Humanoid robots stopped being a demo business in 2025. Chinese makers shipped them by the thousands, while the best-known US names shipped them by the hundred. In the first half of 2026, global shipments reached about 19,100 units, more than triple the 5,100 a year earlier, and Chinese companies made over 97% of them. This is where the market stands, who leads, and the price per working hour that will decide how fast it grows.
Who Is Actually Shipping
Unitree shipped more than 5,500 humanoids in 2025, about a third of the world total, by its own count. AgiBot shipped 5,168 by Omdia’s count. Figure AI, Agility Robotics and Tesla each shipped somewhere around 150, and Tesla has never published a production number for Optimus. In the first half of 2026 AgiBot moved into first place with roughly 8,400 to 9,700 units depending on the tracker, and Unitree shipped over 7,000.

| Company | Country | Status in late 2026 |
|---|---|---|
| Unitree | China | 5,500+ humanoids in 2025; went public on Shanghai’s STAR Market in August 2026; G1 starts near $12,000 to $16,000 |
| AgiBot | China | Largest shipper in 1H 2026; preparing a Hong Kong IPO |
| Figure AI | US | More than 350 Figure 03 units built by April 2026; BMW Spartanburg deployment |
| Agility Robotics | US | Digit in warehouse use; multi-year deal covering 1,000 robots; service contracts from about $8,500 a month |
| Tesla | US | Optimus lines being installed at Fremont; no production count published |
| Boston Dynamics | US (Hyundai) | Electric Atlas, strongest on mobility and dexterity; Hyundai plants are the first customer |
Where the Robots Go Today
Unitree’s IPO filing gives the clearest picture. In the first nine months of 2025, 73.6% of its humanoid revenue came from research and education, 17.4% from commercial and consumer uses, and only about 9% from industry. Its average humanoid price fell from about $39,000 in 2024 to about $25,000 in 2025. Unitree made a profit, with gross margin near 60%, but its customers are mostly labs, universities and showrooms. Factory work is still the small slice.
In the US the model is different. Figure, Agility and Apptronik sell fewer, more capable machines into specific jobs: tote moving, parts kitting, case handling. Figure’s robot finished an eleven-month posting at BMW’s Spartanburg plant. Agility’s Digit moves bins in fulfillment centers. Tesla’s Elon Musk has said Optimus will be the hardest product Tesla has ever scaled, because there is no existing supply chain for it.
The Price That Unlocks Scale
The real question for buyers is simple. Can a humanoid deliver a productive hour for less than a person does? US warehouse workers earn about $18 to $23 an hour in base pay, and close to $30 with benefits. Across all US jobs, fully loaded employer cost is around $45 an hour.
A robot that works two shifts can deliver 3,000 to 5,000 productive hours a year. Over a five-year life, at a target of $15 to $20 per hour all-in, a buyer can justify paying roughly $75,000 to $200,000 for the machine, depending on service costs and uptime. Chinese list prices are already far below that. US makers are still above it once integration and support are counted. So cost is less of a barrier than reliability. A $16,000 robot that fails half its tasks is worth nothing on a production line.
- Logistics first. Structured tasks, controlled floors and multiple shifts make the math work soonest.
- Manufacturing next. Needs better hands and faster task switching.
- Healthcare, elder care and homes last. Huge demand, but much higher bars for safety and judgment.
Why China Leads
China has the supply chain for motors, reducers, sensors and batteries, and it can build at low cost. Morgan Stanley has pointed out that Tesla buys Optimus harmonic reducers from a Chinese supplier, and one analysis estimated that taking Chinese parts out could roughly triple Optimus’s cost. Beijing treats humanoids as a strategic industry, and local investors are paying for it: Unitree’s retail IPO tranche was oversubscribed more than 8,000 times.
The US edge is in robot intelligence. Foundation models for robots, like those from Skild AI and Physical Intelligence, and Nvidia’s robotics stack, are mostly American. Read more in Skild AI’s funding round and Project Prometheus. Whoever pairs cheap Chinese-style hardware with the best robot brains will set the pace.
Forecasts and the Hype Check
Forecasts for 2026 range from about 50,000 to 60,000 units worldwide. Long-range numbers are much larger. Goldman Sachs projects about $38 billion in annual sales by 2035, and Morgan Stanley sees a $5 trillion industry by 2050. Skeptics such as Rodney Brooks point out that human-level hand dexterity is still far off, and that robots struggle in cluttered, changing spaces.
The turning point will be a dull one: a large company showing in its results that humanoids cut labor cost in a warehouse by a real margin with uptime close to human crews. Until then, most humanoids will keep going to labs and pilots. The supply side is ready. Demand from real factories is what is still being tested. For more on robotics and automation trends see technologies.org.