Walking Was the Easy Part: Why the Humanoid Robot Market Is Now a Cost-Per-Part Business

The humanoid industry spent 2026 learning to ship. Pheonix Research sizes the global humanoid robot market at USD 1.15 billion in 2025, reaching USD 3.42 billion by 2033 — a 2.97-times expansion (Pheonix Research calculation: USD 3.42bn divided by USD 1.15bn). Units are compounding several times faster than revenue. That gap, not the walking, is the market’s real story.

At the end of June 2026, UBTECH opened orders in Shenzhen for a full-size humanoid line it says is built for the production line rather than the trade-show stage. By the close of launch day the company reported cumulative orders of 13,361 units for its UWORLD U1 series, priced from RMB 119,800. That is not a technology event. It is an order book.

Unitree, the volume leader in this category, had already made the same turn. In January 2026 the company clarified its 2025 figures: more than 5,500 pure humanoid robots actually shipped to end customers, and more than 6,500 units off the production line — a gap of roughly a thousand machines between what was built and what was sold (Pheonix Research calculation: 6,500 minus 5,500). The factory is running ahead of the order book. Time reported in July 2026 that those 5,500 units represented more than a quarter of the entire global humanoid market by volume. Unitree’s stated ambition for 2026 is as many as 20,000 units — a 3.64-times step-up on 2025 (Pheonix Research calculation: 20,000 divided by 5,500), an increase of 263.6%.

Two caveats belong in the same sentence as those order numbers. UBTECH’s 13,361 figure counts orders placed ahead of delivery, and the company’s release does not separate binding contracts from reservations; even taken at face value, it is 2.43 times Unitree’s entire 2025 shipment (Pheonix Research calculation: 13,361 divided by 5,500) — an order book measured against a delivery record. Unit growth of this kind is real. It is not yet revenue.

Why the dollars lag the units

Pheonix Research’s Global Humanoid Robot Market report values the market at USD 1.15 billion in 2025 and projects USD 3.42 billion by 2033, a CAGR of 13.2% across a forecast window beginning in 2026. Over the eight years from the 2025 base that is a 2.97-times expansion (Pheonix Research calculation: USD 3.42bn divided by USD 1.15bn).

Set that against Pheonix Research’s wider coverage of autonomous robots, the much larger pool that shares the same buyers and the same deployment sites. That market runs from USD 18.90 billion in 2025 to USD 67.80 billion in 2033 at a 17.30% CAGR — a 3.59-times expansion (Pheonix Research calculation: USD 67.80bn divided by USD 18.90bn). On those figures the humanoid category is 6.1% of the autonomous-robot market it sits inside (Pheonix Research calculation: USD 1.15bn divided by USD 18.90bn), and it grows more slowly in percentage terms than the machines it is meant to displace.

There is no mystery in that arithmetic. When average selling prices fall, dollar growth decelerates even as unit growth accelerates. Unitree cut the pretax price of its flagship G1 from USD 16,000 to USD 13,500 in eighteen months — a 16% reduction, with the current price equal to 84% of the earlier one (Pheonix Research calculation: USD 13,500 divided by USD 16,000) — and now sells the smaller R1 biped for under USD 5,000. The same company’s four-legged robots fell from USD 45,000 to under USD 2,000 over six years. Deflation is how a hardware category gets adopted. It is also why a market can multiply its unit count and add comparatively little to its revenue line.

Three forecasts, three boundaries

Rarely has one market been described in figures this far apart. Goldman Sachs Research projects the humanoid total addressable market at USD 38 billion by 2035, on roughly 1.4 million units shipped, with a base case of more than 250,000 units in 2030 — almost all of them industrial. Morgan Stanley Research puts the market above USD 5 trillion by 2050, but that figure explicitly includes supply chains, repair, maintenance and support, with more than a billion units in use and 90% of them industrial or commercial. Pheonix Research’s number is USD 3.42 billion by 2033.

These are not competing claims so much as different boundaries drawn around the same machine. One is hardware to 2035. One is an installed-base service economy a quarter-century out. One is a hardware market on a 2025 base with a 2033 horizon. The discipline worth carrying into any board conversation is not which number is right — it is which number answers the question being asked. A USD 3.42 billion hardware market and a USD 5 trillion installed-base economy can both be true.

What a paid shift actually looks like

The evidence that now matters is counted on customer lines rather than in laboratories.

Figure ran its second-generation humanoid at BMW Group Plant Spartanburg for eleven months: ten-hour shifts, five days a week, more than 90,000 parts loaded across more than 1,250 hours of runtime, contributing to the production of over 30,000 X3 vehicles. That works out at approximately 72 parts loaded per operating hour (Pheonix Research calculation: 90,000 divided by 1,250) — a throughput figure, not a capability figure, and the first kind a plant’s finance function can price. Figure has since retired the robot and folded what it learned into its third generation; the forearm, the machine’s most frequent hardware failure point at BMW, was re-architected as a result.

Agility Robotics, meanwhile, has moved more than 100,000 totes with its Digit humanoid at GXO’s Flowery Branch facility, a deployment the company frames as an ROI milestone rather than a demonstration. In June 2026 it agreed to go public through a merger with Churchill Capital Corp XI at a USD 2.5 billion pre-money equity value, with more than USD 620 million of expected gross proceeds including a USD 200 million PIPE led by Foxconn, per the transaction announcement filed with the SEC. The company expects the deal to close in 2026.

The rulebook is still a draft

Every deployment described above is an industrial one — a workplace from which the public is excluded or its access restricted. That is precisely the scope of ISO 25785-1, Safety requirements for dynamically stable industrial mobile robots (legged, wheeled, or other forms of locomotion) — Part 1: Robots, under development by ISO’s robotics committee, TC 299. As of 2026 the standard remains a committee draft, at stage 30.20, consultation initiated; Part 2, covering the integration of applications, has yet to be developed. Mobile robots used outside industrial environments sit under a different standard, ISO 13482.

Read that sequence against the deployment record and it is uncomfortable. Fleets are running full shifts on live assembly lines while the international safety framework that will govern them is unfinished. Until it lands, each operator writes its own risk assessment, its own intervention protocol and its own insurance case — a fixed cost per site, and the one kind of cost that does not fall when volume rises.

Pheonix Research’s own assessment of this market carries the same tension. Competitive intensity is high across a fragmented field of eleven tier-one players; the technology sits at an emerging maturity stage despite very high patent activity; regulatory pathways are standardised but complex; and overall risk is assessed as high. Vertically integrated operation is the norm, and distribution runs direct to the customer.

The cost curve has a home address

The forces pushing prices down are concentrated in one place. China’s Ministry of Industry and Information Technology set the frame in its 2023 guidance on humanoid robotics: an initial innovation system and batch production at internationally advanced levels by 2025, and by 2027 a world-advanced industry capable of scaled development, anchored by two or three “ecosystem” enterprises of global influence. Reuters reported in May 2025 on how Chinese developers were combining that hardware base with domestic AI models to make the machines economically useful, and noted President Xi Jinping inspecting AgiBot’s robots in Shanghai.

Vertical integration is doing the rest of the work. BYD self-produces around 80% of core components — harmonic reducers and torque sensors among them — which it credits with a 30–40% cost reduction, and has said it plans to deploy 2,000 robots on its own lines. XPeng’s Iron robot reuses its autonomous-driving and cabin-intelligence stack for perception and interaction. Unitree manufactures its own actuators, the components that account for 50–70% of a humanoid’s cost.

Pheonix Research expects Asia-Pacific to be the fastest-growing region on the strength of exactly this — government robotics programmes and industrial modernisation — while North America holds the largest share today on the back of research spending and early adoption. The demand anchor and the cost base are in different places, and for now the cost base is winning the argument.

What to watch

Three indicators will separate the companies that ship from the companies that present.

Parts per hour, per robot, on a paid line. Figure’s roughly 72-per-hour benchmark at Spartanburg is the first figure of its kind in general circulation. If it climbs while human interventions fall, the unit economics work. If it does not, the deployment stays a pilot with better marketing.

Whether ISO 25785-1 stops being a draft. Publication of Part 1, and then Part 2’s integration requirements, converts a bespoke per-site risk case into a standard one. That is the point at which humanoids become underwritable at scale.

The unit-versus-dollar divergence. Pheonix Research’s dollar CAGR for this market is 13.2% through 2033. If unit volumes keep compounding in the tens or hundreds of percent inside that envelope, value migrates to whoever owns the cost base — actuators, reducers, sensing, control — rather than whoever owns the most compelling demonstration. Solar and batteries ran the same pattern; the humanoid is now running it at speed.

The humanoid has passed the walking test. What it has not yet passed is the invoice test — and the standard that lets a plant manager sign one.

Pheonix Research covers this cluster in its Robotics & Intelligent Automation segment, including the Global Humanoid Robot Market outlook and the Global Autonomous Robot Market outlook to 2033. See also the 2026 Global Humanoid Robot Market press release.

Frequently Asked Questions

How large is the global humanoid robot market?

Pheonix Research values the global humanoid robot market at USD 1.15 billion in 2025 and projects USD 3.42 billion by 2033, a CAGR of 13.2% across a forecast window beginning in 2026 — a 2.97-times expansion over the period, a Pheonix Research calculation from the two levels (USD 3.42bn divided by USD 1.15bn).

Which regions lead and grow fastest in humanoid robot adoption?

North America is the largest market on the strength of research investment and early industrial adoption, while Asia-Pacific is the fastest-growing region, supported by government robotics programmes and industrial modernisation.

What are the main applications of humanoid robots?

Pheonix Research covers deployment across healthcare and elder care, education and research, industrial and manufacturing, retail and hospitality, and defence and security, with industrial material handling the leading commercial use case today.

Who are the key players in the humanoid robot market?

Pheonix Research identifies SoftBank Robotics, Honda, Toyota, Boston Dynamics, PAL Robotics, FANUC and Kawada Robotics among the leading participants, within a fragmented field of eleven tier-one players and high competitive intensity.