks (the three that kill the deal)
- Unit economics never clear the labor bar. A humanoid must beat ~$25–35/hr fully loaded labor. That requires ~$50–100K COGS, multi-year lifespan, high uptime, and minimal teleoperation/supervision. If robots need one human minder per two robots, the model collapses. Today, no public evidence Figure clears this bar — and physics (hands, batteries, reliability) is unforgiving on venture timelines.
- Manipulation autonomy is unsolved. Locomotion is largely done; dexterous, reliable, generalizable manipulation in cluttered real environments is the frontier. If it takes 7–10 more years, Figure burns through capital selling demos. VLA models are promising but demo-to-deployment gaps in robotics have historically been graveyard-wide.
- Capital war against Tesla and China. This market may be won by whoever can subsidize losses longest and manufacture cheapest. Tesla has both advantages; Chinese players will price hardware at levels Figure cannot match. Figure's survival requires perpetual mega-fundraising — dilution risk is severe and a single failed round is fatal given burn.
Verdict
Invest — at a true Series A price, with sized conviction (not max check). The founder-market fit, iteration speed, and end-to-end AI approach make Figure a top-3 credible contender in a category where the winner is worth $100B+. The risks are existential but symmetric across all players; you're buying a lottery ticket with better-than-lottery odds. At current real-world valuations, this becomes a much harder call — the risk hasn't retired proportionally to the price.
The one question I need answered first: In the BMW deployment, what is the true fully-autonomous task completion rate and cost per completed task versus a human worker — with no teleoperation, no cherry-picked demo runs? That single number tells you whether this is a business or a very expensive research lab.
Investment Memo: Figure AI (hypothetical Series A)
Thesis Figure is betting that general-purpose humanoid robots will be the next platform after smartphones and EVs: a single form factor that can enter existing human-designed factories and warehouses without retooling the world. If they ship a reliable, cost-effective bipedal worker that can do multi-hour shifts of material handling, kitting, and simple assembly, they capture a slice of the $1T+ global labor cost in logistics and manufacturing. A fund-returner outcome requires them to become the default “body + brain” stack (hardware + foundation model for embodiment) before Tesla Optimus, Agility, or Chinese players lock in the category. The asymmetry is real: labor shortages are structural, wages keep rising, and a working humanoid at <$50k fully loaded with 2–3 year payback would be adopted faster than most VCs currently model. Failure modes are also binary—hardware reliability, data flywheel, and capital intensity—so this is a high-conviction, high-risk bet, not a diversified robotics play.
Product & wedge Figure’s product is a bipedal humanoid (Figure 01 → 02) designed for unstructured human environments: two arms, dexterous hands, vision + language, and an end-to-end neural policy rather than classical robotics stacks. The wedge is warehouse and light manufacturing first—tasks that already have clear ROI (palletizing, tote handling, line feeding) and where the robot can be tele-operated or supervised initially, then autonomously. They are pairing the body with a large multimodal model (OpenAI collaboration publicly announced) so the same robot can be instructed in natural language and improve via fleet data. Differentiation vs. wheeled AMRs or single-arm cobots is the ability to use existing aisles, stairs, and human tools without facility redesign. The long-term vision is a generalist that graduates from factory to home; the near-term product is a specialized labor replacement that looks general.
Market & competition The addressable market is the replacement of repetitive physical labor in logistics (Amazon, 3PLs, automotive OEMs) and discrete manufacturing. Even 5–10% penetration of warehouse FTE in the US/EU is a multi-billion-dollar hardware + recurring software opportunity. Real competitors: Tesla Optimus (scale, vertical integration, Elon distribution), Agility Robotics (Digit, already in Amazon trials, more conservative wheeled-biped design), Boston Dynamics (Atlas, technically impressive but historically not commercially focused), Apptronik (Apollo, NASA/Google heritage), Sanctuary AI, 1X (NEO), and a wave of well-funded Chinese players (Fourier, Unitree, Agibot). Figure’s claimed edge is speed of iteration + AI partnership + manufacturing-first GTM vs. research-lab heritage. The market will likely support 2–3 winners; winner-take-most dynamics appear in the foundation model + data flywheel, not the actuators.
Traction & business signal Publicly known: Figure has raised large subsequent rounds (hundreds of millions) from Microsoft, OpenAI, NVIDIA, Jeff Bezos, Parkway, etc., implying strong later-stage conviction. They have a public BMW manufacturing partnership and have shown videos of Figure 02 walking, picking objects, and conversing. They have demonstrated end-to-end neural control rather than purely tele-op. Exact unit shipments, revenue, gross margin, customer concentration, and production yield are unknown. Fleet hours, mean-time-between-failure, and real-world autonomy percentage are unknown. Hiring velocity and manufacturing scale-up progress beyond demos are unknown. No public Series A metrics exist in this hypothetical; later capital raises and OEM logos are the main observable signals.
Risks (the three that actually kill the deal)
- Hardware reliability and unit economics never close. Humanoids have thousands of degrees of freedom, high power draw, and catastrophic failure modes (falls, actuator wear). If Figure cannot hit 20k+ hour MTBF at a BOM that allows <$30–40k ASP with positive contribution margin, they remain a science project while cheaper wheeled or single-arm solutions take the warehouse. Tesla and Chinese players have structural cost advantages.
- Data and autonomy flywheel does not materialize. Language models transfer poorly to contact-rich, long-horizon physical tasks. If they stay in the “impressive 30-second demo / supervised 4-hour shift” regime, customers will not pay for robots that still need humans in the loop. Competitors with more real-world hours (Agility at Amazon, Tesla’s own factories) could pull ahead.
- Capital intensity + timing. Building a humanoid at scale requires hundreds of millions before meaningful revenue. A 2025–2027 recession, slower OEM adoption, or a Tesla Optimus “good enough” launch could starve them of the next round or force a down-round that wipes early investors. Labor markets could also ease, reducing urgency.
Verdict Pass at a hypothetical Series A. The vision is correct and the team has subsequently proven they can raise and attract partners, but at true Series A the technical and manufacturing risk is still too high relative to capital required, and publicly observable traction (beyond videos and later logos) is insufficient to underwrite a fund-returner probability. The one question I would need answered first: “What is the current mean productive hours per robot per day in a real customer facility without tele-operation, and what is the all-in cost (hardware + energy + supervision) versus the human it replaces?” Until that number is both known and improving on a steep curve, this remains a research bet, not a Series A company.