AgiBot rolls out its 10,000th humanoid robot, doubling its fleet in 112 days

Unit counts are the first hard evidence that humanoid robots have left the prototype stage, and AgiBot is one of the first private companies to publish a five-figure total with a dated trail of milestones behind it. Investors pricing humanoid makers have had funding rounds and demonstrations to go on; a run rate of 44.6 units a day, rising to 55.6 a day by June, is a manufacturing fact that can be checked against each new announcement.

The second reason is the mix. AgiBot's December breakdown is one of the few published splits of a humanoid fleet by size class, and it shows that a market leader's volume is built largely on half-size and task-optimised machines. That should change how readers interpret every other volume claim in the sector, including the 100,000-unit expectation the Chinese industry ministry later set for embodied AI products in 2026.

Third, the milestone shifts the competitive question from whether Chinese makers can build humanoids at scale to who pays for them. The answer will decide whether the industry's next phase is a margin story or a subsidy story.

UBTech is the most comparable listed rival and the only one that reports full-size humanoid sales in a filing, with 1,079 units in 2025 and 921 in the first half of 2026, with capacity of more than 6,000 a year. UBTech concentrates on the Walker S series for car plants and factories, so its count is smaller but almost entirely full-size. AgiBot's like-for-like full-size lead, 1,742 cumulative A-series units against UBTech's 1,079 sold in 2025, is real but modest.

Unitree, which listed on Shanghai's STAR Market in August 2026, sells large volumes of quadrupeds and humanoids to research, education and entertainment buyers and reported 2025 revenue of RMB1,699.27 million, but its listing announcement gave no humanoid unit numbers. AgiBot competes with it directly in the half-size and research tiers, which is where price pressure will hit first.

In the United States the reference points are capacity plans rather than volumes. Figure AI had delivered more than 350 Figure 03 robots by 29 April 2026, 1X has capacity for up to 10,000 NEO units a year in Hayward, and Tesla is converting Fremont floor space to an Optimus line designed for one million a year. On units built and delivered, AgiBot is well ahead of every Western rival that has published a delivery figure.

AgiBot has not disclosed a valuation, and the two funding rows ROBOTNESS holds for it, a Series A extension announced on 3 September 2024 and a strategic round on 1 August 2025, carry no disclosed amounts. Any figure circulating in the press is therefore unverified for our purposes.

The listed comparables give a sense of what public investors pay for Chinese robot volume. Unitree's STAR Market issue price of RMB150.80 valued the company at about RMB60.99 billion, which its listing document put at 35.89 times 2025 revenue, against 19.37 times for UBTech and 20.12 times for Dobot. AgiBot's reported 2025 revenue of more than RMB1 billion sits in the same bracket as Unitree's RMB1.70 billion and UBTech's RMB2.0 billion.

The unit data matters for valuation because it lets investors separate volume from value. A company whose fleet is two thirds smaller robots should earn less revenue per unit than one selling only full-size machines; UBTech's 2025 full-size revenue averaged about RMB760,000 a robot. Until AgiBot publishes revenue by product family, its revenue per unit cannot be computed, and that is the number a listing would have to supply.

A run rate of roughly 16,000 to 20,000 robots a year draws heavily on four component groups, namely frameless torque motors, precision reducers such as strain-wave and planetary gear sets, battery packs and edge AI compute. Full-size machines need the highest-torque joints in hips and knees; half-size and task-optimised units use smaller parts that Chinese suppliers already make in volume for automation and consumer devices.

AgiBot has not named its suppliers. Its scale nonetheless matters to the Chinese reducer and actuator makers that have been building humanoid lines, and to Japanese precision-gear makers whose products have been the reference in industrial robots. If AgiBot's volume rests mainly on domestic parts, it is evidence that the cost curve for humanoid joints is now being set in China.

The OmniHand dexterous hand and the VR teleoperation kit point to a second chain, in tactile sensing and data collection. Data-collection training is one of AgiBot's eight listed application areas, so part of its demand is internal to the AI stack, with robots producing training data for the company's own models.

A listing application would be the single most informative event. Hong Kong or mainland prospectuses require audited unit sales, average selling prices, customer concentration and gross margin, which would answer whether the 10,000 and 15,000 milestones translate into proportionate revenue.

The 31 December 2026 deadline in the MIIT and SASAC special action on real-world training of humanoids will show how much of 2026 demand comes from state-designated scenario sites. A large share of AgiBot's second-half volume going to such sites would support the bear case on demand quality.

UBTech's 2026 annual results, due by 31 March 2027, and full-year 2026 rankings from research firms such as Omdia will show whether AgiBot's lead holds and whether the full-size share of the market is rising. A further AgiBot milestone release with a size breakdown would be the cheapest test of our thesis.

Thesis: AgiBot has proved it can manufacture at a pace no other humanoid maker has published, but its volume is weighted towards smaller and task-specific machines, so the milestone validates China's supply chain more than it validates factory demand for full-size humanoids. Confidence: medium.

The confidence is medium rather than high because the only size breakdown is from December 2025, and the mix may have shifted towards the A3 and Genie G2 since. The 15,000th unit, a G2 doing inspection work on a tablet line, hints that industrial use is growing. Against that, AgiBot has given no prices, no revenue per unit and no utilisation data, so the volume could also be flattered by demonstration and data-collection fleets.

Two things would change our view. A disclosed full-size share above 50%, or audited revenue per unit close to UBTech's roughly RMB760,000, would move us to a high-confidence bull reading. A listing document showing heavy reliance on government-linked buyers or rental fleets would move us to the bear case.

What share of the 10,000 and 15,000 units were full-size A-series robots, and how has that share moved since the December 2025 breakdown?

How many of the units rolled out have been sold and invoiced, how many are in rental or demonstration fleets, and how many are used internally for data collection?

What is the average selling price by family, and what gross margin does each family earn?

Which factories produce the robots, what is current annual capacity, and which joint components are sourced outside China?

How many robots are deployed outside China, and in which of the named regions are there paying industrial customers?