How many humanoids were shipped and sold, what makers report in their own filings, what they cost, and which you can get today. Every figure links to its source.
Statistics only. No affiliation, sponsorship or endorsement. Not investment advice.
Each figure comes from a different publisher and counts something different, so they can’t be added up into one robot total. Read the note under each one.
Humanoids · Analyst estimate
≈19,100About 19,100
Half-year shipments
SAG · units H1 2026 · January–June · Worldwide
Includes wheeled and smaller humanoids. Other trackers count differently, and this is an estimate, not a count of deliveries.
Separate published estimates and supplier-survey observations for full year 2025; the bars share a zero baseline.
Different tracker definitions are separate observations, not a confidence interval or a total to average.
Sales, shipments, installations and factory production are different measures.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars start at zero and share one scale. Where a source says “about” or “more than”, the bar is drawn at that number. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Source: Smart Analytics Global (); Counterpoint Research (); IDC () · Compilation as of . Inspect data contains full source titles, access dates, limitations and related atlas records.Inspect data
Evidence rows for “How big is the humanoid robot market?”. The values and definitions match the chart and CSV.
Compare the attributed H1 snapshots independently; approximate values and lower bounds are plotting anchors, not exact counts.
Different tracker definitions are separate observations, not a confidence interval or a total to average.
Sales, shipments, installations and factory production are different measures.
No verified Q3 or October-year-to-date industry total was found in the reviewed research.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars start at zero and share one scale. Where a source says “about” or “more than”, the bar is drawn at that number. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
When you quote a figure, keep its publisher, its wording (“about”, “more than”) and the period it covers.
What this means: IDC counts more humanoids shipped in the first half of 2026 (nearly 25,000) than in all of 2025 (nearly 18,000): fast growth from a small base.
Forecasters expect about 50,000 to 60,000 humanoids to ship worldwide in 2026. SAG counted about 19,100 in the first half and expects close to 60,000 for the year, which means it expects the second half to be about twice the first. Third-quarter results, due from late October, will be the first test.
Older forecasts, such as Goldman Sachs’ 2024 estimate of about 1.4 million humanoids a year by 2035, are left out because their publishers have since issued newer views or the figures are out of date.
What revenue do humanoid market estimates cover?
IDC reports revenue estimates for two different observation periods. They do not form a like-for-like annual trend.
Source: IDC (); IDC () · Compilation as of . Inspect data contains full source titles, access dates, limitations and related atlas records.Inspect data
Evidence rows for “What revenue do humanoid market estimates cover?”. The values and definitions match the chart and CSV.
Retain the original current-period tracker estimates separately from long-term forecasts.
Full-year and half-year measures must remain separate.
No consensus revenue total is computed.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Source: Unitree Robotics (Undated); Unitree Robotics (Undated); Unitree Robotics (Undated) · Compilation as of . Inspect data contains full source titles, access dates, limitations and related atlas records.Inspect data
Evidence rows for “How much does a humanoid robot cost?”. The values and definitions match the chart and CSV.
Compare the configurations named in each row; keep their dated maker source and exclusions with the value.
R1 AIR is not standard R1; G1 base excludes secondary development and EDU is distinct; H2 base is separate from EDU/Plus.
No matched-configuration price index or deployment-cost total is implied.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars start at zero and share one scale. Where a source says “about” or “more than”, the bar is drawn at that number. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
One company filing, same annual revenue-recognized sales basis, in original currency. Source pages 145 and 206.
The 2023 sample contains only five recognized sales.
Do not infer a global transaction-price trend or the cost of a working deployment.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars start at zero and share one scale. Where a source says “about” or “more than”, the bar is drawn at that number. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
One tracker and one period; do not insert a company filing into this tracker’s numerator or denominator.
SAG includes full-size and smaller bipeds, wheeled and ultra-bionic humanoids; Leju covers Kuavo only, excluding Aelos and Roban.
Selected disclosed estimates, not a universal vendor ranking.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Only a few makers publish audited unit counts, mostly in stock-exchange filings. Unitree sold 5,215 humanoids in 2025; UBTECH sold 921 full-size humanoids in the first half of 2026.
Company filingCovers: Full year 2025WorldwideAs of
UBTECH sales volume of full-size embodied intelligent humanoid robots, all scenarios, 2025. Separately, 12,759 non-embodied small humanoid units were sold
Company filingAs publishedCovers: Full year 2025WorldwideChecked
UBTECH sales volume of full-size embodied intelligent humanoid robots, all scenarios, 2025. Separately, 12,759 non-embodied small humanoid units were sold
Sales volume, not orders.
Do not add the 12,759 small educational/consumer humanoids; they are a different product class.
Each maker’s own figure from its prospectus, annual or interim report, or official post. Periods differ, so the rows are not drawn to one scale.
Shipped, sold and built are different measures.
Makers define “humanoid” differently; check each row’s scope.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Filed revenue is still small: Unitree reported RMB867.8 million from humanoids in 2025 and UBTECH RMB820.6 million from full-size humanoids. Agility’s 2025 robot sales were about US$1.55 million.
Currencies, periods and revenue definitions differ (some include services or robot dogs).
Revenue is not the same as units sold.
Each row keeps its own unit, so the rows are not drawn to one scale.
Unitree
Humanoid product revenue
RMB867,831,900RMB
Company filingCovers: Full year 2025WorldwideAs of
Unitree main-business revenue from humanoid robots (RMB86,783.19 x 10,000), 51.78% of main-business revenue. 2024: RMB107.3 million; 2023: RMB2.97 million
Unitree only; excludes quadrupeds (RMB697.6 million) and components.
Company filingCovers: H1 2026 · January–JuneWorldwideAs of
UBTECH revenue from full-size embodied intelligent humanoid robot products and services (Walker S series, Cruzr Y1, Walker C1, U1 series), all scenarios; 46.5% of group revenue; H1 2025: RMB38.2 million
Company filingAs publishedCovers: Full year 2025WorldwideChecked
Unitree main-business revenue from humanoid robots (RMB86,783.19 x 10,000), 51.78% of main-business revenue. 2024: RMB107.3 million; 2023: RMB2.97 million
Unitree only; excludes quadrupeds (RMB697.6 million) and components.
Audited full-year figure; no product split published for 2026 so far.
Company filingAs publishedCovers: H1 2026 · January–JuneWorldwideChecked
UBTECH revenue from full-size embodied intelligent humanoid robot products and services (Walker S series, Cruzr Y1, Walker C1, U1 series), all scenarios; 46.5% of group revenue; H1 2025: RMB38.2 million
Revenue from each maker’s filing, in the currency it reports. Not converted or added up.
Currencies, periods and revenue definitions differ (some include services or robot dogs).
Revenue is not the same as units sold.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
More than US$300 millionUSD, multi-year contract value
Company filingCovers: As of May 2026Not statedAs of
Committed multi-year orders for Digit 5, mainly one customer: about 1,000 Digit 5 robots on three-year RaaS terms, with warrants vesting as robots are deployed
Not revenue; depends on contractual milestones, product features and specifications.
Company reportedCovers: First 5 days after 28 October 2025 launchMainly United StatesAs of
NEO home humanoid pre-orders that 1X says filled its next year of production capacity; stated in its 30 April 2026 factory post. Factory capacity stated as up to 10,000 NEOs per year
Pre-orders, not deliveries or revenue; refundable-deposit terms not given in the post.
Company reportedCovers: Cumulative to September 2026WorldwideAs of
All AgiBot robots produced since founding (Feb 2023), including full-size and half-size humanoids, wheeled G-series and other forms; earlier milestones: 10,000th on 30 Mar 2026, 15,000th on 28 Jun 2026
More than US$300 million USD, multi-year contract value
Company filingMore thanCovers: As of May 2026Not statedChecked
Committed multi-year orders for Digit 5, mainly one customer: about 1,000 Digit 5 robots on three-year RaaS terms, with warrants vesting as robots are deployed
Not revenue; depends on contractual milestones, product features and specifications.
Concentrated in a single customer.
Digit 5 was unveiled on 15 Sep 2026; no Digit 5 deliveries are reported yet.
1X (NEO)Pre-orders booked after launch1x-neo-preorders-2025
10,000 pre-orders
Company reportedAs publishedCovers: First 5 days after 28 October 2025 launchMainly United StatesChecked
NEO home humanoid pre-orders that 1X says filled its next year of production capacity; stated in its 30 April 2026 factory post. Factory capacity stated as up to 10,000 NEOs per year
Pre-orders, not deliveries or revenue; refundable-deposit terms not given in the post.
Private company; unaudited. The post says some buyers get NEO in 2026, others later.
Company reportedAs publishedCovers: Cumulative to September 2026WorldwideChecked
All AgiBot robots produced since founding (Feb 2023), including full-size and half-size humanoids, wheeled G-series and other forms; earlier milestones: 10,000th on 30 Mar 2026, 15,000th on 28 Jun 2026
Production, not shipments or sales.
Includes non-humanoid forms (the 15,000th unit was a wheeled AgiBot G2).
AgiBot is unlisted; no audited figures. Swancor (SSE 688585), which AgiBot controls, states it operates independently and discloses no AgiBot figures.
Company announcements and filings, kept separate because orders, pre-orders, contract value and production measure different things.
Orders can be cancelled or delayed; production is not sales.
These are company statements, not independent counts.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
IDC’s definition of a humanoid includes wheeled robots.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Each row keeps its own unit, so the rows are not drawn to one scale. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Source: Simbeing (Undated); Simbeing (Undated) · Compilation as of . Inspect data contains full source titles, access dates, limitations and related atlas records.Inspect data
Evidence rows for “How much of the atlas has a usable price?”. The values and definitions match the chart and CSV.
Count positive finite USD figures and apply the reviewed exclusion rules. Show the entire category denominator; missing documentation is not zero cost.
Screened does not mean verified current sale price.
The two rows per category overlap; their values must not be added.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars show a share of the robots Simbeing lists in each group. They are not market shares. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
Source: Simbeing (Undated) · Compilation as of . Inspect data contains full source titles, access dates, limitations and related atlas records.Inspect data
Evidence rows for “What do humanoid availability labels mean?”. The values and definitions match the chart and CSV.
Count the frozen catalogue’s stored status fields; each row uses the full category denominator.
Commercial can include internal use, services and preorders.
Pilot does not establish a paid deployment; historical records remain in the denominator.
The research edition is . Publication dates, measurement periods and research review dates describe different events. Missing observations are not zero. Catalogue counts describe documented records, including historical entries, rather than robots operating or sold.
Bars show a share of the robots Simbeing lists in each group. They are not market shares. A source filter changes the visible rows, table, download and citation together; it keeps the original chart scale.
When you quote a figure, keep its publisher, its wording (“about”, “more than”) and the period it covers.
Edition note · 2026-10-12
New evidence in this edition
In this edition every figure was checked again against its original document: company annual and interim reports, stock-exchange filings and prospectuses, and the research firms’ own releases. None of the published numbers was wrong. Six now cite the original document instead of a news report, one repeated figure was retired, and 65 new figures were added, most of them from company filings.
New sections cover what makers themselves report (units, revenue and orders), robot dogs, home-robot makers, warehouse and delivery robots, driverless trucks, mining trucks and drone deliveries. The previous edition stays available. We don’t treat new robots added to the atlas as market growth.
Quote each number with its publisher, what it measures and the period it covers. Every chart has a data table, a CSV download and ready-to-copy citation text.
What the labels mean
Analyst estimate: a research firm’s estimate, using its own definition. Company reported and company filing: the company’s own number, from a statement or an official filing. Published offer: one listed price for one version, on the date shown. Atlas calculation: a count from Simbeing’s own records, not a market share.
Dates, gaps and comparisons
Each figure shows three dates: the period it covers, when the source published it, and when we checked it. “About” and “more than” stay as the source wrote them. A missing number means it wasn’t published, not zero. Fleets, shipments, sales, installations, trips and forecasts measure different things, so they are never added together.
No public source yet gives a complete 2026 humanoid total, a count of deliveries by every maker, or a fair worldwide average price. A maker’s page today may differ from the dated figure here.
How the atlas snapshot is counted
The counts are fixed on the edition date, so they stay citable. Retired models are included. “On sale” and “in pilots” are how we classify each model; they don’t prove a robot is being delivered today, and a pilot isn’t always paid.
Downloads hold the figures, their sources and their notes, never copied reports or images. The original publishers keep the rights to their own material.