Figure AIFigure AI
LatestSeries C (Sep 2025, $39B post-money valuation)
$39B (Series C post-money, Sep 2025; Bloomberg had reported in Feb 2025 a $1.5B raise at a $39.5B valuation, which did not close)
Figure AI, Inc.
The world's most-funded and highest-valued humanoid robotics startup; self-developed Helix VLA model; Figure 03 in mass production (one unit per hour) with an open developer SDK
数据截至2026-08-24
About
Figure AI was founded in San Jose, California in 2022 by Brett Adcock. It is the world's most-funded (over $1.9B raised cumulatively) and highest-valued ($39B) humanoid robotics startup. Three product generations in three years: Figure 01 was the prototype; Figure 02 was deployed at a BMW plant for 11 months and helped produce 30,000 X3s; Figure 03 (released October 2025) is in mass production at the self-built BotQ factory, where capacity rose from one unit per day to one unit per hour within 120 days, with over 350 units cumulatively delivered and the Figure 02 developer SDK released in early 2026. Its self-developed Helix VLA model: Helix 02 released January 2026 (a single three-layer neural network controlling the whole body, 4 minutes of continuous autonomous tasks), a fully autonomous living-room tidying demo in March, 200 hours of continuous autonomous operation plus autonomous ladder climbing in May, deployment at a BMW logistics sorting station in June, and Project Go-Big with Brookfield to build the world's largest humanoid robot pre-training dataset. The May 2026 human-vs-robot live sorting showdown: triggered by Scott Walter's public skepticism about robot usefulness, Figure 03 livestreamed for 5 consecutive days (after already working 96 hours straight before the match); in the 10-hour duel the human intern sorted 12,924 pieces versus about 12,732 for the robot, with humans winning by about 190 pieces (2.79 s vs 2.83 s per piece), interpreted as "possibly the last human victory"; the average 1,248 pieces/hour became the industry's public efficiency benchmark. Currently ahead in capability validation but behind in shipments.
Technical Approach
Self-developed unified Helix VLA model with a three-layer architecture: System 2 (~1 Hz) handles semantic understanding and instruction decomposition, System 1 (~200 Hz) handles visuo-motor mapping, and System 0 (~1,000 Hz, ~10M parameters) is the whole-body controller; trained on 1,000 hours of human motion data plus 200,000 parallel simulation environments, replacing 109,000 lines of hand-written C++ control code. Vertically integrated, manufacturing at its own BotQ factory.
Commercialization（6）
- BotQ manufacturing factory (vertical integration): raised output from 1 unit/day to 1 unit/hour in 120 days (24x), with over 350 units cumulatively delivered; current capacity about 8,700 units/year, targeting 12,000 units/year and 100,000 units/year in four years
- BMW partnership: Figure 02 deployed for 11 months helping produce over 30,000 X3s and moving 90,000+ sheet-metal parts at over 99% accuracy per shift; in June 2026 Figure 03 was upgraded to logistics sorting (parts sequencing) under a multi-year commercial deployment; note that BMW's Leipzig European pilot chose Hexagon AEON, downgrading Figure to one of multiple vendors
- Human-vs-robot live sorting showdown (May 2026): the average 1,248 pieces/hour became the industry's public efficiency benchmark; in Aug 2026 Chinese vendor 自变量 (Zibianliang) broke this benchmark at roughly 30% of the cost with over 45% higher efficiency — the chasers have arrived
- Retail logistics: commercial partnership with Catalyst Brands in May 2026; Figure 02 developer SDK released in early 2026, opening a secondary-development ecosystem
- Cost: about $90,000–100,000 per unit, payback period about 18–24 months
- Shipment standing: absent from global humanoid deployment share statistics (AgiBot about 31%, Unitree about 27%); in a phase of leading capability validation but lagging shipments
Position & Strengths（3）
- The world's most-funded (over $1.9B cumulative) and highest-valued ($39B) humanoid robotics startup; fastest product iteration (three generations in three years); once named in TIME's Best Inventions of 2025
- Differentiation: vertical integration (self-built BotQ factory); Helix three-layer architecture (System 0/1/2); scaled commercial validation at a BMW plant (30,000 X3s over 11 months); the human-vs-robot sorting showdown, though narrowly lost by about 190 pieces, went viral and made 1,248 pieces/hour a quantifiable industry benchmark; Project Go-Big dataset built with Brookfield; a star-studded investor lineup (OpenAI, NVIDIA, Microsoft, Bezos, Intel, etc.)
- Controversies: a wide gap between the $39B valuation and commercialization progress (350 units shipped vs Unitree's 5,500, about a 6x valuation gap); the founder's track record; insufficient revenue transparency