Figure AI completed its Series C funding round in September 2025, securing over $1 billion and reaching a valuation of $39 billion (approximately 280 billion yuan).
That figure is already quite staggering for the humanoid robotics sector.
But more noteworthy than the money is a strategic decision: completely severing its partnership with OpenAI and building its own AI brain from scratch.
Why Figure AI Dropped OpenAI
Figure AI and OpenAI announced their partnership in early 2024, with OpenAI providing language capabilities to Figure's robots—enabling them to understand speech and execute verbal commands.
Then, in February 2025, the collaboration ended.
Figure CEO Brett Adcock offered a pragmatic explanation in interviews: large language models are no longer the primary bottleneck. What they truly need is for the robot to make motion decisions at a frequency exceeding 200 times per second—a level of high-speed real-time control that LLMs simply cannot reach.
So they developed Helix.
Helix: A Three-Layer Neural Network Architecture
Helix is Figure's self-developed Vision-Language-Action (VLA) model, essentially a hierarchical control system:
| Layer | Frequency | Responsibility |
|---|---|---|
| System 2 | Visual reasoning | 7-billion-parameter VLA model for high-level planning and language understanding |
| System 1 | 200 Hz | Real-time motion strategy, rapid response to environmental changes |
| System 0 | 1000 Hz | Low-level balance control and physical contact handling |
The core logic of this three-layer design is to separate "what to think" from "how to move," preventing reasoning latency from dragging down real-time control.
System 2 runs a 7-billion-parameter language-vision-action model responsible for scene understanding, instruction comprehension, and action sequence planning. System 1 converts those plans into actual motion commands at 200 Hz. System 0 manages balance and force feedback at 1000 Hz—a rate roughly on par with human muscle reflexes.
"Large language models are no longer the main bottleneck. The real challenge is enabling robots to stably control their bodies at high frequencies." —Brett Adcock, Figure AI CEO
Figure 03: Aimed at the Home Market
The Figure 03 is Figure's third-generation robot, targeting home environments.
Key specifications:
- Height: 168 cm, Weight: 60 kg
- Full-body soft, washable textile mesh covering, with no pinch points, suitable for environments with children and pets
- 20 degrees of freedom per hand, fingertip tactile sensors capable of sensing forces as low as 3 grams
- Battery life of approximately 5 hours
The tactile sensing capability deserves special mention. A perception accuracy of 3 grams means the robot can handle fragile items like eggs without explicit programming—not by pre-setting "how much force for an egg," but by using sensor feedback to determine force in real time. This previously required extensive manual calibration; now it's handled through perception feedback.
Figure released a 4-minute unedited video showing the robot unloading a dishwasher, folding laundry, and loading a dishwasher—with motion fluidity that appears more natural than most competitors' demonstrations.
Figure plans to produce 12,000 units in 2026, with a target price of $20,000 (approximately 145,000 yuan).
Who Is Investing
The Series C investor lineup is notable: NVIDIA, Intel Capital, Qualcomm Ventures, and Salesforce—a mix of hardware and software giants.
Brookfield Asset Management also participated, an institution managing over $900 billion in assets, representing more traditional large capital beginning to seriously consider the robotics sector.
Overall funding: total raised exceeds $1.9 billion, with a valuation of $39 billion.
BotQ Factory and Scaling Challenges
Figure has built a robot factory called BotQ in California, with a target of producing 100,000 units over four years, averaging 25,000 per year.
For comparison:
- Boston Dynamics' Atlas has announced a production target of 30,000 units per year at Hyundai's facility.
- Figure's BotQ factory is still in its ramp-up phase.
- There is no public data on how many of the planned 12,000 units in 2026 will actually reach customers.
$39 billion valuation ÷ 100,000 units over 4 years = a capital burden of approximately $390,000 per robot.
This doesn't mean Figure isn't worth that amount, but it does mean they need to prove themselves simultaneously across technology, production capacity, and pricing.
The Home Market Is the Toughest Path
Humanoid robotics currently feels like a "must push forward regardless of outcome" atmosphere. Tesla's Optimus is progressing slower than expected (still in R&D phase in Q1 2026), Boston Dynamics' Atlas is taking an industrial route, and 1X Technologies is pursuing a lightweight approach.
Figure has chosen the most difficult entry point: the home.
Home environments have far more random object placement than factory assembly lines. Children and pets create unexpected situations. Repairs are more troublesome. Ordinary users won't tolerate lengthy debugging and waiting. All of these are harder to handle than industrial deployments.
The $20,000 price point: cheaper than a Tesla Model 3, but significantly more expensive than high-end home appliances. Who will ultimately pay remains to be seen.
However, Helix's three-layer architecture and Figure's decision to abandon the "use LLMs for everything" approach currently make it one of the companies with the clearest technical path in this sector.
Sources: Figure AI Series C Announcement (figure.ai); The Domestic Robotics Revolution: Figure AI Unveils Figure 03 and the Helix Unified Brain (Financial Content / TokenRing); CocoLoop, Figure AI: The $39B humanoid robot startup explained (eesel.ai)