A manufacturer ships 100,000 humanoid robots in a year

Where things stand (mid-October 2026)

No company is close to 100,000 qualifying robots in a calendar year. The leaders are at roughly 20,000–45,000 a year:

• AgiBot: about 5,100 humanoids shipped in 2025.
  • Its cumulative production passed 10,000 in March 2026, 15,000 in June and 20,000 on 24 September 2026. That is roughly flat at about 5,000 a quarter, so about 20,000 for 2026.
  • It targets more than CNY 10bn revenue in 2027 and "several hundred thousand units a year in a few years".
  • Its Hong Kong IPO has slipped to 2027 (36Kr/Caijing, 13–14 Oct 2026; IDC via Robotics & Automation News, 2 Oct 2026).
• Unitree (listed on Shanghai's STAR Market since 19 Aug 2026):
  • More than 5,500 humanoids in 2025 and about 6,000–7,000 in the first half of 2026 (Counterpoint, 20 Aug 2026).
  • Revenue growth is slowing.
  • The factory funded by its IPO is designed for 75,000 humanoids a year (CLS, 10 Aug 2026).
• UBTech (listed in Hong Kong):
  • Its 2025 annual results (Hong Kong exchange filing, 31 Mar 2026) report 1,079 full-size humanoids and 12,759 small "non-embodied" humanoids.
  • Media citing UBTech put first-half 2026 sales at 16,123 units, of which 921 were full-size.
  • Most of the volume is the AI Wukong desktop humanoid. It has 14 servos, walks, and supports graphical and Python programming, so it meets the question's literal definition (size is not a requirement; only non-programmable toys are excluded).
  • A China Telecom programme targets 100,000 AI Wukong units (STCN, Nov 2025).
• Tesla:
  • Its Q2 2026 update says the first Optimus lines are being installed and Optimus capacity is still listed as "Construction"; Tesla has published no unit counts.
  • The Information (25 Sep 2026) reports output went from a few dozen a week in Q2 to several hundred a week in August. Tesla is aiming for more than 1,000 a week by end-2026 and has told suppliers to plan for 2,000–2,500 a week.
  • Problems reported: unreliable hands, robots that can't generalise to new tasks, China's magnet export curbs, and memory-chip shortages. Musk says the early ramp will be "very slow".
• Others are smaller: 1X targets 50,000 NEO robots in 2027, and XPeng targets more than 1,000 IRON robots a month by end-2026.

Reference class and base rate

The relevant comparison is a new hardware category where the leading firm goes from about 20,000–40,000 units a year to 100,000 within roughly four years.

• Consumer devices (drones, robot vacuums, VR headsets) usually make that jump quickly.
• Industrial and professional robots (cobots, service robots) often never do for a single firm.

Humanoids sit between the two. Mixing these cases gives a base rate of about 50–60%.

Forecasts point upward: Goldman Sachs expects about 890,000 humanoids worldwide in 2030, IDC more than 750,000, and Morgan Stanley 446,000 for China. Any of these would put the leader above 100,000. Analysts' near-term forecasts have also been revised up repeatedly.

Paths to YES

1. UBTech, 2027–28. Its small consumer humanoids plus full-size units, published in annual results by 31 March of the following year. I put about a 15–25% chance on 2027 alone. There is a risk the resolver treats the small-humanoid category as a mixed total, because UBTech says it includes remote-controlled, pre-programmed and toy units.
2. Tesla Optimus, 2028–29. If the ramp reaches several thousand a week. Disclosure is uncertain: a Manifold market gives only about 14% that Tesla officially reports even 10,000 units for 2026. Tesla's 2025 CEO award tracks "Bots delivered", which makes eventual quarterly reporting likely.
3. AgiBot, 2028–30. It reportedly has more than 100,000 units of capacity, though this is unverified. It would report through post-IPO filings or press releases.
4. Others: Unitree, XPeng, 1X, Figure, BYD, and edge cases such as robot vacuums fitted with arms.

Paths to NO

• Most current demand is research, entertainment and data-collection centres, and the International Federation of Robotics counted only about 7,000 humanoids sold for industrial or professional use in 2025.
• AgiBot's output has plateaued, Unitree's valuation has fallen, China's securities regulator is tightening IPOs for humanoid makers, and UBTech's receivables are rising.
• Tesla has a long record of slipping deadlines.
• Strict resolution rules (no cumulative totals, no mixed totals, no capacity figures) could block reports that would otherwise qualify.

Model

I assigned probabilities to the first qualifying calendar year: 2026 about 0.6%, 2027 about 17%, 2028 about 22%, 2029 about 13%, 2030 about 9%, and 2031 about 6% (only statements published within 2031 count). These total about 68% before a 5% haircut for overconfidence. I then mapped each year onto when the report would be published:

• 50–70% in the first quarter of the next year (Tesla's results deck in January; Hong Kong annual results by 31 March).
• About 20% in the second quarter (A-share annual reports by 30 April).
• 5–20% as year-to-date statements in the same year.

I added a small background chance for unexpected routes, then took cumulative products in Python. The result is about 14% by Q1 2028, 35% by Q1 2029, 51% by Q1 2030 and 64% by end-2031.

Sources

• counterpointresearch.com (20 Aug 2026)
• roboticsandautomationnews.com (2 Oct 2026)
• eu.36kr.com/en/p/4025612762763138
• UBTech filings on hkexnews.hk: 2026033102607.pdf, 2026082801691.pdf
• stcn.com/article/detail/3496898.html
• assets-ir.tesla.com TSLA-Q2-2026-Update.pdf
• electrek.co (25 Sep 2026)
• driveteslacanada.ca (Oct 2026)
• cnbc.com (24 Jun 2026)
• finance.biggo.com (Goldman Sachs forecast)
• cls.cn/detail/2449568
• Tesla proxy statement on sec.gov
• Manifold API

Question. By the last day of each quarter, has any manufacturer reported producing or shipping at least 100,000 humanoid or general-purpose mobile robots in a calendar year?

Definitions.

• Manufacturer: any company worldwide that manufactures robots or has robots manufactured under its own brand. A consolidated corporate group counts as one manufacturer; unrelated manufacturers cannot be combined.
• Humanoid robot: a programmable physical robot with a human-like torso, two articulated arms, and two legs designed for locomotion. Size, commercial deployment, and demonstrated general-purpose capability are not requirements. Fixed mannequins, animatronic displays, and nonprogrammable toys do not count.
• General-purpose mobile robot: a programmable physical robot with powered locomotion and one or more manipulators, designed to perform multiple distinct kinds of physical work by changing software, instructions, or end-effectors rather than replacing its main hardware. Wheeled, tracked, and legged platforms qualify. Product documentation must identify at least two distinct work functions, such as picking and placing objects, cleaning, assembly, or operating tools. Multiple versions of transport, navigation, inspection, or the same manipulation task do not by themselves establish general-purpose capability. A humanoid meeting the preceding definition need not independently meet this definition.
• Producing: completing assembly of new, whole, functional robots, including units retained for internal use. Components, unassembled kits, and refurbishment of existing robots do not count.
• Shipping: physically dispatching new, whole robots to customers, distributors, or external operators, including leased units. Orders, reservations, undelivered sales, and transfers within the same corporate group do not count. Each robot counts once within a shipment total.
• Calendar year: January 1 through December 31 of a named year. An actual year-to-date count reaching the threshold establishes the threshold for that calendar year; an annualized rate does not.

Resolves YES if a qualifying manufacturer report states that actual production alone or actual shipments alone reached at least 100,000 qualifying robots within one calendar year. Totals may combine qualifying models and both robot categories, without double-counting robots that fall into both categories. Production and shipments cannot be added together. Nonoverlapping reported subperiods within the same calendar year may be summed. A mixed total including excluded robot types counts only if the qualifying portion is separately reported or has an explicit lower bound of at least 100,000. An unqualified reported figure of 100,000 counts; an estimate, approximate figure, or range counts only if its stated lower bound reaches 100,000.

Exclusions (resolve NO).

• Production capacity, targets, forecasts, planned deliveries, order backlogs, installed fleet counts, and lifetime cumulative totals that cannot establish the count within one calendar year.
• Industry-wide or country-wide totals, or analyst estimates not adopted in a qualifying manufacturer report.
• Dedicated vacuum cleaners, lawnmowers, transport-only warehouse robots, delivery vehicles, drones, robotaxis, stationary industrial arms, and other single-purpose machines, unless they independently meet one of the qualifying robot definitions.
• Reports covering only a non-calendar fiscal year or a rolling twelve-month period, unless calendar-year totals can be established from the reported breakdown.
• At each horizon, resolve NO if no qualifying report has been published by that horizon.

Resolving source. Manufacturer-authored annual reports, securities filings, prospectuses, production or shipment releases, and official statements published on the manufacturer's own website or filed with a securities regulator or stock exchange. This is a narrower source requirement than Public confirmation (A2); interviews, employee posts, press estimates, and third-party reports alone do not qualify. Official product specifications or manuals establish robot eligibility. Apply the Resolution window, Observation period, and Resolution date (A3) conventions; a later qualifying revision resolves on its publication date.