Figure AI Review 2026: Team, Investors, History, Robots & $39B Valuation

Figure AI has earned its place in the top group of humanoid robotics companies. Since 2022 it has built three robot generations, developed an in-house vision-language-action system, opened a manufacturing line and put robots into live production work at BMW. That is a serious record for a four-year-old company. It is also nowhere near enough to validate a $39 billion valuation on operating results alone.

Our verdict: Figure is one of the most technically credible and commercially ambitious humanoid startups, but the public evidence is lopsided. The BMW deployment is real and measurable. The BotQ production ramp is meaningful but self-reported. Home demonstrations show breadth, not consumer readiness. Revenue, unit economics, fleet uptime, intervention rates and current customer concentration remain largely undisclosed. Investors are paying for the possibility that Figure becomes a full-stack robotics platform, not for a mature business that already exists.

This review covers the company behind the robots: its history, founder, team, funding, investors, AI strategy, manufacturing plan, customers and risks. Detailed Figure 03 specifications, price rumors and release-status updates stay in the separate product review. For the catalogue view, see the Figure AI manufacturer record.

Figure AI quick facts

What Figure AI actually does

Figure is building general-purpose humanoid robots for physical work. The basic thesis is straightforward: homes, warehouses and factories were designed around the human body, so a machine with legs, arms and hands can use existing doors, shelves, tools and workstations without every site being rebuilt for automation. Figure initially emphasized structured jobs in manufacturing and logistics. Figure 03 and Helix have widened the story to household work.

The company is unusually vertically integrated. It designs the robot body, hands, actuators, battery, sensing and charging system; trains the Helix AI models that interpret instructions and control movement; operates data-collection programs; and runs BotQ manufacturing. A defect found on a factory shift can therefore inform a mechanical redesign, a new test at the end of the assembly line and the next training cycle. In theory, more deployed robots create more data, which improves the models, which makes the robots more useful and easier to sell.

What Figure has not explained publicly is the commercial wrapper. It has not published a hardware price, lease rate, robot-as-a-service fee, maintenance contract, gross margin or expected payback period. BMW has not disclosed the financial terms of its relationship. The business may eventually combine hardware, recurring software and service revenue, but that is a plausible model rather than a disclosed one. Today, the most defensible description is a private robotics developer moving from paid industrial trials toward repeatable commercial deployments.

That distinction matters. A robot can complete a useful task and still be a poor product if it needs too much human supervision, fails often, takes too long to repair or costs more per productive hour than the alternatives. Figure publishes more engineering detail than many startups, but the numbers needed to judge the business are still thin.

Figure AI history: from basement idea to BotQ

The speed is the striking part of this history. Figure launched publicly with 40 employees in March 2023, reported 50 by its Series A that May and 80 by the Series B in February 2024. TIME reported a 360-person company in October 2025. Within the same period, Figure moved through three complete robot designs and shifted from an external AI collaboration to its own embodied-AI stack.

The sequence also shows how the strategy changed. Figure 01 was pitched primarily at warehouses, manufacturing and retail. Figure 02 became the industrial proof point at BMW. Figure 03 was redesigned with softer materials, wireless charging, quieter operation and a more domestic appearance, even though the strongest evidence still comes from factories. Figure is trying to preserve the near-term industrial route while aiming at the much larger, much harder home market.

Founder Brett Adcock

Brett Adcock is Figure's founder and chief executive. Before humanoid robots, he co-founded the recruiting marketplace Vettery, which was sold to Adecco in 2018 for about $110 million. He then co-founded electric-aircraft company Archer Aviation, which went public through a SPAC transaction in 2021. Those companies gave him experience raising capital, recruiting technical teams, selling an enterprise platform and taking a hardware story to public markets.

Figure's master plan, dated May 20, 2022, frames humanoids as a response to labor shortages and unsafe or undesirable work. Adcock's argument is that a general-purpose machine should eventually spread its development cost across many tasks rather than remain trapped in one narrow application. He has since expanded the claim to homes and has predicted that humanoids could become one of the world's largest markets. Those forecasts explain the scale of Figure's fundraising, but they should be read as founder ambition, not demand forecasts backed by disclosed orders.

Adcock is a high-velocity operator and an unusually effective fundraiser. Figure's product cadence supports that reputation. It also creates key-person risk. TIME reported that a group of colleagues followed him from Vettery and Archer into Figure, including senior operations and growth leaders. Business Insider reported in May 2026 that he had also launched and become CEO of a new AI hardware company, Hark, after raising $700 million. Running another heavily funded company raises a fair question about attention, governance and how responsibilities at Figure are divided.

There is also a disputed point in his prior history. Adcock told TIME that he chose to leave Archer in 2022 to start Figure; an Archer spokesperson said his departure followed a board decision, while Figure said the resignation was voluntary. The disagreement does not determine Figure's prospects, but it belongs in a complete founder profile rather than a sanitized biography.

Leadership, team and hiring

Figure does not publish a current executive-team page. Its company site names Adcock and describes the collective team's experience, while the careers page lists open roles. That opacity makes it easy for third-party profiles to preserve old titles. The clearest example is Jerry Pratt: Figure identified the veteran roboticist as CTO in 2023, but TechCrunch reported in June 2024 that he had left to co-found Persona AI. He should not be presented as Figure's current CTO.

The team has grown quickly, but the best point estimate is dated. TIME counted 360 people around the Figure 03 launch in October 2025. LinkedIn currently places Figure in its broad 201-500 employee band. Neither figure tells us how many people are permanent employees, contractors or hourly robot operators, and Figure has not published a July 2026 headcount. A precise current number would be false confidence.

The hiring mix is more revealing than the total. Figure's current openings span reinforcement learning, whole-body control, AI training infrastructure, electrical and actuator engineering, manufacturing, quality, health and safety, deployment logistics, field service, robot operation and commercial site leadership. Roles in Reno and Los Angeles, alongside San Jose, suggest the company is building the operating layer needed to stage, deploy and support robots outside headquarters.

Figure's early releases emphasized recruits from Boston Dynamics, Tesla, IHMC, Google X, DeepMind, Cruise, Apple and Archer. That pedigree is useful, but brand-name résumés do not replace current accountability. For a company preparing machines to work near people, a published leadership roster covering engineering, AI, manufacturing, safety and commercial operations would materially improve transparency.

Corporate structure and the board disclosed in 2024

Figure describes itself as founded in 2022, which is when Brett Adcock published the master plan and began assembling the company. The legal issuer used for the Series B was Figure AI Inc., a Delaware corporation whose Form D lists 2023 as its year of incorporation. Those statements are compatible: a company can begin operating before the financing entity is formally incorporated. The distinction is worth recording because company histories often collapse the operating start date and legal-entity date into one number.

The March 2024 Form D is also the best public snapshot of Figure's formal governance. It listed Brett Adcock as both an executive officer and director; Dana Berlin, Lee Randaccio and Logan Berkowitz as executive officers; and Colby Adcock, Jesse Coors-Blankenship, Grant Hosking, Peter Welinder and Gregg Hill as directors. The filing did not give their job titles, committee assignments, voting rights or independence status. It should not be treated as a July 2026 board roster: Form D records the related persons relevant to that offering at the time it was filed.

The composition provides useful context for the Series B. It included the founder, operating executives, representatives connected to Figure's repeat lead investor and a senior figure from its then-current AI partner. That can help a young company move quickly and connect financing to technical resources. It does not tell outsiders whether the board had a lead independent director, a dedicated safety committee or formal succession arrangements.

Figure remains a private company, so it does not file annual proxy statements or public 10-K governance disclosures. It also does not publish a complete current leadership or board page. For a company valued at $39 billion, building machines intended to work beside people and potentially inside homes, that is a meaningful transparency gap. A current roster covering the board, senior engineering, manufacturing, safety, legal and commercial leadership would make accountability easier to assess.

Culture, hiring bar and retention risk

Figure's culture page is unusually explicit. It says the company will hire, recognize, reward and fire based on its values, warns that the culture is not for everyone and says it is not seeking candidates whose priority is minimizing workload or maximizing compensation. Its five published values are Move Fast & Be Technically Fearless, Product First Mission Focused, Aggressively Optimistic, Maximize Future Impact and Championship Mindset.

That operating philosophy helps explain the product cadence. Figure completed three robot generations, built an AI stack and stood up a factory system in roughly four years. The company also says it wants in-person collaboration, minimal bureaucracy and a team oriented around one mission. Clear expectations can be an advantage in a technically demanding startup because candidates know the intensity before joining.

The trade-off is organizational. A culture built around speed, exceptional commitment and aggressive optimism can make it harder to surface dissent, retain people through repeated hardware cycles or distinguish a prudent safety objection from resistance to execution. That does not establish a problem at Figure; it identifies a governance question that matters more in physical robotics than in ordinary software. The healthiest version of this culture would pair speed with strong stop-work authority, documented escalation paths and leaders rewarded for finding failures early.

Adcock reported in December 2025 that Figure had received 176,000 job applications over three years and hired about 425 people. If those self-reported figures use the same denominator, that is roughly a 0.24% hiring rate. It shows enormous applicant interest and a narrow funnel, but it is not an employee-retention metric. Figure has not published voluntary attrition, average tenure, offer-acceptance rates or the share of hires assigned to AI, hardware, manufacturing, deployment and robot operations.

Funding history

The disclosed rounds total more than $1.745 billion. That sum is based on company announcements and does not assume undisclosed seed financing. The progression is extraordinary: Figure's valuation moved from $2.6 billion at Series B to $39 billion post-money roughly 19 months later, a fifteen-fold step-up. Hardware progress and the Helix story advanced during that period, but public operating disclosures did not grow at the same rate.

Parkway Venture Capital is the central financial backer. It led the Series A and Series C and participated in the Series B. The repeat support matters because it signals sustained conviction across hardware generations. It also means the investor map is not a collection of one-off celebrity checks; Figure has a lead investor willing to finance the company through multiple stages.

The Series B brought a second kind of value. Microsoft offered Azure infrastructure; NVIDIA and Intel connected Figure to compute and silicon ecosystems; OpenAI supplied brand and an announced model-development partnership; Jeff Bezos invested through Bezos Expeditions. The OpenAI collaboration ended within a year, which is a useful warning: a strategic investor logo can create optionality without guaranteeing a lasting operating relationship.

Who invested in Figure AI, and why the mix matters

Figure's investors fall into four groups. Financial investors such as Parkway, Align, Tamarack, ARK, Aliya, Bold and FJ Labs are underwriting company value. Compute and component companies such as NVIDIA, Intel, Qualcomm, Microsoft and LG can support training infrastructure, chips, connectivity and manufacturing relationships. Brookfield can provide real estate, operating environments and potential deployment sites. Salesforce, T-Mobile and Macquarie bring enterprise distribution, communications and capital-markets reach.

Brookfield is the clearest example of money tied to an operating thesis. Its 2025 agreement covers human-video data collection across varied properties, infrastructure for Helix and potential commercial deployments. That could give Figure access to homes, offices, logistics sites and other real-world environments that are difficult for a robotics startup to assemble alone. It is still a partnership plan, not a published fleet result.

NVIDIA and Intel participated in more than one round, while Qualcomm and LG joined the Series C. Their presence is strategically sensible because humanoids require training compute, on-device inference, sensors, communications and power electronics. It does not mean those companies have validated Figure's unit economics or safety performance. Investors validate willingness to invest at negotiated terms; customers validate willingness to pay for useful work.

What the Series B securities filing adds

The Series B press release rounded the round to $675 million. Figure's SEC notice supplies the transaction mechanics: the company offered $675 million of equity under Rule 506(b), reported $674,981,967 sold, only $18,033 remaining and 45 investors in the offering. The first sale was recorded on February 27, 2024, and the notice was signed by Adcock as CEO on March 12.

The Form D is a notice of an exempt offering, not an audited financial statement. The SEC explicitly warns that it has not necessarily reviewed the information for accuracy or completeness. Still, it is a primary legal filing and more precise than copying the financing announcement. It confirms the investor count, exact amount sold, legal issuer and related persons while leaving revenue, share rights and operating results undisclosed.

Valuation, private stock and IPO status

The $39 billion figure is a post-money private valuation from the September 2025 Series C. It is not market capitalization, revenue or a daily price set by public trading. Private valuations can reflect preferred-share rights, negotiated terms and a small amount of capital relative to the full implied company value. Without audited financials, readers cannot calculate a credible revenue multiple or compare the valuation with mature robotics businesses.

The Wall Street Journal reported in April 2025 that investor materials showed no revenue in the prior year, only a few dozen robots in production and forecasts of more than 200,000 robots and $9 billion in revenue by 2029. Those figures predate the later Series C and BotQ ramp. They are useful as evidence of how aggressively the company was being marketed, not as current results or guidance that Figure has publicly reaffirmed.

Figure AI is not publicly traded and has no ticker. In July 2026 the company warned that unauthorized attempts to sell its shares may be invalid and said transfers require board approval. Secondary-market listings and special-purpose vehicles do not create the same liquidity, disclosure or investor protections as a public exchange. Figure has not announced a confirmed IPO date.

Private-share marks: useful evidence, not a public price

A few regulated investment funds disclose their Figure holdings. Those filings offer rare per-share reference points, but none is equivalent to a quoted market price. The securities are restricted, different preferred series may carry different rights, and the funds classify the positions as Level 3 or otherwise valued with significant unobservable inputs. A reported fair value is an accounting estimate at a date, not proof that the entire company could trade there.

The spread between $135 and about $157 for two Series B references illustrates why secondary headlines need care. The marks come from different holders and dates, and they may use different valuation procedures. The Series C figure is higher still, but preferred-share seniority and contractual rights can differ across rounds. Dividing a fund's value by its share count is informative; multiplying that result by an unknown fully diluted share count would not produce a defensible company valuation.

The filings do support one conclusion: some regulated funds have materially marked up Figure's earlier shares, and investor demand extends beyond the companies named in press releases. They do not show liquidity. Figure's own July 2026 notice says transfers require board approval and warns that unauthorized sale attempts may be invalid. A buyer in an SPV may own an interest in the vehicle rather than Figure shares directly, with additional fees, transfer restrictions and counterparty risk.

The $39 billion valuation also needs the right denominator. The 2025 investor materials reported by The Wall Street Journal projected $9 billion of revenue in 2029. Dividing $39 billion by that projection gives about 4.3 times a four-year-forward revenue target, not a current revenue multiple. The same valuation is 15 times Figure's $2.6 billion Series B value. Without current revenue, gross margin, cash burn and dilution data, neither comparison is enough to decide whether the Series C price was justified.

Figure's robot portfolio

The Figure 01 review covers the first platform. Its importance is historical: Figure says it took its first steps in May 2023, roughly a year after the company was founded. Figure 01 supported early proof-of-concept work, but it was not a commercial consumer product.

The Figure 02 review covers the model that moved into BMW's Spartanburg plant. Figure integrated the battery into the torso, improved the hands and refined the package for deployment. The company later described forearm electronics and dynamic cabling as reliability lessons that informed the Figure 03 wrist design.

Figure 03 is a ground-up redesign with lower-cost manufacturing processes, palm cameras, tactile fingertips, a lighter body, softer external materials and wireless charging. Those changes serve both the factory and home stories. For current dimensions, specifications, availability and price status, use the dedicated Figure 03 review and Figure 03 database record.

Helix: Figure's in-house AI strategy

Figure's strategic pivot in 2025 was to make robot intelligence a core internal product. The company had announced an OpenAI collaboration alongside the Series B in February 2024. By February 2025 Adcock said Figure would end the arrangement and rely on its own models. He argued that large language models were becoming more commoditized while high-frequency robot control remained the harder problem. Whatever the relationship's internal history, Helix made Figure responsible for the entire perception-to-action stack.

The first Helix release used two layers. System 2 was a 7-billion-parameter vision-language model that interpreted images and instructions at roughly 7 to 9 Hz. System 1 was an 80-million-parameter visuomotor policy that converted that context into upper-body action at 200 Hz. Figure said it used about 500 hours of teleoperated training data and controlled 35 degrees of freedom across the head, torso, wrists, hands and fingers.

Helix 02 added a third layer for whole-body autonomy. System 0 is a 10-million-parameter controller running at 1 kHz, trained in simulation using more than 1,000 hours of retargeted human-motion data and 200,000 simulated environments. System 2 selects and interprets the goal, System 1 maps perception to desired movement, and System 0 handles balance, contact and fast joint control. The design lets walking and manipulation become one continuous learned behavior rather than separate scripted modes.

The demonstrations are broad. Figure has shown robots sorting packages, putting away groceries, folding towels, loading a dishwasher, making a bed and coordinating a two-robot room reset. That breadth is relevant because general-purpose economics depend on reusing the same hardware and model across many tasks. The weak point is measurement. Figure rarely publishes task-success distributions, intervention rates or performance over long unscripted periods in homes.

TIME's visit before the Figure 03 launch gives useful context. Reporters saw towel-folding and laundry demonstrations fail before later succeeding, and Figure acknowledged that the robot was not ready for domestic use at launch. That does not negate the successful demonstrations. It shows why edited clips and best-case runs cannot stand in for reliability data. A home robot must handle clutter, children, pets, reflective surfaces, bad lighting, dropped objects and thousands of edge cases without a technician standing nearby.

Patents and the intellectual-property moat

Figure's public patent record has become substantial enough to examine. Justia's Figure AI assignee page listed 13 U.S. grants as of July 16, 2026, with grant dates from July 2025 through June 2026. The list spans industrial design and mechanical architecture, not merely a logo or outer shell. It includes hands and end effectors, head and neck assemblies, central-body and hip geometry, ankle mechanics, whole-body control and an AI action-model architecture.

The bipedal-action-model grant is particularly relevant to Figure's full-stack claim. It describes a larger model that processes language and a smaller model that produces robot-control output, trained in a unified framework. The named inventors include Corey Lynch, Yevgen Chebotar, Toki Migimatsu and Michael Ahn. Other grants identify a wider bench across mechanical design, hands, locomotion, industrial design and control.

The pending torso application adds details that marketing pages do not emphasize. Its claims cover a rechargeable battery in a 1.5 to 5 kWh range, internal airflow past compute or battery cooling devices, and energy-absorbing protection around an arm interface. Patent claims are drafted broadly and do not confirm the exact production configuration, but they show which engineering problems Figure considered worth protecting.

A patent portfolio is evidence of original work and a potential defensive asset. It is not proof of technical superiority, freedom to operate, enforceability or customer value. Competitors can design around claims, patents can be challenged, and the hardest advantage may sit in manufacturing know-how, fleet data and model-training operations that never appear in a patent. Figure's likely moat is the combination of hardware IP, tacit production knowledge, deployed-robot data and capital, not any single grant.

BotQ manufacturing

BotQ is Figure's attempt to solve the hardware scale problem in-house. Announced in March 2025, the facility was designed for initial capacity of up to 12,000 humanoids per year. Figure also set a goal of producing 100,000 robots over four years and eventually using its own robots in the manufacturing process. Capacity and goals are not the same as output, orders or customer deliveries.

The April 2026 production update was more concrete. Figure said it had produced more than 350 Figure 03 robots, improved final assembly from one robot per day to a one-hour cycle and reached end-of-line first-pass yield above 80%. It also reported 99.3% first-pass yield on the battery line, more than 500 battery packs, over 9,000 actuators and more than 80 functional tests on each completed robot.

Those numbers show a real production system, not a lab assembling one hero prototype. They are still company-reported and do not reveal the commercial mix. Figure allocates robots to AI research, data collection, home testing and commercial development, so more than 350 produced does not mean more than 350 sold. The decisive manufacturing metrics will be field failure rates, warranty cost, repair time, component life and the percentage of output accepted by paying customers.

Factory capacity, quality control and the service burden

BotQ's headline capacity is 12,000 robots per year, but capacity, demonstrated line speed and completed output are different measurements. Figure reported more than 350 Figure 03 units produced by April 29, 2026 and said it had demonstrated a one-robot-per-hour cycle time. One robot per hour sustained for one eight-hour shift on 250 working days would equal 2,000 robots a year; sustained around the clock for 365 days would equal 8,760. Reaching 12,000 would require more parallel capacity, a faster effective cycle or additional shifts and lines. Figure calls 12,000 the capability of its first-generation manufacturing line, not current annual deliveries.

The manufacturing architecture is more sophisticated than a prototype workshop. Figure says its custom manufacturing execution system links production data with more than 150 networked workstations, part genealogy and component test results. It uses product-lifecycle, enterprise-resource and warehouse-management software, qualifies suppliers against incoming inspection criteria and performs thousands of-cycle burn-in exercises on completed robots.

Figure 03 was redesigned around injection molding, die casting, metal injection molding and stamping instead of relying heavily on slow CNC machining. Figure says parts that took more than a week to machine can be produced in under 20 seconds with tooling. That shift is necessary for cost and volume, but it moves expenditure forward into molds, supplier qualification and line equipment. The economics work only if volume becomes large enough to amortize the tooling and designs remain stable long enough to use it.

Service may be as difficult as assembly. Figure says it has built field-service and fleet-management systems, over-the-air updates, alerting, failure analysis, fallback ladders and processes for fleet-wide upgrades and recalls. Those are the right systems for commercial deployment. They are also indirect evidence of the support burden: a robot in a customer's factory or home creates maintenance logistics, spare-parts inventory, software rollback, technician training and incident-response obligations that do not exist for an internal demo.

The missing unit-economic bridge is labor. Figure has not disclosed assembly hours per robot, service technicians per deployed unit, average repair time, warranty reserve, cost of spare parts or the share of robot-operating hours that require remote assistance. A one-hour factory cycle does not mean one labor-hour of total cost, and an 80% first-pass yield does not reveal the cost of bringing the remaining units to acceptance.

From eight hours of training data to an eight-hour shift

Package handling provides a useful view of how Figure develops a task. In February 2025 the company said eight hours of carefully selected human demonstrations were enough to train a Helix policy to grasp, reorient and move varied packages on a conveyor. The task forced the model to handle boxes and deformable bags, find labels, judge depth and timing, transfer across robots and recover when a grasp was imperfect.

In May 2026 Figure streamed three Figure 03 robots working through an eight-hour package-sorting shift under Helix 02. Adcock described the run as fully autonomous and at human performance levels, with packages oriented barcode-down for scanning. TechRadar reported that the robots appeared to complete the work mostly as required, noted a small number of exceptions in the footage its writer reviewed and recorded continuing skepticism from observers about speed and possible teleoperation.

The livestream is stronger evidence than a short edited clip because viewers can observe extended operation and imperfect cases. It is still not a customer audit. Figure did not publish the complete run log, intervention count, throughput distribution, downtime, energy use, robot swaps, package mix or exception taxonomy. Three robots taking turns also makes it important to separate cell throughput from per-robot throughput.

The right next step is not a longer promotional video. It is a dated operating table showing total packages attempted, success without human touch, recoveries, interventions, faults, mean repair time, energy per productive hour and throughput at each percentile. The same standard applies at BMW: the move from Figure 02 sheet-metal loading to Figure 03 sequencing is technically meaningful, but commercial judgment requires repeatable output, not capability alone.

Customers, deployments and partnerships

BMW is Figure's strongest customer reference. The companies announced a commercial agreement in January 2024 and began by identifying suitable automotive tasks. Figure later reported that Figure 02 robots worked 10-hour weekday shifts over an 11-month deployment, logged more than 1,250 operating hours, handled over 90,000 sheet-metal parts and contributed to production of more than 30,000 BMW X3 vehicles.

BMW independently confirmed the underlying production work. It described Figure 02 inserting sheet-metal parts into fixtures at Plant Spartanburg and later said the robot supported production of more than 30,000 X3 vehicles. Figure's detailed hours, step count and parts totals remain company-reported, but this is much stronger evidence than an unnamed pilot or a demo inside Figure's office.

Figure 03 arrived at the same plant in June 2026 for a sequencing workflow in Hall 52. The task combines part selection, dexterous handling, carrying and whole-body movement, including moving a wheeled cart. It appears more complex than Figure 02's earlier placement task. Figure has not yet published an equivalent multi-month report showing Figure 03 uptime, interventions and sustained cycle performance.

Catalyst Brands is the next named deployment path. Its May 2026 agreement targets physically demanding distribution and logistics work at a Reno facility. Catalyst's retail portfolio includes JCPenney, Aéropostale and Brooks Brothers. The announcement does not specify fleet size, commercial terms, deployment dates or measured results, so it should be classified as a signed agreement rather than a proven operating fleet.

Brookfield is an investor and strategic partner, not yet a comparable customer case study. Its agreement covers data collection across varied environments, AI infrastructure and potential deployments in Brookfield properties. The partnership may be strategically valuable because real homes and commercial buildings provide more diverse training data than a robotics lab. Public metrics will determine whether it becomes a deployment channel or stays at the development stage.

What is independently verified?

Figure's evidence is better than its critics sometimes imply and weaker than its marketing can make it appear. A named automaker allowed robots onto a live production line and later confirmed the work. That matters. The company has also published specific architecture and manufacturing numbers rather than relying only on cinematic videos. At the same time, most details originate with Figure and few have been audited.

The standard should rise as the valuation and claims rise. A startup proving its first robot can move parts deserves latitude. A $39 billion company discussing homes, mass manufacturing and hundreds of thousands of robots should publish fleet uptime, intervention frequency, safety incidents, customer concentration and economics per productive hour.

Business model and commercial prospects

Figure has chosen a sensible first market. Factories and distribution centers offer repeatable workflows, controlled access, professional maintenance and measurable labor costs. A robot does not need human-level general intelligence to create value if it can perform one high-frequency task safely for most of a shift. BMW provides Figure with an environment where reliability, cycle time and part quality are visible.

The industrial route can also finance learning. Each deployment produces failure cases, maintenance records and task data. If Figure can reuse that learning across customers without expensive site-specific engineering, margins should improve as the fleet grows. If every installation needs a large integration team and persistent teleoperation, the supposed general-purpose advantage shrinks.

Homes offer a much larger theoretical market and a harsher product test. Consumers need an affordable machine that is quiet, safe, privacy-preserving, easy to insure and useful across many chores. They will not tolerate factory-style safety cages, frequent technician visits or long calibration procedures. Figure 03's softer design and wireless charging are steps toward that product, but there is no public price, order page, service plan or broad release schedule.

Commercial success will be visible in a few simple metrics: repeat orders from named customers, growing paid fleet size, declining human interventions, high uptime, shorter repair cycles and a cost per useful hour below human or fixed-automation alternatives. Figure has not disclosed enough of these metrics to model the business with confidence.

Figure AI competitors

Figure's main advantage is integration plus deployment evidence. Tesla has far greater capital and manufacturing infrastructure, but its Optimus program still exposes limited external customer data. Agility has the strongest pure-play logistics posture and more named commercial deployments. Apptronik can use Jabil's global manufacturing network rather than build every production capability internally. Each model tests a different scaling strategy.

The home race is particularly revealing. 1X accepts orders at disclosed prices and explicitly describes a remote Expert Mode for tasks the robot cannot perform alone. Figure has published stronger industrial evidence but no consumer offer. Unitree sells lower-cost hardware to developers, which can build ecosystem and data even when the product is not a turnkey household worker. Figure is asking the market to believe it can bridge enterprise reliability and home generality before either market is mature.

Risks and unresolved questions

Valuation risk. A $39 billion post-money price assumes Figure captures a substantial share of a future humanoid market. Public revenue, gross margin, cash burn and backlog are absent. Even strong engineering progress can coexist with a future down round if commercialization takes longer than investors expect.

Safety risk. A humanoid is a heavy, powered machine operating around people. A former head of product safety sued Figure in federal court in November 2025, alleging he was fired after raising concerns about injury risks. Figure disputes the claims in active litigation; the court has not adjudicated them. The Northern District of California docket showed continuing filings through July 6, 2026. The allegations should not be treated as findings, but they make Figure's limited public safety documentation more consequential.

Home privacy risk. A useful household robot sees rooms, faces, possessions, routines and conversations. Adcock told TIME that Figure intended to use data from home robots to improve future models and described plans to scrub personal information. Figure has not published a detailed consumer data-retention, consent or model-training policy tied to a retail product because that product is not yet on sale.

Reliability risk. Home demos and short clips do not reveal failure distributions. BMW proves that a narrow task can run in production, but a general household worker needs far broader reliability. Figure should publish task success over hundreds of attempts, mean time between failures, remote-assistance rates and recovery behavior after dropped or misplaced objects.

Customer-concentration risk. BMW is the only customer with substantial public operating metrics. Catalyst is an announced agreement and Brookfield is a strategic pathway. Until several independent customers expand from pilots to repeat orders, Figure's commercial traction remains concentrated.

Manufacturing risk. BotQ's ramp is impressive, but producing hundreds of units is different from supporting tens of thousands in the field. Component durability, supplier quality, repair logistics, warranty reserves and spare-parts availability become harder as deployments spread.

Leadership and governance risk. Figure is founder-led, does not publish a current leadership roster and has already seen a prominent CTO depart. Adcock's simultaneous leadership of Hark increases the need for clarity about executive responsibility, conflicts, board oversight and how technology or people move between the companies.

AI-strategy risk. Bringing Helix in-house gives Figure control and differentiation, but it also concentrates model, data and safety responsibility inside one company. The ended OpenAI collaboration shows that strategic relationships can change quickly. Figure must fund frontier AI, robot hardware and manufacturing at the same time.

Safety lawsuit: procedural status in July 2026

A former principal robotics safety engineer, Robert Gruendel, sued Figure in November 2025, alleging retaliation after raising safety concerns. Figure disputes the claims and has brought related counterclaims. The case is Gruendel v. Figure AI, Inc., No. 5:25-cv-10094, in the U.S. District Court for the Northern District of California. The allegations are not findings of fact, and the case has not reached a merits judgment.

A July 6, 2026 discovery order addressed a collection of Figure documents retained by Gruendel. The magistrate judge directed Figure to review and produce responsive, non-privileged materials under specified procedures, ordered the collection preserved, set July 17 for production of non-privileged documents referenced in parts of the complaint and required the parties to submit a discovery plan by July 24. Fact discovery is scheduled to close November 14, 2026.

That order is procedural. It does not validate the complaint's safety allegations or Figure's counterclaims. Its relevance to this review is narrower: the dispute is active, discovery is under way and public information may change as documents are produced or later motions are decided. Any article that says the lawsuit was resolved, dismissed or proved without a later docket entry would be inaccurate.

The broader diligence issue remains safety governance. Figure says it created dedicated safety and reliability teams during the Figure 03 manufacturing redesign, and a granted control patent includes a safety-related limit. Those are positive signals. Public reporting still lacks incident-rate data, external safety certification, a named senior safety owner, red-team results for home use and a clear explanation of who can halt a deployment.

The diligence questions Figure still does not answer

Figure does not need to publish every customer contract or proprietary model detail. It could answer most of these questions with ranges, standardized fleet metrics and an annually refreshed governance page. Until then, the company deserves credit for technical velocity and real deployment evidence, while the valuation and scale claims deserve a larger discount for missing operating data.

What to watch next

The most important next disclosure is not another household video. It is a Figure 03 fleet report covering several months of paid work: robot count, scheduled hours, autonomous hours, interventions, task success, safety events, repair time and cost per productive hour. A comparable Catalyst report would show that BMW was not a one-customer exception.

Watch BotQ for customer shipment mix rather than gross output. Watch the careers page for field-service and deployment hiring outside California. Watch for a public safety framework before any broad home program, and for consumer terms that explain price, remote assistance, data use, warranty and liability. On the financing side, a tender, new round or IPO filing would reveal whether the private valuation can withstand greater disclosure.

Final verdict

Figure AI is a credible robotics company with an aggressive valuation, not vaporware and not a proven mass-market business. The team has moved unusually fast, BMW validates that its robots can perform useful production work, Helix gives the company a coherent AI architecture, and BotQ shows preparation for scale. Very few humanoid startups can point to all four.

The weak side of the case is commercial transparency. Figure does not disclose enough about revenue, pricing, paid fleet size, customer concentration, interventions, field reliability or unit economics. Its home story is several steps ahead of the evidence. The active safety litigation, sparse leadership disclosure and Adcock's second CEO role add governance questions that a comprehensive review cannot ignore.

So the rating is conditional: top-tier technical contender, early commercial operator, speculative valuation. Figure deserves close attention because it has built and deployed real machines. It does not yet deserve the assumption that mass household adoption, attractive margins or a $39 billion outcome are inevitable.

Methodology and source standard

We checked Figure's company, careers, product, AI, manufacturing, funding, partnership and stock-transfer materials; BMW's independent statements; contemporaneous reporting from AP, TIME, Forbes, TechCrunch, The Wall Street Journal and Bloomberg Law; the federal court docket; and official competitor materials. Company figures are labeled as company-reported unless a customer or independent source confirms them. Funding announcements establish transaction terms, not business performance. Allegations in active litigation are identified as allegations.

This page was fact-checked on July 16, 2026. Figure is private and the market moves quickly, so customer metrics, team size, financing and legal status may change. The dedicated robot pages remain the source for model-level specifications and availability updates.