Executive Intelligence · Startup Spotlight | October 4, 2026
Figure AI entered a market that had plenty of AI model competition and AI infrastructure investment, but almost no proof that humanoid robots could move from viral demonstrations to commercially viable industrial deployment. This Startup Spotlight examines whether Figure can convert early leadership in physical AI into a durable platform for enterprise robotics — or whether humanoid robotics becomes a capital-intensive race that larger manufacturers ultimately absorb.
Figure AI was founded in 2022 and reached a $39 billion valuation within three years. It has raised $3.68 billion, built a manufacturing facility capable of producing 12,000 robots per year, and deployed humanoid robots at BMW's Spartanburg plant in a commercial engagement that supported the production of more than 30,000 vehicles. But its most recent capital event — a $3.5 billion compute commitment with Nscale — exceeds its total lifetime fundraising, and it has never disclosed a revenue figure.
The strategic question is not whether Figure can build a humanoid robot that walks, balances, and manipulates objects. It can. The question is whether Figure can convert early leadership in general-purpose humanoid robotics into a durable commercial platform for industrial labor — or whether humanoid robotics becomes a capital-intensive hardware race that larger manufacturers or Chinese competitors ultimately win on scale and cost.
1. Why Humanoid Robotics Is Becoming More Important
The manufacturing labor landscape has undergone a structural transformation over the past decade, and the pace is accelerating. Three converging forces are driving demand for general-purpose humanoid robots.
Chronic labor shortages in manufacturing and logistics. Aging workforces across the US, Europe, Japan, and South Korea have created persistent vacancies in physically demanding, repetitive roles — the exact tasks that are hardest to automate with fixed machinery and easiest to justify with a general-purpose robot.
AI's transition from digital to physical environments. Foundation models have made enormous progress in perception, planning, and language understanding. The next frontier is embodied AI — systems that act in the physical world. Humanoid form factor is attractive because it can operate in environments already designed for humans without requiring factories to be re-engineered.
The economics of general-purpose automation. Fixed automation is expensive to deploy and inflexible once installed. A general-purpose humanoid that can be reassigned to different tasks — sequencing, machine tending, kitting, inspection — changes the ROI calculation for manufacturers who cannot justify single-purpose robots for every task.
Barclays Research estimates the global humanoid robotics market will grow from $2–3 billion today to $200 billion by 2035 under its most optimistic scenario. More conservative estimates put the 2035 market at $40 billion in a base case. Counterpoint Research projects cumulative humanoid robot shipments will exceed 100,000 units by 2028 — a seven-fold increase from 2025 levels — with a 73% CAGR during 2026–2035.
The CODEW Lens: Humanoid robotics used to be a research curiosity. In the AI era, it becomes an operational lever — because a general-purpose robot can be redeployed across tasks that fixed automation cannot economically address.
2. What Figure AI Does
Figure's platform addresses a deceptively simple question that most manufacturers cannot answer: Can a single robot perform the physical tasks we currently need humans for, without being re-engineered for every new job?
The company started by building a humanoid robot — working out the mechanical design, actuation, battery, and control systems required for a bipedal machine that can walk, balance, and manipulate objects. That foundation has expanded into a full-stack physical AI platform spanning several distinct capabilities.
Figure 03 humanoid robot. The core hardware platform. Redesigned cameras, palm vision, tactile hands, wireless charging, a softer exterior, and an in-house battery architecture. The robot is designed for industrial environments, not demonstrations.
Helix vision-language-action AI. Figure's proprietary VLA system that links visual observations and language instructions to continuous humanoid motion. Helix 2.5 extended neural control to the entire robot — walking, manipulating, and balancing as one continuous system — and demonstrated zero-shot generalization across 30 unseen Bay Area homes.
Index data platform. A global crowdsourcing platform that pays ordinary people to record first-person operational videos using smartphones or head-mounted devices. 264,000 app downloads across 108 countries, 44,000 weekly active contributors, and more than 16 million uploaded videos as of 2026.
BotQ manufacturing facility. A high-volume production facility in San Jose that has become central to Figure's production scaling strategy. Over 150 networked workstations with custom manufacturing execution software, and each robot undergoes more than 80 functional verification tests before sign-off.
The CODEW Lens: Figure did not start as an AI company. It started as a robotics company and built the AI because outsourcing the model to a third party meant surrendering the robot's brain. That ordering matters — it is much harder to move from AI security backwards into hardware than the other way around.
3. The Physical AI Opportunity
The emergence of foundation models and autonomous agents represents both the largest opportunity and the most significant challenge for robotics. Figure's thesis is that AI changes the fundamental nature of physical labor in five ways.
Robots can be programmed with language, not code. Traditional industrial robots require engineers to write explicit motion plans. A VLA system allows a supervisor to describe a task in natural language and have the robot attempt it — a fundamental shift in how automation is deployed and reassigned.
Data scales faster than hardware. Figure's Index platform collects human behavioral video at a rate that does not depend on how many robots it has manufactured. If the human-to-humanoid transfer scaling law holds, model quality improves predictably with data volume — a structurally different growth curve than robot-only data collection.
General-purpose beats single-purpose at the margin. A humanoid that can sequence parts, tend machines, and inspect output can be reassigned as production needs change — a flexibility that fixed automation cannot match without significant re-engineering.
Labor shortages create urgency. Manufacturers facing persistent vacancies in physically demanding roles have a stronger incentive to deploy general-purpose robots than they did when labor was abundant.
The supply chain is the bottleneck, not the AI. Chips are only 5–10% of the bill of materials. The other 90% consists of actuators, gearboxes, motors, sensors, and structural systems — none of which follow Moore's Law. A foundation model breakthrough does not make the mechanical stack cheaper or easier to produce at volume.
The CODEW Lens: Every physical AI problem eventually resolves into a hardware problem. You cannot govern what a robot does without knowing what its actuators and sensors can physically accomplish — and most enterprises do not know that today.
4. The Business Model
Figure operates an enterprise hardware plus software model with several distinctive characteristics: a vertically integrated stack, a direct enterprise sales motion, and a manufacturing-led scaling strategy.
Target customers. Large manufacturers in automotive, logistics, and industrial production. The flagship customer is BMW Group, which deployed Figure 02 at its Spartanburg plant for ten months and supported the production of more than 30,000 BMW X3 vehicles. Figure 03 arrived at BMW's Hall 52 in June 2026.
The company has not published a standard price list. Available benchmarks from competitors include:
| Company | Model | Price Benchmark |
|---|---|---|
| Agility Robotics | Digit v5 | ~$200,000 per robot + $20,000 deployment |
| Unitree | G1 | ~$16,000 (research/education) |
| IDTechEx projection | Industry average | ~$114,700 (2024) → ~$37,000 (2030) |
Recurring revenue. Figure has not disclosed its revenue model in detail. CEO Brett Adcock has indicated the company will offer a home rental scheme at several hundred dollars per month, suggesting a shift toward subscription-like pricing for consumer applications. Enterprise pricing is custom-quoted.
Expansion opportunities. Each new deployment adds training data and operational experience that can be applied to the next customer. The Index platform creates a data flywheel that compounds with scale — a structurally better position than selling one-off robots without a learning loop.
The CODEW Lens: Figure's pricing is not yet indexed to labor displacement — the metric that matters most for enterprise ROI. Until that happens, the business model remains a capital-intensive hardware bet with an unproven software margin.
5. Funding & Capital
Figure's funding history is remarkable for both velocity and scale. The company has raised more capital in three years than most robotics companies raise in a decade.
| Round | Date | Amount | Valuation | Key Investors |
|---|---|---|---|---|
| Series A | May 2023 | $70M | — | Parkway VC |
| Series B | Mar 2024 | $675M | $2.6B | Nvidia, Microsoft, OpenAI Startup Fund, Jeff Bezos, Intel Capital, Qualcomm Ventures, Salesforce, T-Mobile, LG, Macquarie |
| Series C | Sep 2025 | $1B+ | $39B | Nvidia, Microsoft, Brookfield, Intel Capital, Qualcomm Ventures, Salesforce, T-Mobile, LG, Macquarie |
| Nscale Compute Deal | Sep 2026 | $3.5B+ commitment | $39B | Nscale (became Figure shareholder) |
Total funding now exceeds $3.68 billion. The syndicate includes many of the most prominent venture firms in AI and enterprise software, alongside strategic investors such as Nvidia, Microsoft, Intel Capital, Qualcomm Ventures, and LG.
What the funding enables. The Nscale partnership commits at least $3.5 billion in compute spending, with intent to scale beyond $6 billion, covering up to 100,000 Nvidia Vera Rubin GPUs. Figure has committed more than $1 billion to data and compute over the next 12 months.
Capital intensity. Figure's growth strategy is capital-intensive. It runs a direct enterprise sales motion, operates its own manufacturing facility, and is investing heavily in data collection and compute. The company remains unprofitable and has not disclosed revenue. The Nscale deal — $3.5 billion against $3.68 billion total lifetime fundraising — suggests either ambitious near-term revenue projections or a financing structure that warrants scrutiny.
The CODEW Lens: Figure's funding trajectory is a bet on category ownership, not current economics. The company is buying speed — in data, manufacturing, and AI capability — because it believes humanoid robotics will consolidate around two or three platforms. The Nscale arrangement resembles the circular financing patterns that have drawn scrutiny in the AI industry.
6. The Competitive Landscape
Figure competes across several overlapping categories, and the dynamics are fluid rather than static. The table below positions the company by category rather than by ranking.
| Category | Representative Players | Figure's Position |
|---|---|---|
| US humanoid peers | Tesla Optimus, Boston Dynamics, Agility, Apptronik, 1X | Only US company with confirmed external enterprise deployment at scale (BMW) |
| Chinese humanoid leaders | Agibot, Unitree, UBTECH, XPeng Robotics | Premium positioning vs. cost leadership; supply chain diversification away from China |
| Industrial automation incumbents | ABB, Fanuc, KUKA, Yaskawa | General-purpose flexibility vs. fixed automation reliability |
| Foundation model providers | OpenAI, Google DeepMind, Nvidia | In-house Helix model; terminated OpenAI partnership in 2025 |
| Large manufacturing platforms | Tesla, Hyundai/Boston Dynamics, Amazon | Positions as hardware+AI supplier rather than internal automation division |
The competitive landscape is not zero-sum. Many manufacturers will deploy multiple robotic systems, and Figure's integrations — with BMW, Nscale, and LG Innotek — position it as a complement to broader automation investments rather than a wholesale replacement.
The CODEW Lens: Figure's most dangerous competitor is not Tesla Optimus. It is the combination of Chinese manufacturers with a 100-kilometer supply chain and cost leadership, and incumbent automation vendors who already have the factory floor relationships.
7. Figure's Competitive Advantage
It is important to separate what Figure has demonstrated from what it claims or projects.
Demonstrated advantages. Figure's manufacturing execution is independently verifiable. BotQ's production rate increased from 1 Figure 03 per day to 1 per hour — a 24x throughput improvement — in under 120 days, with over 350 units delivered by late April 2026. By July 23, 2026, BotQ reached cumulative production of 1,000 Figure 03 units. First-pass yield at end of line exceeds 80%, and the battery line has achieved 99.3% first-pass yield.
Commercial validation provides a second demonstrated advantage. The BMW deployment is the only confirmed external enterprise engagement at scale in the Western humanoid industry. Figure 02 supported the production of more than 30,000 BMW X3 vehicles over ten months, and Figure 03 moved to the more complex sequencing use case in June 2026.
Company claims and future potential. Figure's positioning as the "first commercially viable autonomous humanoid robot" is a strategic narrative that has not yet been proven at scale. The Helix 2.5 zero-shot generalization results — 30 unseen homes, no fine-tuning — are impressive but unpublished in peer-reviewed form. The human-to-humanoid transfer scaling law, if replicable, would be a significant technical advance, but the domain gap between human video and robot execution remains unproven at commercial scale.
Similarly, the expansion into home environments through the rental scheme is a significant bet on consumer adoption. The thesis — that a general-purpose humanoid can perform useful work in unstructured domestic settings — is compelling, but it depends on reliability and cost levels that have not been demonstrated outside controlled environments.
The CODEW Lens: The demonstrable advantage is manufacturing execution and commercial deployment. The claimed advantage is becoming the general-purpose physical AI platform. One is verified today. The other is a bet on where the market goes next.
8. Market Expansion
Figure's trajectory suggests a deliberate strategy to expand from industrial humanoid deployment into a broader physical AI platform.
Industrial manufacturing. The most significant near-term expansion vector: sequencing, machine tending, kitting, inspection, and material handling. The BMW deployment is the reference case, and Figure has stated it plans to expand within automotive and into adjacent industrial verticals.
Home environments. The rental scheme at several hundred dollars per month signals intent to expand beyond industrial applications. The Helix 2.5 zero-shot generalization results — tidying living rooms, folding towels, making beds across 30 unseen homes — are a direct response to this market.
Compute infrastructure. The Nscale partnership extends Figure into the AI cloud layer. The companies agreed to explore using humanoids to scale Nscale's supply chain — a recursive arrangement in which Figure provides robots and Nscale provides compute.
Data platform. The Index platform is a standalone asset that could be licensed or expanded beyond Figure's own robots. Figure plans to commit over $1 billion to data and compute over the next 12 months.
Supply chain localization. CEO Brett Adcock has stated the company plans to eliminate nearly all Chinese components from its supply chain by summer 2026. This is a significant undertaking that carries both strategic independence and execution risk.
The CODEW Lens: Figure is expanding along the same axis that Tesla used in electric vehicles — start with a premium industrial product, then move down-market as costs fall. The question is whether humanoid robotics has the same cost curve as batteries.
9. The Risks
Domain gap risk. If human video data cannot transfer effectively to robot hardware — because the kinematics, sensing, and actuation differ fundamentally — Figure's data advantage becomes less valuable. The human-to-humanoid transfer scaling law is a claim, not a proven theorem.
Supply chain fragility. The global supply chain for precision actuators, cycloidal reducers, and rare-earth magnets is underbuilt relative to deployment ambitions. Figure's plan to remove Chinese suppliers by mid-2026 adds execution risk. On Tesla's Q1 2025 earnings call, Musk acknowledged that Optimus production was directly limited by China's rare earth magnet export controls — "a policy decision about dysprosium and terbium moved faster than any model update".
ROI uncertainty. At current prices and reliability levels, the economics of humanoid deployment remain marginal for many use cases. If ROI does not turn positive within 18–24 months of deployment, enterprise customers may delay orders. The industry is converging on a critical metric: Value per Token — measuring the actual value a robot generates per unit of computational work.
Labor resistance. Hyundai Motor's 40,000-member Korean union has warned that "not a single humanoid robot will be allowed on the production lines without a labor-management agreement". Union opposition in automotive manufacturing could slow adoption even when the technology works.
Capital intensity. The $3.5 billion compute commitment against $3.68 billion total raised suggests Figure is operating with significant financial leverage. Any delay in commercialization could create funding pressure. The company remains unprofitable and has not disclosed revenue.
Competition from Chinese manufacturers. In the first half of 2026, Chinese companies shipped the overwhelming majority of the world's humanoid robots. Agibot led globally with approximately 9,700 units, followed by Unitree with over 7,000 units. Unitree's G1 at ~$16,000 represents a deliberate manufacturing-over-IP strategy designed to capture market share through cost leadership.
Valuation risk. A $39 billion valuation with undisclosed revenue and no profitability creates pressure for continued hypergrowth. A deceleration could trigger a valuation reset of the kind other high-flying hardware startups have experienced.
Execution risk. Scaling manufacturing, expanding geographically, integrating new AI capabilities, and managing a complex supply chain — all while developing new products — is a significant operational challenge.
The CODEW Lens: The biggest risk is not that Figure fails to build a humanoid robot. It is that Figure builds an excellent humanoid robot and still loses the category to a competitor that wins on cost, scale, or supply chain — not on AI capability.
10. What to Watch
Revenue disclosures. Figure has not published absolute revenue figures. Any disclosure of revenue or unit economics would provide critical validation of the business model.
Helix 2.5 replication. Whether the zero-shot generalization results and the human-to-humanoid transfer scaling law can be independently replicated will be a key indicator of Figure's technical advantage.
BMW expansion. Whether Figure expands within BMW's production network — and whether other automotive manufacturers follow — will test the commercial thesis.
Nscale compute deal structure. The $3.5 billion commitment warrants scrutiny. Whether this is a genuine strategic partnership or a circular financing arrangement will shape how the market values Figure.
Supply chain localization. Figure's plan to remove Chinese components by mid-2026 is a significant execution test. Success would reduce geopolitical risk; failure could delay production.
Chinese competitive response. Unitree's A-share listing, Agibot's shipment leadership, and UBTECH's revenue growth will shape the global competitive landscape. Whether Chinese manufacturers can translate cost leadership into Western enterprise deployments is an open question.
Labor negotiations. Union dynamics in automotive manufacturing — particularly Hyundai's warning — will shape the pace of adoption.
The CODEW Lens: The single most important data point to watch is whether Figure's robots are deployed in revenue-generating production roles — not pilot programs, not demonstrations — and whether customers renew and expand after the initial deployment.
The CODEW Take: Can Figure AI Become the Platform for Physical AI?
Can Figure AI become the platform for general-purpose physical AI as manufacturers seek to automate tasks that fixed machinery cannot economically address?
The answer is likely yes — but the company that emerges will be judged on two things it has not yet proven: whether humanoid robots deliver positive ROI in commercial deployments, and whether Figure can win the manufacturing scale race against Chinese competitors and larger incumbents.
Figure has built something genuinely difficult: a vertically integrated humanoid robotics company with in-house AI, a manufacturing facility capable of producing 12,000 robots per year, and a commercial deployment at BMW that supported the production of more than 30,000 vehicles. That foundation is real, and it is the prerequisite for everything else — you cannot deploy general-purpose robots at scale without the ability to manufacture and support them.
The expansion into home environments and compute infrastructure is strategically logical. Data and compute are the two inputs that determine how quickly a physical AI system improves. If Figure owns both — through Index and Nscale — it becomes a platform rather than a hardware vendor. That is a much more defensible position — and a much larger market.
But the risks are structural. Figure's valuation assumes continued hypergrowth and eventual profitability, with no public revenue figures to validate either. Its most formidable competitors — Chinese manufacturers with a 100-kilometer supply chain and incumbent automation vendors with existing factory relationships — have cost and distribution advantages that Figure cannot match. And the physical AI market it is betting on could consolidate around the manufacturers who already own the factory floor.
The CODEW verdict: Figure AI is the most credible independent challenger in humanoid robotics today, with the technology, capital, manufacturing capability, and customer proof points to become a foundational layer of the physical AI stack. The open question is not capability — it is economics. Humanoid robots must deliver positive ROI in commercial deployments, not just impressive demonstrations. If they do, Figure is positioned to define the category. If they do not, Figure becomes an acquisition target rather than a platform.
The three sources of potential advantage — manufacturing execution, data flywheel, and vertical integration — are all present. The question is whether they compound into a platform or dissolve into a collection of expensive capabilities that larger manufacturers replicate for less. The next 24 months will tell.
The CODEW Lens: Figure is not betting on being the best humanoid robot. It is betting that in a physical AI era, whoever controls the data flywheel and the manufacturing scale controls the market. Owning the learning loop is a more durable position than owning any single robot design built on top of it.
The Figure AI Glossary
VLA (Vision-Language-Action) — An AI system that links visual observations and language instructions to continuous physical motion. Figure's Helix is a VLA system.
Physical AI — AI systems that operate in the physical world through robots, autonomous vehicles, and embodied agents, as distinct from purely digital AI.
Zero-shot generalization — The ability of an AI system to perform tasks in environments it has never seen before, without fine-tuning or adaptation.
Human-to-humanoid transfer — The hypothesis that AI models pretrained on human behavioral video can effectively control humanoid robots, despite differences in kinematics and sensing.
BotQ — Figure's high-volume manufacturing facility in San Jose, with over 150 networked workstations and a production capacity of 12,000 robots per year.
Index — Figure's global crowdsourcing platform that pays contributors to record first-person operational videos for AI training.
Sequencing — A manufacturing logistics task in which a robot must handle parts that arrive in unpredictable positions, requiring perception and correction on the fly.
Cycloidal reducer — A precision gear mechanism used in robot joints. Figure has filed seven cycloidal reducer patents as a hedge against harmonic drive supply constraints.
Actuator — The motor-and-gearbox assembly that moves a robot joint. Actuators account for roughly 50% of humanoid production costs.
Value per Token — An emerging metric measuring the actual value a robot generates per unit of computational work, not simply what the robot costs.
First-pass yield — The percentage of manufactured units that pass all quality checks without rework. BotQ's end-of-line first-pass yield exceeds 80%.
Domain gap — The difference between training data (e.g., human video) and deployment conditions (e.g., robot execution), which can limit the effectiveness of transfer learning.
FAQ
Q: What does Figure AI actually do?
Figure builds general-purpose humanoid robots and the AI systems that control them. Its Figure 03 robot is designed for industrial environments, and its Helix vision-language-action AI system allows robots to be programmed with natural language rather than explicit motion plans. The company operates its own manufacturing facility (BotQ) and a crowdsourced data platform (Index).
Q: How much has Figure AI raised, and at what valuation?
Figure has raised $3.68 billion in total across four rounds. Its Series C in September 2025 raised $1B+ at a $39 billion valuation. In September 2026, Figure committed at least $3.5 billion in compute spending to Nscale, with intent to scale beyond $6 billion. Investors include Nvidia, Microsoft, the OpenAI Startup Fund, Jeff Bezos, Brookfield, Intel Capital, Qualcomm Ventures, Salesforce, T-Mobile, LG, and Macquarie.
Q: Is Figure AI profitable?
Figure has not disclosed profitability, and its capital intensity — a direct enterprise sales motion, its own manufacturing facility, and heavy investment in data and compute — suggests the company is prioritizing growth over near-term margin. It remains unprofitable based on the information available, and it has never disclosed a revenue figure.
Q: Who are Figure AI's main competitors?
In the US, the closest competitors are Tesla Optimus, Boston Dynamics, Agility, Apptronik, and 1X. Globally, Chinese manufacturers Agibot, Unitree, and UBTECH are the volume leaders. Figure also competes indirectly with industrial automation incumbents including ABB, Fanuc, KUKA, and Yaskawa, and with foundation model providers such as OpenAI and Google DeepMind.
Q: What are Figure AI's biggest risks?
The biggest risks are domain gap risk (human video not transferring effectively to robot hardware), supply chain fragility (actuators and rare-earth magnets), ROI uncertainty in commercial deployments, labor resistance from unions, capital intensity (the $3.5 billion Nscale commitment against $3.68 billion total raised), competition from Chinese manufacturers with cost leadership, and valuation risk with undisclosed revenue against a $39 billion valuation.
Q: Why does Figure AI matter for the AI era specifically?
Because physical AI represents the next frontier after digital AI. Foundation models have transformed perception, planning, and language understanding. Figure's argument is that general-purpose humanoid robots can bring those capabilities into the physical world — performing tasks that fixed automation cannot economically address. If that argument holds, physical AI becomes a foundational technology for manufacturing and logistics, not just a research curiosity.
The CODEW Stat
$3.68B raised · $39B valuation · 1,000+ robots produced Figure AI has raised $3.68 billion, reached a $39 billion valuation, and produced more than 1,000 Figure 03 robots — with a manufacturing capacity of 12,000 units per year and a commercial deployment at BMW that supported the production of more than 30,000 vehicles. The capital is real. The customer footprint is real. What remains unproven is whether humanoid robots deliver positive ROI in commercial deployments — and whether Figure can win the manufacturing scale race against Chinese competitors and larger incumbents. That is the central question of the Figure AI thesis.
Reviewed by Erwin Castro
on
Sunday, October 04, 2026
Rating:
No comments: