A humanoid robot that walks on two legs, picks up parts, and assembles products on a factory line used to cost more than a house. In 2023, Unitree Robotics sold its humanoid units at an average price of around $85,000. By mid-2025, that average selling price had dropped to roughly $25,000, and the company’s entry-level G1 model lists at around $13,500. The humanoid robot economics that once made these machines a lab curiosity are shifting fast enough to make factory owners pay attention.
This is not a story about a distant future. BMW has already completed a pilot deployment of humanoid robots at its Spartanburg plant in the United States and is now expanding to its Leipzig plant in Germany. Agility Robotics’ Digit has moved over 100,000 totes at a GXO logistics facility. Unitree’s G1 is assembling robot parts inside the company’s own factory. The question is no longer whether humanoid robots will reach the factory floor. It is how fast the economics will push them there.
Why Prices Are Crashing
Two forces are driving the collapse in humanoid robot prices: component commoditization and vertical integration.
The single most expensive part of a humanoid robot is its actuatorsMotorized components that move a robot's joints, combining an electric motor, gearbox, and encoder. They account for 40-60% of a humanoid robot's total cost., the motors and gearboxes that move each joint. McKinsey estimates that actuation accounts for 40% to 60% of a humanoid’s total bill of materials. Unitree manufactures its own actuators in-house, which is a major reason its gross margin actually rose to 59.8% even as selling prices dropped by more than 70%. Most hardware companies see margins compress when prices fall. Unitree’s margins improved, which points to a genuine cost advantage in its supply chain.
The second force is China’s electric vehicle supply chain. Companies like BYD, Geely, and NIO already mass-produce electric motors, battery packs, sensors, and power electronics at enormous scale. These are the same components a humanoid robot needs. When an EV factory can supply servo motors by the millions, the component cost for a robot builder drops sharply. This is not coincidence. Beijing has identified the overlap explicitly, calling it “industrial migration” as EV and tech companies enter the humanoid robotics sector.
Goldman Sachs revised its humanoid robot market forecast upward by a factor of more than six, from $6 billion to $38 billion by 2035, citing a 40% reduction in the cost of materials and faster-than-expected AI progress as the primary drivers.
Who Is Actually Buying
Here is the uncomfortable truth behind the optimistic headlines: most humanoid robots are not going to factories yet.
Unitree’s IPO filing reveals the real customer mix. In the first nine months of 2025, 73.6% of its humanoid revenue came from research and education. Another 17.4% went to commercial uses like demonstrations and displays. Only 9.01% came from industrial applications. The company shipped 3,551 humanoids in that period with a 95.95% sell-through rate, which shows real demand, but the demand is mostly from universities and labs, not production lines.
This pattern holds across the industry. Agility Robotics’ Digit is in active commercial deployment at GXO and Amazon, but the total number of units is in the low hundreds, not thousands. Figure AI completed a pilot at BMW Spartanburg where its Figure 02 robot supported the production of more than 30,000 BMW X3s over ten months, handling sheet metal positioning for welding. That is a real result, but a single-task pilot at one plant.
The gap between “units shipped” and “units doing productive factory work” is the central tension in humanoid robot economics right now.
The Math That Matters
The economic case for humanoid robots gets compelling when you compare costs over time. A manufacturing worker in the United States costs roughly $80,000 per year in base wages, with benefits adding another 50%, for a total loaded cost of around $120,000 annually. A humanoid robot priced at $50,000 with a five-year lifespan and $5,000 per year in maintenance works out to $15,000 per year and can operate 20 hours a day without overtime, sick days, or benefits.
But this simple comparison misses the real picture. Traditional industrial robots, the kind bolted to factory floors, already cost $50,000 to $200,000 and have been doing welding, painting, and palletizing for decades. Their total system cost, including integration, safety cages, and tooling, runs $150,000 to $500,000. They are proven and reliable, but rigid. They do one task in one place.
The humanoid’s economic pitch is flexibility. A humanoid can, in theory, move between tasks, navigate human-designed spaces, and use existing tools without retrofitting the entire production line. That flexibility is why BMW and GXO are testing them. The question is whether the flexibility premium justifies the higher complexity and lower reliability compared to a fixed robot arm that has been perfected over 40 years.
Humanoid Robot Economics and the Subsidy Factor
No analysis of humanoid robot economics is complete without understanding the scale of government support behind the Chinese robotics industry.
CSIS reports that the Chinese government allocated over $20 billion in subsidies to its robotics industry in late 2024 and early 2025. A new NDRC guidance fund aims to direct $137 billion into AI and robotics startups over the next 20 years. Provinces and cities are running what the Jamestown Foundation describes as a “subsidy race,” with Beijing and Guangzhou each launching dedicated 10 billion RMB robotics funds, Guangdong offering up to 100 million RMB for approved robotics projects, and Jiangsu providing up to 30 million RMB for manufacturing innovation centers.
In 2024, China installed 295,000 new industrial robots, more than every other country in the world combined. Chinese firms shipped a majority of the world’s humanoid robot units in 2025. The 15th Five-Year Plan, launched in 2026, includes robotics among its priority sectors for accelerated development.
This level of state support means that the sticker price of a Chinese humanoid robot does not necessarily reflect its true production cost. When a government subsidizes R&D, manufacturing capacity, and purchases through state-enterprise procurement, the resulting prices can undercut any competitor operating without that support.
What This Means for Workers
The labor implications are real but uneven. Research by MIT economists Daron Acemoglu and Pascual Restrepo found that adding one robot per 1,000 workers in the U.S. reduces wages by 0.42% and cuts the employment-to-population ratio by 0.2 percentage points. In geographic areas where robots are concentrated, the effect is sharper: one additional robot reduces local employment by about six workers.
But context matters. The National Association of Manufacturers reports over 600,000 unfilled manufacturing jobs in the U.S. as of 2025, with projections of 2.1 million unfilled positions by 2030. The industry is not choosing between human workers and robots in a vacuum. It is choosing between robots and unfilled positions that no one is applying for.
The World Economic Forum’s 2025 Future of Jobs Report projects that technology will create 170 million new jobs globally while displacing 92 million, for a net gain of 78 million jobs by 2030. The catch is that the jobs created and the jobs destroyed are not in the same places, industries, or skill levels. The machinists, assemblers, and material handlers identified by the MIT study as most vulnerable to robot displacement are not the same people who will fill new roles in robot maintenance, AI training, or systems integration.
The Production Ramp Problem
Tesla has announced plans for Optimus that dwarf everyone else’s numbers. The company targets a $20,000 price point once mass production begins, with a Gen 3 reveal in early 2026 and production starting later that year. The roadmap calls for a 1 million unit per year line at its Fremont factory, scaling to 4 million by the end of 2027, and eventually 10 million per year at Giga Texas.
These numbers should be met with the same skepticism applied to any Tesla production timeline. The company has a history of ambitious targets followed by slower actual ramps. But even at a fraction of these numbers, the scale would be transformative. Figure AI’s BotQ manufacturing facility targets 12,000 units per year initially, scaling to 100,000 over four years. Agility Robotics’ RoboFab in Salem, Oregon has capacity for 10,000 Digits per year.
Unitree is the only company that has demonstrated real manufacturing scale so far. With over 5,500 G1 units shipped in 2025 and a $248 million revenue base, it has a working hardware business. Its R1, launched at $5,900, pushes the price floor even lower, though it is designed more as a consumer and research platform than a factory worker.
What Comes Next
The honest assessment is this: humanoid robots are real products with real revenue, but they are not yet proven factory tools at scale. The economics are moving in one direction, prices are falling, capabilities are improving, and the money flowing into the sector is enormous. Goldman Sachs projects more than 250,000 humanoid robot shipments by 2030, almost all for industrial use.
But the gap between a robot that can be demonstrated doing a task and a robot that can reliably perform that task 20 hours a day for five years is vast. Unitree’s CEO has framed this as the “80/80” problem: a humanoid needs to complete 80% of tasks in 80% of unfamiliar environments before it reaches a commercial tipping point. No one is there yet.
What is clear is that the price barrier has fallen. At $13,500 for a Unitree G1, a factory can buy 15 humanoid robots for the price of a single traditional industrial robot system. Even if each one handles a narrow set of tasks, the economics of deploying cheap, flexible labor are starting to pencil out. The automation wave is not coming from a single expensive machine replacing a worker. It is coming from a swarm of inexpensive ones, each doing a fraction of what a human can do, but doing it around the clock.
The factory floor of 2030 will not be the robot-free zone of 2020 or the fully automated “dark factory” of science fiction. It will be something messier and more interesting: humans and bipedal machines working side by side, with the ratio shifting a little more each year as the software catches up to the hardware that is already, improbably, cheap enough to deploy.
A humanoid robot that walks on two legs, picks up parts, and assembles products on a factory line used to cost more than a house. In 2023, Unitree Robotics sold its humanoid units at an average selling price of approximately $85,000 (¥593,400). By mid-2025, that ASP had dropped to roughly $25,000 (¥167,600), a 70% reduction in under two years, while the company’s companywide gross margin rose to 59.8%. The entry-level G1 model lists at approximately $13,500. Understanding the humanoid robot economics behind this price collapse reveals both why the market is accelerating and why skepticism remains warranted.
This is not a story about a distant future. BMW completed a pilot deployment of Figure AI’s Figure 02 at its Spartanburg plant, where the robot supported production of over 30,000 BMW X3 units over ten months, handling sheet metal removal and positioning in the welding process. Agility Robotics’ Digit has moved over 100,000 totes at a GXO logistics facility in Flowery Branch, Georgia. Unitree’s G1 is assembling robot parts in the company’s own factory using its UnifoLM-X1-0 embodied AIArtificial intelligence designed for physical agents that must perceive and act in the real world, handling movement, object manipulation, and navigation in unstructured environments. model. The question is no longer whether humanoid robots will reach the factory floor. It is how fast the underlying economics will push them there.
The Bill of Materials Breakdown
The key to understanding humanoid robot pricing is the bill of materials (BOM), and within it, actuation dominates. McKinsey estimates that actuatorsMotorized components that move a robot's joints, combining an electric motor, gearbox, and encoder. They account for 40-60% of a humanoid robot's total cost., the servo motors, harmonic drives, and gearboxes that power each joint, account for 40% to 60% of a humanoid robot’s total BOM cost. A humanoid with 23 to 43 degrees of freedom (the range for Unitree’s G1, depending on configuration) requires a corresponding number of precision actuators, each combining a brushless DC motor, a strain-wave or planetary gear reducer, an encoder, and a motor driver.
Unitree’s cost advantage stems from vertical integration on this exact component. The company’s IPO prospectus cites “self-developed and self-produced core components” as central to its margin story. By manufacturing actuators in-house rather than sourcing from suppliers like Harmonic Drive Systems (Japan) or Leaderdrive (China), Unitree avoids the typical 40-60% margin stack that component suppliers charge. The result: gross margins improved to 59.8% even as ASPs fell by more than 70%. In most hardware businesses, this is structurally anomalous. It only makes sense if the company is reducing internal cost faster than it is reducing price.
The broader cost tailwind comes from China’s electric vehicle supply chain. EV manufacturers produce brushless motors, lithium battery packs, LiDAR modules, depth cameras, IMUs, and power electronics at volumes that dwarf anything the robotics industry alone could generate. Humanoid robot builders are piggybacking on this existing scale. Beijing has identified this supply chain overlap explicitly, describing “industrial migration” as EV and tech companies enter the humanoid sector, bringing scale, engineering capacity, and cost advantages with them.
Goldman Sachs quantified this effect in its revised forecast, citing a 40% reduction in the cost of materials as a primary driver for increasing its humanoid robot TAM estimate more than sixfold, from $6 billion to $38 billion by 2035, with projected shipments of 1.4 million units.
Revenue Mix: The 9% Problem
The optimistic market projections collide with a stubborn reality in the revenue data. Unitree’s IPO exchange response materials show that in the first nine months of 2025, 73.6% of humanoid revenue came from research and education. Another 17.4% went to commercial consumption (demonstrations, display environments, events). Only 9.01% came from industrial applications.
The sell-through data is healthy: 3,701 humanoids produced, 3,551 sold, yielding a 95.95% sell-through rate. This indicates genuine demand, not channel-stuffing. But the customer composition tells a different story than the one required to justify a $38 billion industrial automation market. Universities and research labs are real customers, but they buy in ones and twos for experimentation, not in hundreds for production deployment.
Across the industry, commercial deployments remain in pilot or limited-scale phases. Agility Robotics has Digit units at Amazon (tote recycling at the Sumner facility) and GXO, but total deployed fleet size is measured in the low hundreds. Figure AI’s BotQ manufacturing facility targets 12,000 units per year initially, with a goal of 100,000 over four years. These are real manufacturing commitments, but they are capacity targets, not deployed-and-working numbers.
Humanoid Robot Economics vs. Traditional Automation
The cost comparison that matters is not humanoid vs. human. It is humanoid vs. the existing industrial automation stack.
A traditional 6-axis industrial robot arm (FANUC, ABB, KUKA) costs $50,000 to $200,000 for the unit alone. Add integration, safety infrastructure (cages, light curtains, area scanners), end-of-arm tooling, and programming, and the total system cost runs $150,000 to $500,000. These systems are mature, with mean time between failure measured in tens of thousands of hours, and they perform single tasks with sub-millimeter repeatability.
A humanoid at $13,500 to $50,000 offers none of that precision or reliability. What it offers is versatility: bipedal locomotion through human-designed spaces, manipulation with dexterous hands (the G1 EDU has up to 43 DOF), and the ability to switch between tasks without physical reconfiguration. The economic pitch is not “cheaper per task” but “deployable across many tasks without retooling.”
For the labor comparison: a U.S. manufacturing worker at $80,000 base salary with 50% benefits loading costs approximately $120,000 per year. A $50,000 humanoid over a five-year lifespan with $5,000 annual maintenance works out to $15,000 per year and can theoretically operate 20 hours per day. The per-hour cost advantage is overwhelming on paper. The question is whether the robot’s effective utilization rate and task success rate justify the comparison. A robot that operates 20 hours per day but completes tasks at 60% the rate of a human worker, with a 5% error rate requiring human intervention, has very different economics than the simple cost-per-hour calculation suggests.
The State Capital Factor
No analysis of humanoid robot economics is complete without quantifying the state support behind the Chinese robotics industry, which constitutes the vast majority of current humanoid production.
CSIS reports that Beijing allocated over $20 billion in subsidies to its robotics industry in late 2024 and early 2025 through grants, loans, tax credits, and state-backed venture capital. The NDRC launched a government guidance fund in March 2025 targeting $137 billion in AI and robotics investment over 20 years. At the provincial level, Beijing and Guangzhou each launched $1.4 billion (10 billion RMB) robotics funds. Shenzhen, Shanghai, and numerous other cities followed with dedicated funds.
The Jamestown Foundation’s analysis of 30 national, provincial, and municipal policy documents found that provinces and cities are engaged in a “subsidy race,” each competing to produce the next national robotics champion. The MIIT established a dedicated Humanoid Robot and Embodied Intelligence Standardization Technical Committee in December 2025, and by March 2026 had released the first national standard system covering the humanoid robot industry’s entire lifecycle.
In 2024, China installed 295,000 new industrial robots, more than every other country in the world combined. Chinese firms shipped a majority of the world’s humanoid robot units in 2025.
This matters for pricing analysis because it means the market price of a Chinese humanoid robot may be substantially below its true fully-loaded production cost. When government subsidizes R&D (the $8.2 billion National AI Industry Investment Fund), manufacturing capacity (provincial grants for innovation centers), and purchases (state-enterprise procurement), the resulting retail price is a policy signal, not purely a market signal.
Labor Market Displacement: What the Data Shows
The most rigorous empirical work on robot-labor displacement comes from MIT economists Daron Acemoglu and Pascual Restrepo. Their study of U.S. labor markets found that for every robot added per 1,000 workers, wages decline by 0.42% and the employment-to-population ratio drops by 0.2 percentage points. In commuting zones with direct robot exposure, one additional robot reduces local employment by approximately six workers.
The displacement effect is concentrated in routine manual occupations: machinists, assemblers, material handlers, welders. The automotive industry, which employs 38% of existing industrial robots at a density of up to 7.5 robots per 1,000 workers, shows the strongest effects. The study found negative wage impacts across all education levels, though workers without college degrees were hit far harder.
However, these findings predate the humanoid robot era and are based on traditional fixed industrial robots. Humanoid robots, if they reach the versatility their proponents claim, could affect a broader range of occupations because they are not confined to a fixed position and can interact with human-designed environments. Conversely, their current reliability limitations mean near-term displacement may be more limited than the hardware price points suggest.
The macro picture is complicated by labor shortages. The National Association of Manufacturers reports over 600,000 unfilled manufacturing positions in the U.S. as of 2025, with projections of 2.1 million by 2030. In this context, robots may be filling a vacuum rather than displacing existing workers, at least initially.
Production Scaling: Targets vs. Reality
Tesla’s Optimus production roadmap is the most aggressive in the industry. At the 2025 shareholder meeting, Tesla announced a $20,000 price target for mass production, with Gen 3 production beginning in 2026, a 1 million unit/year line at Fremont, 4 million units/year by end of 2027, and an eventual 10 million unit/year facility at Giga Texas. Tesla’s argument is that Optimus shares 80% of its technology stack with its vehicles: cameras, AI chips, actuators, battery systems, and neural network training infrastructure.
These targets should be weighted against Tesla’s historical production timeline accuracy. The company’s actual manufacturing capability is formidable, but its public timelines consistently compress what turns out to be a longer development cycle. Even at 10% of the stated targets, however, the volume would exceed the entire current humanoid industry.
Unitree is the only company with demonstrated manufacturing scale. Over 5,500 G1 units shipped in 2025, $248 million in total company revenue, humanoids representing 51.5% of core revenue. Its R&D spending as a share of revenue fell from 31.39% in 2023 to 7.73% in the first nine months of 2025, while absolute R&D spending still rose from $6.9 million to $12.5 million. This is the profile of a company transitioning from development to manufacturing scale.
The R1, at $5,900 with 20-26 DOF, pushes the price floor further. At 25 kg and 123 cm tall with approximately one hour of battery life, it is not an industrial workhorse. But it demonstrates that the humanoid form factor can be manufactured at consumer electronics price points.
The Convergence Ahead
The honest technical assessment is that humanoid robots are real products generating real revenue, with demonstrated cost curves moving in the right direction, but they are not yet proven industrial tools at scale. The gap between “can perform a task in a demo” and “can reliably perform that task 20 hours per day for five years with less than 1% error rate” remains large.
Unitree’s CEO Wang Xingxing has framed this as the “80/80” problem: a humanoid needs to complete 80% of tasks in 80% of unfamiliar environments to reach a commercial tipping point. No current platform meets this bar. The software, specifically the embodied AI models that handle perception, planning, and manipulation in unstructured environments, lags behind the hardware by a significant margin.
Goldman Sachs projects more than 250,000 humanoid shipments by 2030, almost all for industrial use, with consumer robots ramping up afterward. That timeline depends on the AI software closing the gap with the hardware, which is progressing faster than expected but still far from the general-purpose capability the economics ultimately require.
What is clear today: at $13,500 per unit, a factory can deploy 15 humanoid robots for the cost of a single integrated industrial robot system. Even if each humanoid handles only a narrow set of tasks at 30% human efficiency, the aggregate economics of cheap, flexible, around-the-clock labor are beginning to pencil out. The automation wave arriving on the factory floor is not the science fiction vision of a single powerful machine replacing a worker. It is a swarm of inexpensive, limited-capability platforms, each handling a fraction of human work, improving incrementally as the software matures, and getting cheaper every quarter.
