takeoff.watch

A manufacturer ships 1 million humanoid robots in a year

Will any manufacturer report producing or shipping at least 1 million humanoid or general-purpose mobile robots in a calendar year?

Chance by Q4 2031

18%

Cumulative probability

0%25%50%75%100%
  • Latest forecast

Forecast history

Forecast made on 16 Oct 2026

Latest of 5 forecasts · Methodology v2.1

Model reasoning

Forecast made 16 Oct 2026 · Methodology v2.1

Today the biggest humanoid robot makers each build only tens of thousands a year. Even optimistic analysts expect the whole world to sell fewer than a million a year in 2030. A "yes" most likely needs Tesla's new robot factories to reach full speed by about 2030 and Tesla to report that output officially. The development that would most change this forecast is clear evidence that robots have become useful enough for buyers to order hundreds of thousands.

Where things stand (mid-October 2026)

No manufacturer is anywhere near 1 million a year.

  • Industry size. Global humanoid shipments in H1 2026 were about 22,000 (Counterpoint) to 25,000 (IDC). Full-year 2026 is projected at more than 50,000 (Counterpoint) to about 75,000 (Goldman Sachs).
  • Agibot, the leader, shipped 8,600 to 9,700 units in H1. Its cumulative output went from 10k (March) to 15k (June) to 20k (September), so it is running at roughly 5k a quarter with little acceleration.
  • Unitree shipped more than 7,000 humanoids in H1. Its H1 revenue grew only 48.5% year on year, and its prospectus plans for 75,000 humanoids a year "over the next five years". Its founder says the "ChatGPT moment" for robots has not arrived.
  • Tesla:
    • Fremont. The first Optimus line replaced the Model S/X line and is "designed for 1 million robots a year". Tesla is preparing a second-generation line at Giga Texas "designed for long-term annual production capacity of 10 million robots" (Q1 2026 update). By August it was building several hundred units a week, aiming for more than 1,000 a week by year-end and eventually about 20,000 a week (The Information, via Electrek, Sep 25).
    • Problems. The hands are hard to build and break easily, the robot's software cannot yet generalize across tasks, and most units are used internally.
    • Latest from Musk. On Oct 10 he said Chinese magnet export curbs set the program back 2–3 months. He expects the Texas line to start in summer 2027, and external sales are "as early as" the second half of 2027.
    • No unit counts yet. Tesla's Q3 production release did not include any Optimus figure.
    • Nomura's view. The near-term Fremont capacity target is about 70k a year, with about 70k more in Austin by 2028.
  • Other aspirational targets: Xpeng aims for 1M a year by 2030, Agibot for CNY 100B revenue by 2030 (roughly 1M units), and 1X for 50k NEO units in 2027.
  • Analyst forecasts for the whole world in 2030: IDC more than 750k, Goldman about 890k (raised from 256k), and Morgan Stanley 446k for China's external sales. So one maker reaching 1M in 2030 would beat the consensus forecast for the entire market.

Reference class and model

I used two reference classes:

  • Emerging hardware categories. Consumer electronics with clear uses (drones, robot vacuums, VR headsets) reached 1M a year for a single maker within about 3–5 years of reaching tens of thousands a year. Complex machines (cars, industrial robots, Segway) took about a decade, or never got there.
  • Musk's Optimus targets. The early-2025 plan was 5–10k in 2025, 50–100k in 2026 and 500k–1M in 2027. Actual output was a few hundred in 2025 and roughly 15–25k in 2026, a slip of about a year that is widening.

The top maker would need to grow about 2.5 times a year from roughly 25k in 2026 to reach 1M in calendar 2030, or about 3.4 times a year to reach it in 2029. That matches the best historical cases, and the leaders' current growth is already slowing.

My estimates for the first calendar year in which any single maker reaches 1M:

Year Probability
2027 0.15%
2028 0.8%
2029 3%
2030 7.8%
2031 (only year-to-date reports published within 2031) 4.8%

On top of that I added about 1.2% for a robot vacuum with a pick-up arm (the "cleaning and fetching" example), which would only count if that model's units were reported separately, plus 0.3% for edge cases.

Reports lag the output. Tesla publishes its Q4 production in early January. Chinese listed companies publish annual reports by the end of April. Year-to-date reports cross the line within the year only when output is well above 1M. I split each year's probability 5% Q3 / 10% Q4 / 70% Q1 of the next year / 15% Q2 of the next year. I calculated the cumulative series in Python and it rises steeply at each Q1.

Readings I adopted: posts by Musk on X do not count, and figures must come from Tesla's investor-relations site or press releases. Robot vacuums whose "arm" only extends a mop or brush do not qualify.

Paths and the case against

  • Most likely YES: Tesla's Fremont line reaches its designed rate and the Texas line ramps in 2028–29, robot software becomes useful enough to drive demand, and Tesla reports 2030 output in January 2031.
  • Second path: a Chinese leader such as Agibot or Xpeng actually hits its 2030 target.
  • NO paths: demand stays mostly research, entertainment and data collection. China's planning agency has warned of a bubble, there are rumours of a crackdown on robot IPOs, and Unitree's share price has halved from its peak. Hand and supply-chain bottlenecks (Epoch AI estimates current reducer supply caps humanoids at about 500k a year in total) or US–China decoupling could also cap growth.
  • Strongest case for a higher number: if robot AI makes a demand jump by around 2028, Epoch's scenario gives 10–30M a year by 2030. The Texas building's steel frame is nearly complete, and Tesla's capital spending exceeds $25B. Thin Manifold markets show 17% that Tesla sells 1M cumulatively by 2030.
  • Why I stay below that: that Manifold market is about cumulative sales, and the Manifold crowd is visibly over-optimistic (17% for 1M cumulative humanoids by the start of 2028).

Sources

Resolution criteria

Opened 16 Sep 2026Not yet resolved

Question. By the last day of each quarter, has any manufacturer reported producing or shipping at least 1 million humanoid or general-purpose mobile robots in a calendar year?

Definitions.

  • Manufacturer: any organization worldwide that manufactures complete robots, including a manufacturer using contract assembly. A parent and its consolidated subsidiaries may be counted as one manufacturer; unrelated manufacturers may not be pooled. Production for internal use qualifies.
  • Humanoid robot: a programmable physical robot with a human-like torso, two articulated arms capable of manipulating objects, and either legs or a mobile wheeled or tracked base. A head, human-sized dimensions, bipedal walking, and general-purpose capability are not required. Stationary humanoid torsos, passive dolls, and animatronics limited to fixed entertainment sequences do not qualify.
  • General-purpose mobile robot: a programmable physical robot that can move through its environment and is capable of multiple materially different physical task types, rather than variations of one specialized function. Qualifying examples include a mobile platform capable of both object manipulation and inspection, or of both cleaning and fetching objects. Transporting different loads, cleaning different surfaces, or inspecting different assets alone does not establish general-purpose capability. Software updates, tools, or end-effectors may change between tasks, but replacement of the main robot platform may not be necessary. Manufacturer technical documentation must substantiate the capabilities; the label “general-purpose” alone is insufficient.
  • Robots meeting either definition qualify. Remote supervision or teleoperation does not by itself disqualify a robot. Dedicated robot vacuums, lawnmowers, warehouse transport robots, delivery vehicles, passenger vehicles, inspection drones, and stationary industrial arms do not qualify unless they independently meet a definition above.
  • Producing: completing assembly of new, operational robot units, not starting assembly, manufacturing components, or refurbishing existing robots.
  • Shipping: physically dispatching newly manufactured complete robots to customers, distributors, or deployment sites, including the manufacturer's own operating sites. Orders, reservations, bookings, software licenses, and transfers of used robots do not qualify. Count each unit only once within the shipment total.
  • Calendar year: January 1 through December 31 of a named year. A year-to-date report qualifies if it establishes that the threshold has already been reached within that year. A fiscal year qualifies only if it coincides with the calendar year, or the published data permit an exact calendar-year total.

Resolves YES if an eligible manufacturer reports at least 1,000,000 qualifying units produced, or at least 1,000,000 qualifying units shipped, within one calendar year. Qualifying models and both robot categories may be combined, without double-counting units. Production and shipments may not be added together to reach the threshold. Nonoverlapping monthly or quarterly figures from the same manufacturer may be summed if their definitions are consistent. A mixed-category total qualifies only when the qualifying portion is separately reported or its documented lower bound reaches 1,000,000. An unqualified reported figure of 1 million counts; an approximate figure or range counts only if its stated lower bound reaches the threshold.

Exclusions (resolve NO). Announced capacity, future targets, forecasts, annualized run rates, cumulative multiyear production or deployments, installed fleet size, and global industry totals do not qualify. For example, Amazon's June 2025 millionth-robot announcement is a cumulative deployment milestone, not qualifying calendar-year output. Resolve NO for a horizon if no qualifying report has been published by its end.

Resolving source. Manufacturer-authored annual reports, securities filings, production or shipment releases, and official statements on the manufacturer's own domain or issued through its official newswire account. These are the eligible company-authored sources under Public confirmation (A2); independent industry estimates, press paraphrases, and executive interviews alone do not count. Product documentation from the manufacturer may establish robot eligibility. Apply Resolution date (A3), the Resolution window, and the Observation period conventions. The output period must end after the opening date; later republication of an earlier year's figures does not qualify.

Download all forecasts (JSON)