Prices start below the cost of a small car, and banks forecast millions of sales. Yet the arithmetic of replacing a worker is harder than the sticker price suggests.
Whether humanoid robots take over human work will be decided less by engineers than by accountants. A machine replaces a person when it does the same task, reliably, for less money. So the useful question is not whether a robot can do a job, but what it costs when it does.
The price list
Published prices vary enormously. At the low end, the Chinese maker Unitree lists its G1 humanoid at $13,500 before tax and shipping. The American firm 1X sells its Neo home robot for $20,000, or rents it for $499 a month. Tesla has not put Optimus on sale; Elon Musk has named a target of $20,000 to $30,000 and has pointed to public sales around the end of 2027.
The robots actually working in factories and warehouses cost far more, and their makers do not publish prices. Agility’s Digit, used in logistics, has been estimated by outside analysts at around $250,000, a figure the company has not confirmed. Figure’s robots are sold by private quotation. Its chief executive has floated the idea of a home model at roughly $600 a month, which shows how the industry hopes to price robots: like a salary, not like a machine.
The cheap robots and the working robots are, for now, different machines. A $13,500 humanoid is mostly bought by researchers and developers. The models trusted on a production line cost many times as much.
The costs that do not appear on the label
The purchase price is only the beginning. A buyer must also pay for shipping and import duty, for the engineering work of fitting the robot into an existing process, for safety systems, for staff training, and for maintenance and spare parts. Then there is downtime: a robot that stops costs money for every hour it stands idle.
There is also supervision. Today’s humanoids need people nearby to reset them, clear their errors and, in many cases, take over by remote control. A robot that requires one human minder for every few machines has not eliminated labour. It has changed its kind.
Price guides for the industry therefore advise buyers to ignore the sticker and calculate the cost, over about three years, of getting a given quantity of work done. On that measure a robot pays only if it solves a real constraint better than the alternatives, which include hiring people, changing the process, or installing simpler automation.
The promise of two dollars an hour
The industry’s optimism rests on a projection. A study by the consultancy Roland Berger suggests that at large scale the operating cost of a humanoid could fall to about $2 an hour. Set against wages in the United States, Europe or Japan, that would be decisive.
Two cautions apply. The figure is a forecast for mass production, not a description of today. And it covers running costs only, excluding the purchase or lease of the robot. Nobody deploying humanoids in 2026 is paying anything like $2 an hour for useful work.
The quiet change in the business model
One development may matter more than any price. Increasingly, customers do not buy the robot. They buy the work.
Agility supplies Digit to the logistics company GXO as a service, charging for the robots’ use. The Dutch firm Monumental goes further. Its bricklaying robots are never sold; the company acts as a subcontractor and is paid for finished walls.
This changes the nature of the decision. A manager who buys a robot and lays off a worker has made a visible choice. A manager who hires a subcontractor has simply chosen a supplier. If that supplier happens to use machines, no job was formally replaced; it was never created. Automation of this kind can spread without anyone deciding to dismiss anyone.
The forecasts
Investment banks expect explosive growth. In September 2026 Goldman Sachs raised its forecast to about 75,000 humanoids shipped in 2026, 890,000 in 2030 and 6.5 million in 2035, a market it values at roughly $138 billion. Its previous estimate for 2035 had been 1.4 million units and $38 billion. The bank cites falling hardware costs as one reason for the revision.
Against these numbers stands the present. Independent trackers that count confirmed deployments find hundreds to a few thousand humanoids at work worldwide. The largest fleet confirmed by a manufacturer and its customer in a Western factory is forty robots.
The forecasts may yet be right. Costs of motors, batteries and chips are falling, and software is improving quickly. But a forecast is an argument for investment, written by institutions with clients to advise. The distance between 75,000 shipments and a few thousand verified workers is a measure of how much is still expectation.
What the jobs studies say
The most widely quoted study of employment, the World Economic Forum’s Future of Jobs Report 2025, projects that by 2030 about 170 million jobs will be created worldwide and 92 million displaced, a net gain of 78 million. Those figures are often cited to show that robots create more work than they destroy.
They show something less tidy. The numbers cover every force acting on employment, including ageing populations, the energy transition and economic pressure, not robots alone. And the same report finds that of every hundred workers, fifty-nine will need retraining by 2030, and that eleven of those are unlikely to receive it. Four in ten employers surveyed expect to reduce staff as AI takes over tasks.
A net gain across the world is little comfort to the particular worker whose task disappears. The jobs created and the jobs lost are rarely in the same place, or open to the same people.
The honest sum
For now the arithmetic favours humans in most settings, and overwhelmingly so where wages are low. A robot wins where labour is scarce, expensive, or exposed to danger, and where the task is simple and endlessly repeated. That describes some factory and warehouse work in rich, ageing countries. It does not yet describe most of the world’s work.



