Robotics / Logistics / Manufacturing · Deep dive

Figure

Sunnyvale humanoid-robot company building a bipedal general-purpose robot — Figure 01, 02 and 03 — plus Helix, the vertically-integrated vision-language-action model it built after walking away from OpenAI, with a $1B BMW pilot, a Brookfield data partnership, and a $39B mark that bakes in a decade of execution the company has not yet done.

emerging

The question that decides it: Does Figure's end-to-end Helix VLA plus vertically-integrated hardware stack reach commercial unit economics — hundreds of autonomously-shift-working robots at a documented $/hour, with margin — before (a) foundation-model-first shops like Physical Intelligence, Skild AI, and Nvidia's open GR00T N1 commoditise the brain layer; (b) Boston Dynamics-Hyundai Atlas, Tesla Optimus, Agility Digit and Apptronik Apollo ship at scale into the same automotive and logistics customers; and (c) the $39B mark becomes uninvestable at the next priced round? A yes looks like: Helix-controlled Figure 03s running sustained, unteleoperated shifts at BMW and a second named customer by end-2027, BotQ output trending toward a four-figure annual run-rate, and a down-round-proof commercial contract disclosed to investors. A no looks like: BMW stays a demo, UPS stays a Bloomberg rumour, the Helix split from OpenAI opens a quality gap that Physical Intelligence or Skild closes first, and the Series C mark gets reset the way Humane's did.

My take ★★ Very interesting ★ Interesting − Maybe ✗ Not interesting

Snapshot

Figure is Brett Adcock’s humanoid-robot company, founded January 2022 in Sunnyvale and one of the two highest-valued private robotics companies in the world. It builds a bipedal general-purpose humanoid — Figure 01 (Oct 2023), 02 (Aug 2024), 03 (Oct 2025) — and, since Adcock’s February 2025 split from OpenAI, trains the policy in-house: a dual-system VLA called Helix. ~$2.3B+ raised across three rounds, closing a $39B Series C in September 2025 — a ~15x step-up from the $2.6B Series B eighteen months earlier. BMW is the one publicly named manufacturing customer (Jan 2024, Spartanburg SC); Bloomberg reported UPS talks in April 2025; Brookfield joined in September 2025 as investor and data/infrastructure partner. No disclosed revenue.

Founding story

Figure is Adcock’s third company. He started Vettery, an online recruiting marketplace, in 2013 and sold it to Adecco in 2018 for a reported ~$110M. He co-founded Archer Aviation the same year, took it public via SPAC in September 2021 (NYSE:ACHR), and stepped down from co-CEO in April 2022 to focus on Figure — incorporated three months earlier and self-funded for the first year while he built the founding engineering team.

Adcock’s pitch is that labour is the biggest input to the physical economy and bipedal humanoids are the only form factor that drops into human-shaped environments — stairs, doors, aisles, assembly lines — without re-engineering them. The May 2023 $70M Series A from Parkway went almost entirely to engineering hires; by Figure 01’s October 2023 unveil the company had ~80 staff, by Series B in February 2024, ~300.

How it works

A Figure robot is a bipedal ~5-ft-6, ~70 kg electric humanoid with 16-DOF hands, cameras in the head and torso, a battery rated at ~5 hours, and an onboard GPU for inference. Figure 03 (October 2025) added tactile fingertip sensors with embedded cameras — the hardware investment separating Helix’s dexterity ambitions from warehouse-grade pick-and-place.

The brain, since the OpenAI split, is Helix: a dual-system VLA. “System 2” is a ~7B-parameter internet-pre-trained vision-language model that reasons about scene and instruction at ~7-9 Hz; “System 1” is an 80M-parameter visuomotor policy converting latents into continuous joint commands at ~200 Hz. The split addresses the universal humanoid problem: LLM-class reasoning is too slow for real-time control, real-time controllers can’t reason. It is Figure’s main technical bet against the single-policy approaches at Physical Intelligence and Skild.

Hardware is built at BotQ, Figure’s in-house factory unveiled March 2025, nameplate ~12,000 robots/year at scale. Humanoids Daily and Robot Report reporting through 2025-26 puts actual output in the low hundreds — the capacity-to-production gap Agility, Apptronik and Boston Dynamics-Hyundai all share.

Product and business overview

Three product lines, only one in a paying customer’s building. Figure 03 is the flagship, pitched for industrial manufacturing (BMW), logistics (UPS reportedly), and — per Adcock’s February 2025 Fortune interview — home alpha. Helix is positioned as Figure’s second commercial moat after the OpenAI split. BotQ is the vertical-integration play — Adcock argues owning the line is the only way to get BOM below the ~$20-30K where unit economics clear. No SKU page, no logo wall, no case study; GTM is CEO-driven enterprise sales backed by the Brookfield partnership and Nvidia/Microsoft investor-channel relationships.

Business model and pricing

Undisclosed. Two models surface in commentary. Robots-as-a-service: shift hours at a per-robot-per-hour rate (Adcock has publicly floated $8-12/hour long-run, below U.S. manufacturing labour); customer never owns the robot. Direct hardware sale plus Helix licence: pay once for hardware, pay for software updates. BMW hints at a hybrid — Figure-owned robots, revenue per task completed. No pricing published; no third party has audited unit economics.

Traction over time

Headcount: ~80 end-2023 → ~300 at Series B → ~400+ at Series C. Revenue: not disclosed, no public source confirms a paid at-scale deployment.

Market analysis

Goldman Sachs’ September 2026 update puts the humanoid installed base at 6.5M units by 2035 (up from 1.4M); Morgan Stanley has run higher. Near-term markets are narrower: U.S. automotive assembly (BMW-Figure, Hyundai-Boston Dynamics, Mercedes-Apptronik), third-party logistics (Agility-GXO, Apptronik-GXO, Figure-UPS reportedly), e-commerce fulfilment (Amazon captive, Covariant acqui-hire), and consumer home (1X Neo, Figure alpha). Structural forces: LLM-era investor tolerance for compute-hungry pre-revenue stories; a decade of cheap teleop tooling that broke the data wall; and Nvidia’s 2025 open-sourcing of GR00T N1, which commoditised the base model most private humanoid companies had argued was their moat.

Competitive intel

See structured set. Sharpest angles: Tesla Optimus is the only peer with manufacturing-at-scale capability in place; if Musk ships, Figure’s hardware moat collapses. Boston Dynamics-Hyundai Atlas has a captive OEM and no need for priced capital — if the Hyundai Metaplant pilot scales first, the BMW narrative looks less unique. Agility (GXO) and Apptronik ($5.5B + DeepMind) bracket Figure’s bet from the commercial-first and model-partnership sides. Physical Intelligence and Skild attack the brain layer; if their generalist VLAs beat Helix on sample efficiency, the OpenAI split looks like a strategic error at the wrong moment. 1X Neo has already shipped the consumer wedge Figure only teased.

History and evolution

Beyond the traction table: the March 2024 OpenAI partnership generated the viral “coffee-making” demo that became the company’s dominant public image. The February 2025 split, framed by Adcock as “major breakthroughs in fully end-to-end robot AI,” preceded Helix by two weeks. The September 2025 Brookfield deal paired the Series C check with a data/infrastructure partnership for “the world’s largest humanoid pre-training dataset.” The November 2025 Gruendel whistleblower suit alleges he was fired after warning Figure robots could fracture a skull and that safety roadmaps had been walked back post-Series C; Adcock has threatened defamation action over partnership coverage in Fortune through 2026.

What people say

The case for. Robotics press (Robot Report, Humanoids Daily, IEEE Spectrum) rate Figure’s hardware cadence — 01 to 02 to 03 in 24 months — as the fastest in the Western humanoid cohort. The Feb 2025 Helix release video (two Figure 02s putting away groceries zero-shot in a novel kitchen) drew favourable technical commentary. Adcock’s track record (Vettery, Archer IPO) buys unusual founder credit; Series C investors cite BotQ plus Helix as the integrated bet nobody else is making at this capital intensity.

The complaints. Four recurring themes. (1) Staged demos — academics on X and Hacker News argued the March 2024 coffee-maker video was cherry-picked or partially teleoperated; Figure has not published independent benchmarks. (2) BMW is a pilot, not a deployment — Adcock has dodged scope/uptime questions, skipping a live demo at a 2025 conference per Fortune. (3) Safety — the Gruendel whistleblower suit alleging walked-back safety roadmaps is corroborated by Techspot and Robotics.press; the “fracture a skull” claim is on the public record. (4) The $39B mark — bubble pieces (Futurism, EntrepreneurLoop, Live in the Future) argue the step-up prices capability Figure hasn’t shipped, with Humane, WeWork and Argo as down-round comps. Glassdoor clusters on high talent density and ambition versus long hours, political management, and Adcock-driven senior-leadership churn.

Outlook: the open question

The answer-conditions on either side are concrete. A yes requires: Helix-controlled Figure 03s running sustained unteleoperated shifts at BMW with a documented $/hour unit economic and audited utilisation; a second named lighthouse customer paying for at-scale deployment by end-2027 (UPS the obvious candidate); BotQ output trending into a four-figure annual run-rate; and a priced round that holds — or marks up — the $39B. In that world Figure’s integrated bet rhymes with Tesla’s (vertical hardware + proprietary model + captive manufacturing) and the mark is cheap.

A no looks like: BMW stays a visually impressive but limited pilot; UPS stays rumour; the OpenAI split opens a reasoning-quality gap that Physical Intelligence or Skild closes first; BotQ output stays in the hundreds while Boston Dynamics-Hyundai and Tesla ramp into the automotive base; and the whistleblower suit drags on Fortune 500 procurement. Then the $39B is unmarketable at the next round and Figure follows the Humane trajectory — a venture tragedy whose hardware was never the problem. In October 2026, with the Series C mark fourteen months old and Figure 03 fewer than twelve months in the field, nobody has the deployment data to answer this yet. The 2026-2028 BMW cohort and the UPS decision are the answer.

How to attack it

The attacker’s wedge is narrow-vertical, labour-SLA, Nvidia-GR00T-based. Figure is trying to sell one humanoid plus one model to every blue-collar job on earth; the near-term buyer is a plant manager or 3PL ops director who does not want a model and does not want a humanoid — they want a predictable hourly rate for a specific task (trailer unload, kitting, bin-to-tote picking) with uptime guarantees. A credible attacker builds: (a) a vertically-focused robot (not necessarily bipedal — Agility’s semi-humanoid Digit proved this) tuned to one task family, (b) a policy stack on top of Nvidia’s open GR00T N1 and openpi, (c) a per-hour contract that undercuts U.S. manufacturing labour with hardware financed on the attacker’s balance sheet, and (d) a tight services motion for integration. Agility’s GXO contract and Chef Robotics’ per-shift food-prep model are the templates.

Weaknesses to exploit, each sourced. $39B is years ahead of revenue (Futurism Nov 2025, EntrepreneurLoop 2026): a well-capitalised attacker can wait for the markdown and buy talent. Helix is a strategic bet made at the moment of maximum optionality lost — the OpenAI split (Humanoids Daily, Feb 2025) traded reasoning quality for ownership; Physical Intelligence’s π₀.₇ (Apr 2026) and Skild’s $1.4B at $14B (Jan 2026) suggest the brain layer moves fastest. BotQ yield is unproven at the stated 12K-unit cadence (Humanoids Daily production tracking, 2026); competitors with captive manufacturing (Hyundai via Boston Dynamics, LG’s partner factories) have a structural cost advantage. Automotive union politics: BMW’s non-union Spartanburg plant masks friction that UAW-coded Toyota, Ford and GM deployments would surface. Safety overhang: the Gruendel whistleblower suit (Techspot, Nov 2025) creates procurement-committee friction at Fortune 500 buyers. Culture: Adcock’s repeated defamation threats at reporters (Humanoids Daily 2025-26) signal a founder-fragile communications posture that invites more scrutiny, not less.

Adjacent-segment play

Three adjacent wedges where Figure’s stack repackages.

Humanoid-as-a-service rentals for narrow verticals. Per-shift rental on the operator’s balance sheet, aimed at mid-market 3PLs and co-manufacturers that cannot amortise a $30K+ purchase. Agility’s GXO contract is the proof point; comps are Formic (robotics-as-a-service in food/CPG) and a verticalised PlusOne Robotics.

Foundation-model licensing with Helix as the product. Spin Helix out the way Argo could have spun its stack from Ford/VW. Physical Intelligence and Skild are the comps. Obstacle: OEM-captive customers (Tesla, Hyundai, 1X) will build their own, so TAM is smaller than the humanoid buildout.

Defence / SOCOM / dirty-dull-dangerous. Humanoids in CBRN inspection, battlefield logistics, bomb disposal and tunnel reconnaissance are the one bidder set that will pay $500K+ per unit for ruggedised hardware. Boston Dynamics has run SOCOM demos for a decade; comps are Shield AI and Ghost Robotics. Figure has not publicly pursued it; this is where the wedge monetises fastest and least diluted.

Elder care and home consumer — Adcock’s February 2025 gesture — is the wrong adjacent. 1X Neo already ships at $20K, consumers will not pay more, and liability friction (the Gruendel suit) makes the home a terrible first market for a 70kg robot that can fracture a skull.

Sources and further reading

• “Figure raises $1B+ in Series C, hits $39B valuation” — Seeking Alpha, Sep 24, 2025
• “Arcapita and Parkway-backed Figure AI raises over USD 1 billion in Series C” — Arcapita, Sep 24, 2025
• “Figure announces strategic partnership with Brookfield” — Finviz, Sep 2025
• “BMW partners with Figure to deploy humanoids in its US plant” — S&P, Jan 2024
• “Figure AI reorganizes around new Helix AI model” — Humanoids Daily, Feb 2025
• “Figure will start alpha testing humanoids in homes in 2025” — TechCrunch, Feb 27, 2025
• “UPS in talks with startup Figure AI to deploy humanoid robots” — Bloomberg, Apr 28, 2025
• “Launch of Figure 03” — Robophilosophy, Oct 2025
• “Figure AI hit with whistleblower lawsuit over robot safety” — Robotics.press, Nov 2025
• “AI robots have power to crush skulls, whistleblower lawsuit alleges” — Techspot, Nov 2025
• “Goldman Sachs now sees 6.5 million humanoid robots by 2035” — 24/7 Wall St, Sep 13, 2026
• “Apptronik triples valuation to $5.5B+” — Bloomberg, Feb 11, 2026
• “GXO signs industry-first multi-year agreement with Agility Robotics” — Stock Titan, Jun 2024
• “Brett Adcock” — Wikipedia