Who Has Actually Built How Many Humanoids
Updated: November 4, 2026 (7531-11-14 in the Bulgarian calendar)
The twelve companies in this section are usually ranked by valuation, which measures what investors expect. This page ranks them by what exists. The two orders are almost the reverse of each other, and that inversion is the most useful thing to understand about the field right now.
Humanoids shipped, first half of 2026
The most current ranking available, and the only one where every vendor is counted by the same firm on the same basis — Smart Analytics Global's Robotics 360, August 2026. All figures are estimates.
KUAVO series only
These five are 86% of the world's humanoid shipments. Total market 19,100 units, up 272% on the 5,100 of H1 2025. Counterpoint Research put the same half-year above 22,000, and IDC, reporting on 30 September, near 25,000 (up 432% year on year, Chinese vendors 95%) — three firms, a spread of a third, and AgiBot first in all three. That spread is the honest state of the data.
Chinese vendors accounted for 97% of every humanoid robot shipped in the world in those six months. Not one American or European maker appears in the top five — and the fifth, Leju, is a company this section does not otherwise cover, which says something about how much of this industry is invisible from outside China.
AgiBot took first place from Unitree by growing 562% year on year, and shipped more in six months than anyone managed in the whole of 2025.
The last complete year: 2025
A full twelve months, so this — and only this — can be set against the annual capacity claims further down. Two of the three figures are audited: Unitree's comes from its IPO prospectus, UBTECH's from its annual report.
5,500 humanoids sold in 2025, from the IPO prospectus — audited
1,079 full-size humanoids delivered in 2025 for RMB 820.6m — audited, from the annual report
All three are Chinese. No American or European maker published a 2025 humanoid shipment figure at all.
None of this is a statement about capability. A Unitree R1 sold to a university laboratory is not doing a shift at BMW. It is a statement about manufacturing, and manufacturing is what drives component prices down for everybody.
Cumulative output, for context
Counted since founding rather than per year, the August 2026 tally of bipedal humanoids runs Unitree 18,000, AgiBot 7,000, UBTECH 3,000, Galbot 1,000, Figure AI 1,000. These totals are kept out of the charts on purpose: a cumulative number cannot be set against an annual one, and the whole point of the comparison below is that both sides are measured per year. They also count fewer form factors than the shipment figures: AgiBot’s 7,000 is bipedal robots only, while its 8,400 shipped in H1 2026 includes wheeled models — the company passed its 20,000th robot of all types in September 2026.
Announced, not yet shipping
1X, XPeng, NEURA, Tesla have delivered nothing to outside customers to date (robots built and used in their own facilities do not count as deliveries). Their zeros are real zeros, and they are kept off the chart above rather than drawn as invisible bars.
No published figure
Agility, Boston Dynamics, Apptronik, Leju do not publish a unit count. That is not the same as zero, and nothing here estimates on their behalf — Agility in particular has 65,000 logged operating hours across named customers, which on any measure of useful work outranks most of the bars above.
Annual capacity announced
Same time base as the chart above — robots a year — but claimed rather than shipped. On its own chart rather than a second axis, because two scales on one frame is how this comparison gets misread.
Fremont lines stated as designed for 1,000,000 robots a year; Giga Texas for 10,000,000 from 2027
Hyundai intends 25,000 Atlas units; a US plant sized for 30,000 robots a year; new CEO Rohit Prasad (ex-Amazon Alexa/AGI) from 7 October 2026, output reported at ~4 a month in May
BotQ factory aimed at 12,000 humanoids a year; leasing floated at ~$600/month
Mass production by end 2026, deliveries in China and abroad from 2027
Liuzhou plant rated at 10,000 robots a year, live since September 2026; consumer U1 home deliveries since 16 September 2026 against 13,361 company-reported pre-orders — not counted here
Tesla's bar is not a rendering error. Its stated Fremont capacity is 33× the next largest claim on this chart, and it has delivered nothing to a customer — Optimus units working inside Tesla’s own factories are not sales.
The gap is the point, and it survives being generous about it. The largest verified annual output by any single company is 5,500 robots. The entire world shipped 19,100 in the first half of 2026 by Smart Analytics Global's count, or close to 25,000 by IDC's; double either for a full-year pace and the whole industry is running at roughly 38,000 to 50,000 a year. Tesla alone states a Fremont capacity of 1,000,000 a year — twenty to twenty-six times everything every manufacturer on earth is currently building — and has delivered none.
Unitree and AgiBot, first and second in both shipment charts, make no capacity claim at all.
Where each one is aimed
Unit counts flatten a real difference: these robots are not competing for the same work. Sorted by units built, with unpublished figures last.
| Company | H1 2026 | 2025 | Built total | Aimed at | What is actually deployed | Stated plan |
|---|---|---|---|---|---|---|
| Unitree | 5,900 | 5,500 | 18,000 | Research, developers, education, entertainment | 5,500 humanoids sold in 2025; profitable on 60% gross margin | No announced capacity figure; average price already down to ~$25,000 |
| AgiBot | 8,400 | 5,100 | 7,000 | Labs, developers, exhibitions, light industry | First in H1 2026 by all three trackers: 8,400 (SAG, 44%), 8,600+ (IDC, 35%), 9,700 (Counterpoint, 43%) | Hong Kong IPO in progress; ~4bn yuan of revenue expected this year |
| UBTECH | 700 | 1,079 | 3,000 | Automotive plants, smart factories, logistics | 1,079 full-size humanoids delivered in 2025 for RMB 820.6m — the only audited humanoid revenue anywhere. NIO, BYD, Foxconn, Audi-FAW, Zeekr | Liuzhou plant rated at 10,000 robots a year, live since September 2026; consumer U1 home deliveries since 16 September 2026 against 13,361 company-reported pre-orders — not counted here |
| Galbot | 900 | — | 1,000 | Retail and pharmacy — serving the public directly | 170+ unstaffed Galbot Stores across 40+ cities; first robot pharmacy licence in China | Hong Kong listing prepared; ET1 biped pre-orders opened September 2026 |
| Figure AI | — | — | 1,000 | Automotive logistics, and eventually the home | BMW Spartanburg — a Figure 02 supported 30,000+ X3 vehicles over ten months | BotQ factory aimed at 12,000 humanoids a year; leasing floated at ~$600/month |
| 1X | — | — | none yet | The home — the only consumer product here | Nothing delivered yet. 10,000+ pre-orders sold out in five days | First year of production sold out; $20,000 or $499/month |
| XPeng | — | — | none yet | Own retail stores first, then general commercial | Nothing yet. Production line running since September 2026 at 80%+ automation | Mass production by end 2026, deliveries in China and abroad from 2027 |
| NEURA | — | — | none yet | Industrial, plus domestic service via MiPA | Nothing yet for 4NE-1. Its cobot range is already in service | ~€1bn order book; Reger has stated five million robots by 2030 |
| Tesla | — | — | none yet | Its own factories first | Nothing outside Tesla's own facilities, and no customer. Q3 2026 update: production began on Fremont's high-volume lines ('Early Ramp'), no unit count published; outside customers 'towards the end of 2027' | Fremont lines stated as designed for 1,000,000 robots a year; Giga Texas for 10,000,000 from 2027 |
| Agility | — | — | not published | Warehouse and logistics | 65,000+ operating hours across nine customer sites — Schaeffler, GXO, Toyota, Mercado Libre. $300m+ of orders | Shareholders approved the SPAC merger on 2 Nov 2026 (84%+ in favour); Nasdaq listing as AGLT expected from 5 Nov (~$465m gross, Foxconn-led PIPE); GXO adds two sites and AGCO pilots Digit from 2027 |
| Boston Dynamics | — | — | not published | Automotive manufacturing | 2026 output committed to two customers: Hyundai's Metaplant centre and Google DeepMind | Hyundai intends 25,000 Atlas units; a US plant sized for 30,000 robots a year; new CEO Rohit Prasad (ex-Amazon Alexa/AGI) from 7 October 2026, output reported at ~4 a month in May |
| Apptronik | — | — | not published | Manufacturing and logistics | Pilots at Mercedes-Benz (Berlin and Kecskemét), GXO and Jabil | Apollo 3 in 2027 described as its first true commercial product |
| Leju | 600 | — | not published | Education, research and government demonstration | Fifth largest humanoid shipper in the world in H1 2026 — and not otherwise covered in this section, which is itself worth knowing | No published capacity figure |
What the numbers do not say
Three cautions, because a ranking invites the wrong conclusion:
- Annual and cumulative are different questions. The two charts are both per year so they can be compared; the cumulative totals are reported separately and never set against a capacity claim. Mixing the two makes any company look whatever you want it to look.
- Units built is not work done. A large share of the Chinese volume goes to universities, developers, showrooms and exhibitions. Agility's 65,000 logged industrial operating hours and UBTECH's audited RMB 820.6m of humanoid revenue are harder currencies than a unit count.
- The totals are not always comparable. The August tally counts bipeds; AgiBot has produced around 15,000 robots of all form factors against roughly 7,000 bipeds, and Galbot's shipping machines are wheeled. Headline numbers in this industry routinely compare different things.
- Announced capacity is a sentence, not a factory. Tesla's own record is the caution: production-ready by 2023, limited production in 2025, 10,000 robots that year — and by January 2026 none doing useful work.
Every figure is drawn from the twelve company profiles in this section, each of which cites its sources. Where a company publishes nothing, this page says so rather than estimating.