Humanoid robot shipments in 2026 explained: 7,000 full-size units sold in 2025, Chinese makers ship thousands, US makers count hundreds

Humanoid unit counts are now used to justify valuations, factory investments and national policy targets, yet the three kinds of number in circulation (statistics-body sales, company shipments and factory capacity) are routinely mixed. An investor comparing Unitree's 5,500 with Figure's 350 and Tesla's one million is comparing a sales count that includes sub-140 cm robots, a cumulative delivery figure and a design capacity.

The practical consequence is that the market looks both bigger and more American than it is. On disclosed shipments the industry is Chinese, and on paid industrial deployments it is still small everywhere.

Unitree leads on volume with cheaper, shorter robots and a research and education customer base, which its prospectus makes explicit. UBTech leads on disclosed full-size volume with Walker-class robots sold to industrial and government-linked buyers, and its revenue per unit is falling as volume rises. AgiBot reports the largest production milestone but across product lines.

In the US, Figure has the clearest delivery disclosure and a named factory customer in BMW, 1X has a consumer-first plan with a 10,000-unit plant, Agility has orders but general availability only at the end of 2027, and Tesla has the largest stated capacity and no unit disclosure. Apptronik is the best-funded US challenger in 2026 by disclosed amount.

Unitree listed at an issue-price valuation of about RMB61 billion, 219 times 2025 earnings, against an industry average of 38.6 times cited in its own document. Figure's last disclosed valuation was USD 39 billion after its Series C. Neither price is explained by today's unit count: at 5,500 units and 350 units respectively, both are priced on future volume.

UBTech's falling revenue per unit is the number to put next to those valuations. If hardware prices fall faster than volume grows, revenue growth slows even as unit counts rise.

Thousands of full-size humanoids a year already create real demand for precision reducers, frameless motors, screws and dexterous hands, much of it met by Chinese suppliers. Our earlier analysis of humanoid joints found Tesla's stated Fremont capacity alone implies tens of millions of joint actuators a year, far beyond today's reducer output at Japanese incumbents.

The near-term bottleneck is less parts than integration labour and quality: Figure's move from one robot a day to one an hour shows how quickly assembly can scale once lines exist.

Tesla's third-quarter 2026 update in late October: first Optimus unit count or not. UBTech's 2026 annual results in March 2027: whether full-year units exceed twice the first half. Agility's listing via Churchill Capital Corp XI in the fourth quarter: the S-4 should disclose deployed fleet size. The IFR's World Robotics 2027 statistics in September 2027: whether the full-size count passes 15,000.

Thesis: humanoid shipments are a Chinese volume story and a US capacity story, and the IFR's 7,000 is the only comparable number. Confidence: medium. The filings are clear on volume, but the mix between paid industrial use and research purchases is poorly disclosed on both sides, and a single large US order book could change the picture within a year.

For Unitree: what share of 2026 humanoid shipments are above 140 cm and sold to industrial customers? For UBTech: how much of the full-size segment is hardware versus services, and what is the order backlog? For Tesla: how many Optimus units are working outside Tesla's own sites? For Figure: how many of the 350-plus delivered robots are on paid customer contracts? For 1X: how many NEO units have shipped against booked preorders?