Figure 03 at $20,000: The Robot That Could Break the Price-Capability Curve

In every emerging hardware category, there is a moment when a product arrives at a price point that reshapes the entire competitive landscape. The iPhone did it to smartphones. The Model 3 did it to electric vehicles. Figure AI is betting that the Figure 03 will do it to humanoid robots.

At $20,000, the Figure 03 is not the cheapest humanoid robot you can buy. That title still belongs to the Unitree G1 at $16,000. But the Figure 03 is arguably the first humanoid that combines industrial-grade capability with a price tag that makes labor substitution economics work outside of a venture capital pitch deck. It offers 42 degrees of freedom, 16-DOF hands with force sensing, the Helix vision-language-action foundation model, a 5-hour battery, wireless inductive charging, and ISO 10218 safety compliance. All for less than a new Honda Civic.

The question is not whether $20,000 is cheap. The question is whether Figure AI can actually build this machine at that price, at scale, without gutting the capabilities that make it worth buying in the first place.

What Figure 03 actually is

Figure 03 is the third-generation humanoid robot from Figure AI, introduced in October 2025 as a complete hardware and software redesign of the Figure 02. Where the 02 was a research-to-deployment bridge, the 03 is designed from the ground up for volume manufacturing and commercial deployment.

The robot stands 173 cm tall and weighs 61 kg, making it roughly human-sized. It carries a 20 kg payload, meaning it can lift and move objects that represent a significant portion of warehouse and assembly tasks. Eight cameras provide 360-degree RGB and depth perception. Connectivity includes Wi-Fi 6, 5G, and Bluetooth 5.3, giving it flexibility across factory, warehouse, and eventually home environments.

But two features define the Figure 03 and separate it from everything else on the market: the hands and the brain.

The hands: 16 DOF with force sensing

Each hand on the Figure 03 has 16 degrees of freedom with integrated force sensors at every fingertip. This is not a gripper with a wrist attached. These are fully articulated hands that can pick up a pen, thread a cable through a clip, or gently handle a fragile component without crushing it.

For context, the human hand has roughly 27 degrees of freedom across its joints. At 16 DOF per hand, the Figure 03 cannot match human dexterity, but it gets close enough for the vast majority of manufacturing manipulation tasks. Tesla’s Optimus Gen 2 offers 11 DOF per hand with tactile sensing. The Unitree G1 ships with a basic gripper in its base variant and an optional 5-finger hand in the EDU configuration.

Force sensing is what makes the difference between a robot hand that can pick things up and one that can actually handle them. Without force feedback, a robotic hand applies the same grip force whether it is holding an aluminum bracket or an egg. With it, the Figure 03 can modulate grip in real time, adapting to the weight, fragility, and surface texture of whatever it is holding. This is essential for mixed-task environments where a robot might handle rigid metal parts one minute and delicate electronics the next.

The brain: Helix vision-language-action model

The Figure 03 runs on Helix, Figure AI’s proprietary vision-language-action (VLA) foundation model. Figure ended its partnership with OpenAI in February 2025 and now develops all of its AI capabilities in-house.

Helix processes visual input from eight cameras, audio from onboard microphones, and proprioceptive data from every joint simultaneously. It can accept natural language commands, reason about spatial relationships, and execute multi-step manipulation tasks without explicit programming for each step.

During Figure 02’s deployment at BMW’s Spartanburg plant, the predecessor version of this system demonstrated the ability to learn new assembly tasks from roughly five human demonstrations. The Figure 03 iteration of Helix is reportedly faster at task acquisition and more robust at generalizing learned behaviors to novel situations.

The price: how Figure plans to hit $20,000

This is where skepticism is warranted. The Figure 02, deployed at BMW, is estimated to cost north of $150,000 per unit. Dropping from $150,000+ to $20,000 in a single generation is not normal in robotics, or in any hardware category.

Figure AI is betting on three structural changes to make this work: BotQ, in-house actuators, and wireless inductive charging.

BotQ: the dedicated robot factory

BotQ is Figure AI’s purpose-built manufacturing facility, designed exclusively for humanoid robot production. The company has publicly stated a target capacity of 12,000 humanoids per year. This matters because the primary cost driver in humanoid robots is not raw materials. It is manufacturing complexity, low production volumes, and the overhead of integrating dozens of subsystems that were never designed to work together.

BotQ is designed to change that equation. By building a factory specifically for humanoid production rather than repurposing an existing facility, Figure can optimize the entire assembly flow for its specific robot architecture. Think of it as the difference between a contract manufacturer who builds whatever walks in the door and a Tesla Gigafactory that is purpose-built for one product line.

At 12,000 units per year, Figure projects per-unit fixed cost allocation that makes $20,000 achievable. At 200 units per year (roughly their current volume), it absolutely is not. This is the single biggest risk in the $20,000 price claim: it only works at scale.

In-house actuators: cutting the middleman

Actuators, the motors and gearboxes that move every joint, typically account for 30 to 40 percent of a humanoid robot’s bill of materials. Most robotics companies buy actuators from third-party suppliers like Maxon, Harmonic Drive, or Chinese manufacturers. Each supplier adds margin, and off-the-shelf actuators are rarely optimized for the specific torque, speed, and form factor requirements of a particular robot design.

Figure AI produces its own actuators in-house. This is a significant capital expenditure decision. Designing, tooling, and manufacturing custom actuators requires deep mechanical engineering expertise and substantial upfront investment. But the payoff is threefold: lower per-unit cost by eliminating supplier margin, actuators optimized specifically for the Figure 03’s joint requirements, and supply chain independence that prevents the kind of component shortages that have plagued other robotics companies.

Wireless inductive charging: an operational cost play

The Figure 03 charges wirelessly via inductive charging, like a smartphone on a charging pad but scaled up dramatically. This seems like a convenience feature, but it is actually an operational cost play that directly supports the $20,000 price argument.

Wired charging requires a human operator or a complex docking system to plug in the robot. In a factory running 24/7, that is downtime and labor cost. Wireless charging lets a robot drive to a charging pad autonomously between tasks, with zero human intervention. It also eliminates the charging port, which is one of the most failure-prone components on any robot that operates in dusty, wet, or industrial environments.

From a manufacturing perspective, removing the charging port and its associated waterproofing, connectors, and cable management simplifies the robot’s housing design and reduces assembly steps. Small savings, but at 12,000 units per year, small savings compound.

How Figure 03 compares to the competition

The $20,000 price point places Figure 03 in direct competition with two very different machines: the Unitree G1, which is cheaper but less capable, and the Tesla Optimus Gen 2, which is more expensive and not yet commercially available at a confirmed price.

The Unitree G1 gap

At first glance, $4,000 separates the Figure 03 and the Unitree G1. But look at the spec sheet and these are fundamentally different machines built for different jobs.

The G1 stands 132 cm tall. That is 4 feet 2 inches, roughly the height of an 8-year-old child. It weighs 35 kg and carries a 3 kg payload. It is fast, agile, and affordable, but it is not built to do the kind of work that justifies a $25-per-hour labor substitution calculation. You cannot ask a G1 to lift a 15 kg engine component off a conveyor belt or reach a shelf at adult human height.

The Figure 03 is 173 cm tall and carries 20 kg. It fits into environments designed for adult human workers. It can reach the same shelves, operate the same tools, and handle the same loads. This is why the $20,000 price is significant: it is the first time a human-scale, industrial-capable humanoid has priced itself in the same conversation as what is essentially a research and education platform.

The Tesla Optimus question

Tesla has not publicly confirmed a commercial price for Optimus Gen 2, though Elon Musk has at various times suggested targets ranging from $20,000 to $25,000. Tesla’s Optimus offers 28 DOF in the body plus 22 DOF in the hands (Gen 3 upgrade), an 8 km/h walking speed, 20 kg payload, and 3 to 5 hours of battery life.

On paper, Tesla and Figure are converging on similar specs at similar target price points. The key difference is production reality. Tesla has deployed roughly 500 Optimus units inside its own Gigafactories but has not opened commercial sales. Figure AI has deployed roughly 200 units at BMW and is now building BotQ to scale production. Neither company is anywhere close to volume manufacturing, but Figure has committed to a public price and a dedicated factory, while Tesla has offered targets and timelines that have repeatedly shifted.

The labor substitution math

The entire economic argument for humanoid robots at scale comes down to one question: is it cheaper to operate a robot than to employ a human for the same task?

Let us run the numbers on the Figure 03.

A US manufacturing worker earning $25 per hour costs roughly $52,000 per year in wages alone. Add benefits, insurance, payroll taxes, and overhead, and the fully loaded cost is closer to $70,000 to $80,000 annually.

A Figure 03 at $20,000 is a one-time capital expenditure. If it operates 16 hours per day (two shifts, with charging during the third), that is roughly 5,800 productive hours per year. Even at conservative utilization rates, the robot pays for itself in under a year on wage savings alone.

But this calculation has asterisks.

First, the robot cannot do everything a human worker can do. The 42 DOF and Helix AI enable a wide range of manipulation tasks, but there will be tasks that require judgment, creativity, or physical capabilities outside the robot’s envelope. A realistic utilization rate for a Figure 03 in a mixed manufacturing environment is probably 40 to 60 percent of a human worker’s task range in year one, improving as the AI learns.

Second, maintenance is not free. Actuators wear out. Sensors need calibration. Software requires updates. Figure AI has not published maintenance cost estimates, but industry benchmarks for industrial robots suggest annual maintenance costs of 5 to 15 percent of the purchase price. At $20,000, that is $1,000 to $3,000 per year, still dramatically cheaper than a human salary.

Third, energy costs. Running a humanoid robot 16 hours per day with wireless charging consumes electricity. The exact power draw depends on task intensity, but a reasonable estimate based on similar systems puts energy costs at $500 to $1,500 per year.

Add it all up: a Figure 03 costs roughly $22,000 to $24,500 in its first year and $1,500 to $4,500 per year thereafter. Compare that to $70,000 to $80,000 per year for a human worker. Even at 50 percent utilization, the economics favor the robot after year one.

What compromises does $20,000 require?

No robot ships at this price without making hard tradeoffs. Here is what Figure appears to have cut, simplified, or deferred to hit $20,000.

Closed ecosystem

The Figure 03 has no public SDK, no API access, and no third-party development support. Everything runs through Figure AI’s proprietary Helix platform. If the Helix model cannot do something, you cannot write custom code to work around it.

Compare this to the Unitree G1’s full ROS2 compatibility, Python and C++ APIs, and support for NVIDIA Isaac Sim and MuJoCo. The G1 is a platform you can build on. The Figure 03 is a product you use as delivered.

For enterprise customers deploying robots in structured manufacturing environments, a closed ecosystem may be acceptable or even preferable, since it means one vendor is responsible for everything. For researchers, hobbyists, or companies that need custom behaviors, the closed system is a dealbreaker.

Unproven production volume

Figure AI has shipped approximately 200 robots. The $20,000 price assumes BotQ will reach 12,000 units per year. That is a 60x increase from current volume. The factory exists and is tooling up, but the gap between announced capacity and proven throughput is enormous.

If Figure delivers only 2,000 units in the first year of BotQ production, the per-unit cost will be significantly higher than $20,000. Either Figure absorbs the loss (burning investor capital), or early buyers pay more than the announced price through some pricing structure that separates the hardware cost from the AI subscription.

Privacy concerns for home deployment

Figure has positioned the 03 for eventual home deployment, not just factory use. But the robot has eight cameras and microphones that continuously map its environment, a closed AI system with no user-accessible privacy controls, and a stated intention to use deployment data for training future AI models.

Figure says it will implement data scrubbing similar to Google Street View face blurring. But “plans to implement” is not the same as “has implemented.” For factory deployments, where the environment is controlled and data policies are negotiated in enterprise contracts, this is manageable. For home deployments, the privacy picture is incomplete.

No track record at this price

The Figure 02 costs an estimated $150,000+. The Figure 03 is priced at $20,000. That is an 87% cost reduction in one generation. There is no precedent for this in humanoid robotics, and the closest analogy in consumer technology (early smartphones, electric vehicles) still required three to five product generations to achieve comparable cost reductions.

Figure is attempting to compress a multi-generational cost curve into a single product jump by simultaneously introducing in-house actuators, a purpose-built factory, simplified charging, and a software-defined capability model. Any one of those innovations could stumble, and the cost target depends on all of them working at once.

Pros and cons of the Figure 03

The competitive landscape in context

The Figure 03 does not exist in a vacuum. It enters a market where Chinese manufacturers are shipping thousands of units, Tesla is ramping internal deployments, and at least a dozen other companies are building humanoid robots at various stages of maturity.

What stands out is convergence. Figure and Tesla are arriving at remarkably similar specifications and price targets from very different starting points. Figure built a robotics company from scratch, raised $1.85 billion, and is designing everything in-house. Tesla took its existing AI infrastructure, computer vision expertise, and manufacturing capability and pivoted toward humanoid form factors.

The convergence suggests that $20,000 for a human-scale, 40+ DOF humanoid with decent AI is likely where the industry cost curve bottoms out for this generation of technology. If two companies with fundamentally different approaches arrive at the same price, it probably reflects a real floor rather than a marketing fantasy.

What this means for the market

Goldman Sachs projects the humanoid robot market will reach $38 billion by 2035. At $20,000 per unit and 12,000 units per year, Figure AI would generate $240 million in annual hardware revenue at full BotQ capacity. That is meaningful but modest relative to the $39 billion valuation.

The real revenue play is not hardware. It is the AI. Figure’s closed ecosystem positions the company to charge for AI capabilities on top of the hardware, whether through software subscriptions, per-task licensing, or deployment management contracts. The $20,000 price may function less as a profit center and more as a customer acquisition cost for a high-margin AI services business.

This mirrors what happened in the smartphone industry. Apple does not sell iPhones at cost, but the iPhone’s primary economic function is to create a captive audience for App Store revenue, iCloud subscriptions, and Apple Services. Similarly, the Figure 03 at $20,000 may be designed to create a captive audience for Helix AI services.

The three-tier market is forming

The humanoid robot market is splitting into three tiers.

Tier 1: Budget platforms ($10,000 to $20,000). Unitree G1 and similar Chinese-manufactured humanoids. Affordable, open-ecosystem, limited capability. Ideal for research, education, and light commercial pilots.

Tier 2: Industrial mid-range ($20,000 to $50,000). Figure 03 and potentially Tesla Optimus. Human-scale, industrial-capable, AI-driven. Designed for labor substitution in structured environments. Closed ecosystems with vendor-managed AI.

Tier 3: Enterprise premium ($100,000+). Boston Dynamics Atlas (electric), Agility Digit, and other purpose-built industrial humanoids. Maximum capability, enterprise support, proven reliability. For companies that need guaranteed uptime and are willing to pay for it.

Figure 03 at $20,000 creates Tier 2. Before its announcement, there was a gap between the $16,000 Unitree G1 and the $100,000+ industrial humanoids. Figure fills that gap with a product that offers industrial capability at near-consumer pricing.

The BMW precedent and what it proves

Figure AI’s deployment at BMW’s Spartanburg plant provides the closest thing to real-world validation for the Figure 03’s capabilities. The Figure 02 completed an 11-month trial at the plant, assisting in the production of over 30,000 BMW X3 vehicles and moving 90,000 components across 1,250 operating hours.

These numbers are modest by factory automation standards, but they establish three things.

First, humanoid robots can function reliably in an active automotive assembly plant. The environment is loud, cluttered, and unforgiving. Components are heavy, tolerances are tight, and downtime is expensive. The Figure 02 operated in this environment for nearly a year without a reported safety incident.

Second, the few-shot learning capability works in practice. BMW workers demonstrated new tasks to the Figure 02, and the robot learned them from roughly five demonstrations. This is significant because it means deploying a Figure robot does not require a team of robotics engineers to program every motion. A line worker can teach the robot a new task in minutes.

Third, the form factor is viable. A 173 cm humanoid can navigate factory aisles, reach conveyor belts, and operate alongside human workers without requiring facility modifications. This is the core advantage of the humanoid form factor over traditional industrial robots: it fits into spaces designed for humans.

The Figure 03 builds on all three of these validated capabilities while adding force-sensing hands, longer battery life, and the updated Helix AI model. If the BMW deployment proved the concept, the Figure 03 is designed to prove the economics.

What to watch for

The Figure 03 is one of the most consequential product announcements in the humanoid robot space. But announcements are not shipments. Here is what will determine whether the $20,000 promise becomes reality.

BotQ production ramp. The single most important signal is how many Figure 03 units ship in the first year of BotQ production. If the number is in the hundreds, the $20,000 price is aspirational. If it is in the thousands, the price is real.

First commercial delivery. Figure has not announced a delivery date for Figure 03 commercial orders. The gap between announcement and delivery will reveal how close BotQ is to production readiness.

Maintenance and TCO data. Total cost of ownership, including maintenance, energy, and AI service fees, will determine whether the labor substitution economics hold up over a 3 to 5 year deployment cycle.

Competitive response from Tesla. If Tesla confirms a commercial Optimus price in the $20,000 to $25,000 range and announces delivery timelines, the resulting competition will be the most important pricing battle in the humanoid robot industry.

Privacy policy for home deployment. If Figure intends to sell the 03 for home use, it needs to publish concrete, enforceable privacy policies covering camera data, audio recording, and AI training data usage. The current “planned” data scrubbing is insufficient for consumer trust.

The bottom line

The Figure 03 at $20,000 is either the product that makes humanoid robots a real industry or an overpromise that joins the long list of robotics hype cycles that failed to deliver.

The specs are impressive. 42 DOF, 16-DOF force-sensing hands, Helix VLA, 5-hour battery, wireless charging, ISO 10218 safety, and a human-scale form factor. At $20,000, these specs describe a machine that, if it works as advertised, fundamentally changes the economics of labor substitution in manufacturing, warehousing, and logistics.

The risks are equally real. The price depends on production volumes 60 times higher than current output. The ecosystem is closed. The privacy picture is incomplete. And an 87% cost reduction in a single generation has no precedent in this industry.

Figure AI has $1.85 billion in funding, a $39 billion valuation, proven BMW deployment results, and a factory designed to produce 12,000 humanoids per year. If any company can pull off this price-capability combination, Figure has the resources and the momentum to do it.

But resources and momentum are not delivery. The next twelve months will tell us whether the Figure 03 is the real thing or whether $20,000 was a number designed to capture headlines rather than reflect manufacturing reality. The humanoid robot industry has seen plenty of both.