Figure AI — humanoid robotics company targeting warehouse and manufacturing labor

Investment Memo: Figure AI

Company: Figure AI (founded 2022, Brett Adcock) Sector: Humanoid robotics — warehouse & manufacturing labor Stage considered: Hypothetical Series A

---

Thesis

The labor market for repetitive physical work is enormous and structurally undersupplied. US warehousing and manufacturing alone employ ~13M people; turnover in warehouse roles routinely exceeds 40% annually, and manufacturers report hundreds of thousands of unfilled positions. If a general-purpose humanoid can perform even a narrow band of these tasks at a fully-loaded cost below ~$15–25/hour equivalent, the addressable spend is measured in hundreds of billions of dollars — not a robotics market, a labor market.

The fund-returner case: humanoids are the rare category where the winner could be a $100B+ company, because the form factor amortizes across every human-built environment without facility retrofits. Recent step-changes in foundation models for vision-language-action (VLA) control have compressed the timeline from "decades" to "plausibly this decade." Figure is one of perhaps four credible Western contenders, founded by a repeat founder (Archer Aviation, Vettery) who has demonstrated he can recruit elite hardware talent and raise capital at scale. In a category where capital access is itself a moat, Figure's ability to raise mega-rounds is a strategic asset, not just dilution.

Product & Wedge

Figure builds general-purpose bipedal humanoids (Figure 01, Figure 02, and the newer Figure 03) with dexterous hands, onboard compute, and a proprietary VLA model ("Helix") after ending its OpenAI collaboration in early 2025 to bring AI in-house. The wedge is deliberate: structured, repetitive, single-station tasks in logistics and manufacturing — tote handling, machine tending, parts sequencing — where the environment is semi-controlled and ROI math is legible. The company has publicly discussed its BotQ manufacturing facility with stated ambitions to produce thousands of units annually, signaling a bet on vertical integration of manufacturing itself.

The wedge logic is sound: win one narrow, high-turnover task class, deploy at customer sites for data flywheel, and expand task coverage via software rather than new hardware SKUs.

Market & Competition

TAM: Global logistics + manufacturing labor spend exceeds $1T. Even a 1% penetration at robot-as-a-service pricing implies a $10B+ revenue opportunity.

Competitors (real and serious):

• Tesla Optimus — the existential threat: manufacturing scale, in-house AI, captive first customer (Tesla factories), and Musk's cost-down discipline.
• Agility Robotics (Digit) — furthest along in actual warehouse deployments (GXO, Amazon pilots); purpose-built for logistics rather than general humanoid ambitions.
• Boston Dynamics (electric Atlas) — Hyundai-backed, best-in-class hardware, now pivoting to commercial manufacturing use.
• 1X Technologies, Apptronik (Apollo) — well-funded; Apptronik has a Mercedes-Benz pilot and Google DeepMind partnership.
• Chinese entrants (Unitree, UBTech, Fourier) — aggressive on cost; Unitree sells humanoids at price points Western firms can't yet touch.

Differentiation among Western players is currently narrative-heavy and data-light. No one has demonstrated durable, unattended commercial operation at scale.

Traction & Business Signal (publicly known)

• BMW Spartanburg partnership (announced Jan 2024): Figure 02 units performing sheet-metal placement tasks; scope and unit count beyond pilot scale — unknown. Media reporting has questioned the depth of deployment.
• Second commercial customer referenced by the company (widely reported to be a major logistics player, unconfirmed by Figure) — details unknown.
• Financing: ~$675M Series B (Feb 2024) at ~$2.6B valuation from OpenAI Startup Fund, Microsoft, NVIDIA, Bezos Expeditions, Intel Capital; a 2025 round reported at ~$39B valuation.
• Revenue, unit economics, units deployed, uptime, tasks-per-hour vs. human baseline: all unknown.
• Ended OpenAI partnership Feb 2025, claiming in-house Helix breakthrough — unverified externally.

Net: strong capital and demo signal; essentially zero public evidence of commercial-scale performance.

Risks (the three that kill the deal)

1. The reliability gap is unsolved and may be far away. Demos ≠ deployment. A robot that succeeds 95% of the time is commercially useless in a line-rate manufacturing environment; humans are 99.9%+ with graceful failure recovery. If closing the last 4.9% takes 7–10 years, Figure burns through even billions before revenue matters. This is the deal-killer.
2. Tesla and Chinese cost curves. If humanoid hardware commoditizes (Unitree already signals this) and the differentiator becomes the AI model plus manufacturing scale, Figure competes against Tesla's factories and Google-scale AI labs with neither advantage. Capital access alone doesn't win a hardware cost war.
3. Governance/narrative risk. Founder-driven hype cycles (a 15x valuation jump in ~18 months with unknown revenue), a disputed partnership scope, and the abrupt OpenAI split raise the question of whether public claims track internal reality. In hardware, narrative-reality gaps compound fatally.

Verdict

Invest at Series A — with sized, eyes-open conviction (1–1.5% of fund, not a concentrated bet). At Series A pricing, the asymmetry works: the category is real, the founder recruits and raises exceptionally, and the downside is a known quantity while the upside is a generational company. At the reported later-stage valuations, pass — the risk is Series A-shaped and the price is not.

The one question I need answered first: At the BMW deployment, what is the current mean-time-between-human-intervention, and how has it trended over the last six months? That single curve tells you whether this is a five-year business or a fifteen-year science project.

MEMORANDUM

To: Investment Committee From: [Analyst Name] Date: October 26, 2023 Subject: Figure AI – Series A Investment Consideration

Thesis

Figure AI represents a binary, high-convexity bet on the physical manifestation of artificial intelligence. The thesis rests on a simple premise: the global economy is built for human form factors, yet the demographic pyramid is inverting, creating a structural deficit in manual labor. If Figure can successfully integrate large language models (LLMs) with a dexterous bipedal platform, they are not building a robot; they are building the hardware shell for AGI. This is a "fund-returner" opportunity because the TAM is effectively the entire global labor market for physical tasks (estimated at $40T+). Unlike narrow vertical automation, a successful general-purpose humanoid is a platform play with winner-take-most dynamics. The upside is not 10x; it is 100x if they achieve "escape velocity" in autonomy.

Product & Wedge

Figure is developing Figure 01, a general-purpose electromechanical humanoid. The key differentiation is not the hardware specs (torque, degrees of freedom), but the software architecture. The company is explicitly leveraging recent breakthroughs in transformer-based AI to enable natural language task planning and "learning by watching."

The Wedge: They are not starting in the home (too chaotic) or on roads (too regulated). They are targeting warehousing and manufacturing. This is the correct wedge for three reasons:

1. Structured Chaos: Warehouses are semi-structured. The lighting is controlled, the floors are flat, and the tasks (picking, placing, palletizing) are repetitive but variable enough to require a humanoid rather than a fixed arm.
2. Labor Pain: Post-COVID, warehouse turnover is >100% annually. The labor is physically destructive and dull. There is immediate budget for CapEx that replaces OpEx.
3. The "BMW Signal": Figure’s partnership with BMW to deploy robots in their Spartanburg plant is a massive signal. Automotive manufacturing is the gold standard for precision, safety, and uptime. If they can survive that environment, they can survive Amazon.

Market & Competition

The humanoid space has transitioned from science fiction to a hotly contested arms race in 18 months.

• Tesla (Optimus): The 800-pound gorilla. Tesla has vertical integration, manufacturing scale, and the FSD inference chips. However, Tesla is a public company distracted by EVs and Twitter. Optimus is currently teleoperated and arguably a PR asset more than a product. Figure’s agility as a private company is an advantage.
• Agility Robotics (Digit): The incumbent in "pragmatic" humanoids. Digit walks like a bird (backward knees) and is designed specifically for tote handling. They have a partnership with Amazon. They are further along on pilot revenue but are less "human" in form factor, potentially limiting the generality of the AI training data.
• 1X Technologies (backed by OpenAI): The dark horse. They are focusing on "safe" androids with soft mechanics. Their tie to OpenAI gives them a potential edge in the "brain," but their hardware appears less robust for heavy industrial payloads.
• Apptronik (Apollo): A spin-out from UT Austin with deep heritage (NASA Valkyrie). They are strong on hardware design and have a GXO logistics pilot.

Figure’s Edge: Figure has assembled the "Avengers" of robotics (ex-Boston Dynamics, Tesla, Apple). Their pace of execution—from founding to a walking, manipulating robot in under 12 months—is unprecedented.

Traction & Business Signal

Note: As a Series A, financial traction is limited. We are evaluating engineering velocity.

• Capital: Raised a $70M Series A led by Parkway, followed by a massive $675M round at a $2.6B valuation from Microsoft, NVIDIA, OpenAI, and Bezos. This is not just capital; it is a strategic moat. They have preferred access to Azure compute, NVIDIA chips, and GPT models.
• Technical Milestones: Public videos show Figure 01 walking dynamically and manipulating objects (picking up a box, placing it on a conveyor). The "coffee machine" video demonstrated end-to-end neural network learning (watching humans make coffee and replicating it).
• Commercial: The BMW agreement is the only named commercial pilot. Revenue is unknown, presumably negligible. Unit economics (BOM cost vs. price) are unknown but likely negative at this stage.

Risks (The Killers)

1. The "Dexterity Wall" (Technical Risk): Walking is solved. The hand is not. The economic value of a humanoid is in fine manipulation (grasping a wire, sorting polybags). Current actuators are either too weak, too slow, or too expensive. If the "end effector" problem is not solved, Figure is left selling a very expensive security guard that walks around but can’t actually do the job. This is the primary risk that kills the deal.
2. Capital Intensity & The "Valley of Death": Scaling humanoid manufacturing requires billions in CapEx for tooling, actuators, and assembly. Unlike software, gross margins on hardware start at 20-30%, not 80%. If the capital markets close (as they did in 2022), Figure will burn through cash before reaching volume production. They are reliant on the "Magnificent 7" continuing to fund their own disruption.
3. The "Uncanny Valley" of Utility: A humanoid costs $50k-$100k. A warehouse worker costs $40k/year. If the robot is 50% as fast as a human, it has a 2-year payback—great. But if it falls down every 4 hours, requires a human supervisor, or can only work 4 hours on a charge, the ROI collapses. The risk is that the robot is a "demo darling" that fails the rigorous uptime requirements of a 24/7 logistics center.

Verdict: PASS (Conditional)

Decision: Pass at the current hypothetical Series A terms.

Rationale: We are late to the party. The $2.6B valuation already prices in the successful execution of the BMW pilot and the resolution of the dexterity problem. We would be paying "platform" prices for a company that currently has "pilot" traction. The risk/reward asymmetry has narrowed. The presence of Microsoft/NVIDIA provides a safety net, but it also caps our influence and potential ownership.

The One Question Needed to Flip to "Invest": "What is the Mean Time Between Failure (MTBF) for the hands, and what is the specific task completion rate (success %) on a standard warehouse pick-and-place task over a continuous 8-hour shift without human intervention?"

If the answer is >95% success rate and >8 hours uptime, the company is under-valued. If the answer is "we are working on it," it is a science project, not a business.