Humanoid Robots: Size and Disputed Shipments

01How the market formed

Humanoid robots spent decades as a research curiosity before anyone treated them as a market. Honda quietly ended its ASIMO program in 2018 after 32 years of development that never reached commercial deployment, and Boston Dynamics' Atlas — the robot behind the viral parkour and backflip videos — remained a DARPA-funded research platform, not a product for sale. That changed on August 19, 2021, when Tesla used its AI Day event to reveal a humanoid concept it called Optimus. The reveal was closer to a statement of intent than a working robot, but it's the moment most industry observers point to as when "humanoid" turned from a research topic into an investable business category. Figure AI, founded the following year by a team that drew engineers from Boston Dynamics, Tesla, and Google DeepMind, was one of the first startups built explicitly around that new framing.

02Funding and IPOs

Within about a year of each other, three very different ways of financing a humanoid robot company all played out. In China, UBTech Robotics listed on the main board of the Hong Kong Stock Exchange on December 29, 2023, becoming the first publicly traded humanoid-robot maker anywhere. In the US, Figure AI raised $675 million in a Series B round in February 2024, at a $2.6 billion valuation, with Microsoft, the OpenAI Startup Fund, Nvidia, Jeff Bezos, and Intel Capital among the investors — alongside a collaboration agreement with OpenAI to build AI models for the robot. And in South Korea, Samsung Electronics took a different route entirely: rather than fund a startup, it built a stake in an existing humanoid maker, buying 10.2% of Rainbow Robotics (a 2011 spin-out from KAIST's Humanoid Robot Research Center) for roughly $40 million in January 2023, then steadily increasing that stake — reaching 35% and making Samsung the company's largest shareholder — while setting up a Future Robotics Office reporting directly to CEO Lee Jae-yong.

03Commercialization

A robot that can walk isn't the same as a robot someone pays to use, and the gap between those two milestones took another two-plus years to close. Agility Robotics' Digit became the first humanoid in a formal commercial deployment when it signed a multi-year Robots-as-a-Service agreement with logistics operator GXO in June 2024 — Robots-as-a-Service, or RaaS, means the customer rents the robot by the month or by the task instead of buying it, so the maker books revenue over years rather than at delivery — by mid-2026, Digit units at GXO's Flowery Branch, Georgia facility had moved over 100,000 totes, and a separate deployment at bearing maker Schaeffler's Cheraw, South Carolina plant had run daily eight-hour factory shifts for roughly 15 consecutive months. Tesla's Optimus program illustrates how wide the gap between target and delivery can still be: the company set a public target of 5,000 units built for internal factory use in 2025, but by its own account had only built "hundreds" by year-end — a shortfall it's now trying to close with public 2026 targets of 50,000 to 100,000 units and a next-generation Optimus 3 slated to enter production at its Fremont factory around July–August 2026.

04Mass production and shipments

While Western makers were still measuring output in the hundreds, Chinese manufacturers scaled into the thousands. Shanghai-based AgiBot, founded in 2023, and Hangzhou-based Unitree — which had spent seven years building quadruped "robot dogs" before entering the humanoid market — both moved from prototype to volume production over 2024–2025, backed by a domestic supply chain for motors, reducers, and sensors that Western makers still mostly import. The result shows up directly in the shipment count: global humanoid shipments reached roughly 13,318 units in 2025, an increase of nearly 480% over 2024 by Omdia's count, with Chinese manufacturers accounting for close to 90% of that total. Analyst firm TrendForce expects the pace to keep accelerating, forecasting that global shipments will exceed 50,000 units in 2026, growth of more than 700% year over year. TrendForce doesn't publish its own 2025 figure, so that growth rate can't be lined up against Omdia's 13,318 units directly. The same forecast projects Unitree and AgiBot alone capturing close to 80% of China's domestic shipments in 2026.

05Key players

Even the basic question of who ships the most humanoid robots doesn't have a single agreed answer. Analyst firm Omdia's General-Purpose Embodied Intelligent Robot report ranked AgiBot first for 2025, crediting it with 5,168 units and a 39% share of global shipments, followed by Unitree at 4,200 units (32%) and UBTech at 1,000. Unitree disputes that count directly, self-reporting more than 5,500 units shipped and claiming the volume crown for itself — a gap that comes down to differences in what each side counts as a shipped, revenue-generating unit rather than a demo or pilot placement. Figure AI, by contrast, shipped roughly 150 robots in 2025 — a fraction of either Chinese company's volume, reflecting a strategy built around a small number of higher-value enterprise deployments rather than manufacturing scale.

06Price and listed makers

Two questions come up more than any other about this segment, and neither is answered by a shipment table. The first is price. Unitree publishes its own, and the spread inside a single maker's lineup is wide: roughly $5,900 for the R1, a stripped-down developer platform, up to around $90,000 at the top of the range, with the G1 at roughly $16,000 sitting in between as the model built for volume. Most Western makers don't publish list prices at all, because they aren't selling units off a shelf — Agility's Digit reaches customers through the Robots-as-a-Service arrangement described above, where the number that matters is a monthly rate rather than a sticker price.

The second is which of these companies are publicly traded. Of the makers named in this article, three are: UBTech Robotics (9880.HK, Hong Kong), Tesla (TSLA), and Rainbow Robotics (277810, KOSDAQ) — the last of which is the route Samsung Electronics (005930, KOSPI) took into the segment rather than building its own. Figure AI, AgiBot, Agility Robotics, and Neura Robotics were all still private as of this writing. Unitree filed for a Shanghai STAR Market listing that received final regulatory approval in July 2026, with the trading debut not yet set. Nothing here is a recommendation — it's just which names have a ticker behind them and which don't.

07Market size

Ask five research firms for the size of the humanoid robot market and expect five different answers — not because any one of them is wrong, but because they're measuring different things. Grand View Research put the global market at roughly $1.55 billion in 2024, counting hardware alone; DataBridge Market Research put the same year at roughly $3.65 billion, apparently using a broader scope. MarketsandMarkets, whose multi-year series is used for the chart below, estimates $2.92 billion in 2025, growing to $15.26 billion by 2030 and $50.27 billion by 2035 — a 39.2% compound annual growth rate over the first stretch. Goldman Sachs, using its own scope and methodology, separately projects the market will reach $38 billion by 2035 (more than sixfold its own earlier $6 billion estimate) on the back of roughly 1.4 million cumulative shipments. For the same year, that's a 1.3x spread against MarketsandMarkets' figure — a useful gauge of how far two credible methodologies can land apart. None of these numbers should be read as more precise than the methodology behind it.

Data note: Market-size estimates for this segment vary by 2–3x across research firms depending on scope (hardware only vs. hardware + software + services) — Grand View Research put the 2024 global market at ~$1.55B, DataBridge Market Research at ~$3.65B for the same year. The shipment ranking above also has an open dispute: analyst firm Omdia ranks AgiBot #1 for 2025 (5,168 units, 39% share), while Unitree self-reports 5,500+ units and claims the volume crown itself. Both figures are shown; neither is presented as the single settled count. The `shipmentVolume` series above also mixes counting methodologies across years — the 2025 point is Omdia's count (13,318 units), while TrendForce's 2026 forecast (>50,000 units) uses TrendForce's own separate, lower count of 2025 shipments as its baseline, so the two points aren't on a strictly comparable year-over-year basis.

Market data as of 2026-07

08By region

The regional split in this segment isn't just about who's ahead — it's about which part of the problem each region is actually betting on. China has bet on manufacturing scale and vertical integration: Unitree makes more than 95% of its components in-house, and the country's dense electronics supply chain lets AgiBot, UBTech, and others iterate and cut cost quickly. The US has bet on software and AI: Figure's OpenAI partnership, 1X's backing from OpenAI, and Tesla's in-house Full Self-Driving stack all treat the robot's "brain" as the differentiator, at the cost of building fewer, more expensive units. Europe, led by Germany's Neura Robotics — which raised roughly $1.4 billion in a single 2026 round, among the largest robotics rounds ever — is betting on cognitive humanoids purpose-built for European manufacturing rather than either the US or Chinese playbook.

Japan, historically the country that built ASIMO, is playing a different position entirely. Analyst firm TrendForce describes Japan as concentrating on the core components — actuators, sensors, control systems — that raise the barrier to entry, citing Harmonic Drive's joint reducers as the example; both US and Chinese humanoid makers still depend on that layer. Complete-system work hasn't stopped: the same report points to Kawasaki Heavy Industries' Kaleido 9 humanoid (30 kg payload, VR-teleoperated) on the Japanese side. It just isn't aimed at the shipment race China and the US are running.

09Challenges ahead

The cost curve

Whether this segment scales may depend less on any market-size forecast than on the manufacturing cost curve. Goldman Sachs found humanoid manufacturing costs fell roughly 40% in a single year, from a $50,000–$250,000 range to $30,000–$150,000, against the 15–20% annual decline its analysts had expected; the driver is mostly the same components-cost decline that has already played out in quadruped robots and electric vehicles. On that basis, the bank estimates a plausible labor-substitution rate of 5–15% for car manufacturing and dangerous jobs like disaster rescue and nuclear reactor work, which would imply global demand of 1.1 million to 3.5 million units — well above its own base-case 2035 shipment forecast.

Profitability at scale

What's still unresolved isn't whether a humanoid can walk, carry a tote, or work an eight-hour shift — Digit and Optimus have each demonstrated that in some form. It's whether any maker, in any region, can build and sell these robots at a large enough scale to be profitable — a question the shipment and market-size numbers above don't yet answer either way.

10Timeline at a glance

11FAQ

12Sources