Industrial humanoid robots vs consumer ones: why the market splits
Every humanoid with a paying customer in 2026 has an industrial one, and the companies advertising home robots publish no consumer price; the entire humanoid category shipped 13,000 to 20,000 units in 2025 against roughly 550,000 conventional industrial robots installed in 2024 alone.
Figure's front page says "The future of home help is here" and calls Figure 03 "a general purpose humanoid robot for every day." Figure's own news index says "F.03 Arrives at BMW," dated June 30, 2026. Both statements are accurate. Only one of them has a customer attached to it.
That gap is the whole structure of the humanoid industry right now. The industrial humanoid robot is marketed at your kitchen and delivered to somebody's plant, and the reasons are boring, physical, and easy to check.
Where the robots actually went
Prices and dates are from each company's own pages.
| Maker | What its site promises | Where the robot actually is | Price to a private buyer |
|---|---|---|---|
| Figure | "The future of home help is here" | BMW (F.02 Nov 19, 2025; F.03 Jun 30, 2026) | None published |
| Apptronik | "manufacturing, warehousing, and logistics" | Enterprise pilots | None published |
| Agility Robotics | "connect islands of automation in your facility" | Amazon, GXO, Schaeffler; Mercado Libre (Dec 10, 2025), Toyota Canada (Feb 19, 2026) | None published |
| Boston Dynamics | "the future of industrial operations" | Hyundai, field testing | None published |
| UBTech | Walker S2, mass production and delivery work | Chinese factory floors | None published |
| Tesla | Optimus, everything, eventually | Nowhere confirmed | None published |
| 1X | NEO, "perfect for people and the home" | Preorder list | $20,000 or $499/mo |
One row on that table has a number in it. Everything else routes to a sales team.
Two robots, engineered for two different rooms
The spec sheets settle it. One category of machine, built two incompatible ways depending on who is buying.
UBTech's Walker S2 page leads with "the world's first humanoid robot capable of autonomous battery swapping." Three minute swap, dual battery dynamic balancing, 15 kg payload, a working envelope from the floor to 1.8 meters. Every one of those choices exists to keep a machine moving across shift changes in a building that runs around the clock.
1X's NEO page leads with a different set of numbers. It weighs 66 lbs. The exterior is a "custom 3D lattice polymer" under a machine washable nylon suit. The joints have "no pinchpoints" and are "covered from outside access." It runs at 22 dB, which 1X describes as quieter than a modern refrigerator, and it publishes a Head Injury Criterion rating below 250.
A factory robot is optimized so it never stops. A house robot is optimized so that if it falls on your dog, the dog is fine. Same silhouette, different machines, and nobody has built one that does both jobs well.
Agility puts the same logic in a leg. Digit's knees bend backward, and the company's stated reason is not that any animal's do: "We don't design for biomimicry. We design for the job."
Agility explains Digit's backward-bending knees as a job requirement, reaching the floor and lifting a heavy tote overhead, rather than a copy of any animal (@agilityrobotics, 2026-07-30)
Why the factory pays first
Four reasons, none of them about how clever the AI is.
The room holds still. A warehouse aisle is measured, lit, repeated, and swept. A living room has a toddler in it. Interact Analysis, in its May 19, 2026 revenue outlook, calls household use "a longer-term opportunity, constrained by safety and environmental complexity." That is the polite version.
The task repeats. Agility's Digit carries 35 lbs, runs about four hours on a charge, and has moved "over 100,000 Totes in Commercial Deployment." One motion, a hundred thousand times, in a building designed around totes. Nothing in a house repeats at that rate.
The hours are there. Two shifts, five days a week, is about 80 hours. A house robot doing an hour of chores a day is seven. Same purchase price, roughly eleven times the work out of it.
Somebody can sign. A plant manager has a budget line, a maintenance department, and a legal team that already accepts machinery risk. A household has none of those and is asked to hand $20,000 to a startup, a category with a documented failure rate.
Morgan Stanley's April 2025 research, the source of that $5 trillion headline, puts roughly 930 million of the more than one billion humanoids it projects for 2050 on "repetitive, simple, and structured work," and about 80 million in homes. Analyst Adam Jonas, same page: "We are not going to see a robot in every home overnight."
"General purpose" describes a research program
Figure keeps posting home demonstrations of its Helix model, each dated on its news index: laundry folding August 12, 2025, the dishwasher September 3, 2025, a living room tidy March 9, 2026, a bedroom tidy May 8, 2026. The revenue events over the same stretch were F.02 contributing to the production of 30,000 cars at BMW (November 19, 2025) and F.03 arriving at BMW (June 30, 2026).
Apptronik states the sequence outright. Apollo's markets are "manufacturing, warehousing, and logistics," and the same foundation "can extend into more complex environments, from hospitals and retail spaces to eldercare and ultimately, the home." Ultimately is doing the work in that sentence.
Boston Dynamics builds Atlas for "part sequencing, machine tending, order building, and more," and has it "taking its first steps into a customer facility with Hyundai for field testing." Nobody is shipping a general purpose humanoid to a general purpose customer.
The exception is 1X NEO, and it is an exception with a person inside it. NEO arrives with "basic autonomy," and for anything harder, "an Expert from 1X can remotely supervise its actions at scheduled times." A human in a headset, by appointment. The full breakdown is in how autonomous humanoids actually are.
The humanoid service robot already had its run
"Humanoid service robot" is a phrase that now mostly describes a market that peaked and closed. SoftBank's Pepper was the greeter in the bank lobby and the phone shop, and it reached roughly 27,000 units before SoftBank paused production in June 2021. Aldebaran, the company that built Pepper and NAO, went into receivership in 2025.
What survived in that slot is rental. Engineered Arts' Ameca works conferences and museums, and rents about as much as it sells. PAL Robotics in Barcelona sells its service platforms through an email address. Real businesses, neither of them a category with a checkout. The full roster is in the company map.
The military humanoid is the one that does not exist
"Military humanoid robot" pulls steady search traffic, and I can find no armed force that has fielded a walking humanoid in an operational role. Defense robotics money goes to things that fly, roll, or swim.
The lineage runs the other way. The best known biped on earth traces back to the DARPA Robotics Challenge, a program for "human-supervised robotic technology for disaster-response operations," now complete. Boston Dynamics still publishes a pledge signed with other mobility robot makers: "We pledge that we will not weaponize our advanced-mobility general-purpose robots or the software we develop that enables advanced robotics and we will not support others to do so." The page carries no date I can verify, and a voluntary pledge is unenforceable. It remains the industry's stated position, with no product line contradicting it.
Even the industrial humanoid is a rounding error
The funding announcements bury the scale. The humanoid is a marginal machine inside industrial robotics, never mind inside houses.
The International Federation of Robotics, in an April 8, 2026 release, counts roughly 550,000 industrial robots installed worldwide in 2024, 295,000 of them in China. The entire humanoid category shipped between 13,000 and 20,000 units across all of 2025, and Interact Analysis reckons only about 10% of everything produced through 2025 had been deployed in real work. The shipment ledger and the 2035 forecasts that disagree by 14x get their own page.
Arms bolted to floors still do the automation. Humanoids are in pilots at a handful of named plants, and homes sit behind that.
What the split means if you are the one shopping
The companies with working robots sell to plants and will not quote you a price. The one company selling to households takes a $200 deposit against $20,000 or $499 a month, states "US Deliveries start 2026," and ships a remote human as part of the product. Tesla has no order page at all.
Watch the enterprise contracts rather than the demo reels. Every dated customer agreement above names a factory, a warehouse, or a car plant, and that holds until a maker publishes a household price with a delivery date beside it.
If one does, or if I have missed a fielded deployment that breaks the pattern, send it over. Dated sources only, same as everything here.
One email when something actually changes: a delivery verified, a price cut, a robot discontinued. No weekly filler, unsubscribe whenever.