Robotics & physical automation

Robotics & physical automation

Industry ID: 13 | Slug: robotics-automation | Researched: 2026-09-15 | Analyst: agent

1. Definition and boundaries

In scope. Robot vendors and robot technology. Industrial articulated arms, SCARA and delta robots and the collaborative-robot (cobot) segment. Humanoid and legged robots. Warehouse and intralogistics automation: goods-to-person systems, autonomous mobile robots, case handling, inventory drones. Agricultural robotics and off-road field autonomy. Surgical and other medical robotics. Construction robotics. Robot software: foundation models for manipulation, simulation and digital twins, fleet orchestration, robot operating systems. Components: precision reducers (strain wave and cycloidal), servo motors and actuators, force-torque and tactile sensors, motor-control semiconductors, and the permanent-magnet supply chain feeding them.

Explicitly out of scope, and who owns it.

• Demand-side adoption of automation by manufacturers and logistics operators → sector 09 (industrial and supply chain). That dossier covers whether operators are buying, what drives the purchase and how it is financed. This one covers the vendors, their unit economics and their technology roadmaps. The boundary is genuinely awkward for Symbotic and AutoStore, which appear in 09 as evidence of adoption and here as market analysis; both treatments are deliberate and the numbers used are the same.
• Autonomous passenger vehicles and on-road trucking autonomy → sector 10 (automotive and mobility). Off-road and field autonomy stays here, because its regulatory profile is completely different: private land, no public-road licensing regime.
• Defence unmanned systems and military UGV/UAV procurement → sector 08 (aerospace, defence and space). Dual-use ground robots such as Ukrainian UGV makers touch both; they are noted here but not analysed.
• The AI foundation-model layer in general → sector 01. Robot-specific manipulation and world models are in scope here; general LLM capability is not.
• Semiconductors as an industry → sector 03. Motor-control and sensing ICs are in scope here only as robot components.
• Rare-earth mining, separation and magnet production as an industry → sector 09. Their effect on robot production is in scope here and is one of this dossier's more important findings.

Boundary disputes worth naming. First, whether humanoids are a robotics category at all or a capital-markets category — on the 2026 evidence they are substantially the latter, and this dossier treats the financing story and the deployment story separately on purpose. Second, whether surgical robotics belongs with robotics or with medical devices; it is included here because its competitive dynamics are now robotics dynamics (platform lock-in, installed base, instrument razor-and-blade) but its regulatory clock is entirely FDA's, which is sector 06's world. Third, whether NVIDIA is a robotics company; it is treated here as a supplier whose claims about the sector are interested evidence, not neutral evidence.

2. Subcategories

1. Industrial articulated robots — the historic core: 6-axis arms for welding, material handling and assembly, sold through integrators, certified under ISO 10218, and characterised by decade-plus service lives and enormous installed-base inertia.
2. Collaborative robots (cobots) — force-limited arms rated to work alongside people without fencing. Lower payload, far lower integration cost, and the segment most open to third-party AI because the vendors deliberately ship open interfaces.
3. Warehouse and intralogistics automation — goods-to-person storage and retrieval, case handling, AMR picking fleets and inventory drones. The segment with the largest audited revenue and the clearest labour-cost business case.
4. Humanoid and legged robots — bipedal and quadrupedal general-purpose platforms. Enormous capital, minimal verified industrial revenue, and the sector's dominant narrative.
5. Surgical and medical robotics — teleoperated and semi-autonomous systems for soft-tissue, orthopaedic and endoluminal procedures. Gated by regulatory clearance, sold on clinical outcomes, monetised through instruments and service.
6. Agricultural robotics and field autonomy — autonomous tractors, orchestration platforms, laser weeding and harvesting. Structured environments, acute seasonal labour scarcity, no public-road regulatory burden.
7. Construction robotics — bricklaying, layout, rebar, excavation. Repeatedly defeated by site variability; now experimenting with outcome pricing rather than machine sales.
8. Robot foundation models and embodied AI — cross-task, cross-embodiment manipulation policies. The most-funded and least-measurable subcategory in the sector.
9. Simulation, digital twins and orchestration software — where robot engineering work is actually being changed by AI compute today, as distinct from where the marketing says it is.
10. Precision components — strain wave and cycloidal reducers, frameless torque motors, harmonic drives, force-torque and tactile sensors. A Japanese-led near-duopoly in reduction gearing and the sector's real chokepoint.
11. Permanent magnets and motor materials — NdFeB magnets and the rare-earth supply behind them. Not a robotics industry, but the input that can stop one.

3. Market structure

Concentration: a stratified oligopoly that is fragmenting from below. Industrial robots have been dominated for thirty years by four groups — FANUC, Yaskawa, ABB and KUKA — whose combined installed base NVIDIA puts at over 2 million units. That structure is now being eroded in the one market that matters for volume. Chinese domestic suppliers took 57% of installations in China in 2024, up from 30% in 2020 and 47% in 2023 (IFR, 2026-05-05 and 2024-09-24), reaching roughly 85% of the domestic metal and machinery segment and 59% of electronics. Because China is 54% of global installations, this is not a regional story.

Surgical robotics is the opposite: a two-decade single-vendor market only now becoming contestable. Warehouse automation is genuinely fragmented across fixed-infrastructure vendors, AMR fleets and integrators. Humanoids are unconcentrated, over-capitalised and pre-revenue. Precision reducers are the most concentrated layer of all — a Japanese duopoly of Harmonic Drive Systems and Nabtesco — and it is the layer with the most pricing power per dollar of revenue.

Where margin actually sits. Not in the arm. Robot hardware is a competitive, increasingly Chinese-priced business. Margin sits in four places: (1) precision components, where Harmonic Drive and Nabtesco have decades of process know-how and no volume substitute; (2) instruments and consumables, which is the entire reason Intuitive Surgical earns what it does on roughly $10.06bn of 2025 revenue; (3) integration and application engineering, historically the integrator's margin and the thing robot foundation models are aimed squarely at destroying; and (4) service, spares and uptime contracts on a 4.66-million-unit installed base. The lesson of the last decade is that whoever sells the razor loses to whoever sells the blade.

Barriers to entry. Extremely high in surgical (clinical evidence, FDA clearance, surgeon training, hospital capital cycles) and in precision reducers (metallurgy and process yield that do not transfer from a datasheet). Moderate in industrial arms — Chinese entrants have demonstrated that. Low in humanoids, which is precisely why there are so many of them and why almost none has revenue. Safety certification under ISO 10218 and ISO/TS 15066 is a real and underrated barrier everywhere, and it is the specific thing that no humanoid has yet cleared for free-roaming work alongside people.

Who has pricing power. Harmonic Drive and Nabtesco, at the component layer. Intuitive, still, at the procedure layer, though its own guidance of 13-15% procedure growth for 2026 against about 18% in 2025 is the first quantitative sign that it expects less. Symbotic, situationally, where a customer has already committed to its architecture. Nobody in industrial arms. Nobody at all in humanoids.

Market size. Handle with care. This research deliberately declines to quote a headline market-size number, because searches for robotics market sizing return almost exclusively undated, unauthored content-farm pages citing unsourced CAGRs — see §13. What can be stated as fact with a named source and date: 542,000 industrial robot installations in 2024 and an operational stock of 4,664,000 units (IFR World Robotics 2025, published 2025-09-25), and 38,000 US installations in 2025, up 11% (IFR, 2026-06-18). IFR's forecast — and it is labelled a forecast — is 575,000 units for 2025 and above 700,000 by 2028.

4. Who matters

Leading companies. FANUC (https://www.fanuc.co.jp) · Yaskawa Electric (https://www.yaskawa-global.com) · ABB Robotics (https://new.abb.com/products/robotics) · KUKA, owned by Midea (https://www.kuka.com) · Kawasaki Heavy Industries · Mitsubishi Electric · Denso · Universal Robots, part of Teradyne (https://www.universal-robots.com) · Intuitive Surgical (https://www.intuitive.com) · Medtronic (https://www.medtronic.com) · Johnson & Johnson MedTech (https://www.jnjmedtech.com) · Stryker · Symbotic (https://www.symbotic.com) · KION Group/Dematic (https://www.kiongroup.com) · Daifuku · AutoStore · Deere & Company (https://www.deere.com) · NVIDIA (https://www.nvidia.com) · Harmonic Drive Systems (https://www.hds.co.jp/english/) · Nabtesco (https://www.nabtesco.com) · Schaeffler (https://www.schaeffler.com) · Allegro MicroSystems (https://www.allegromicro.com) · Estun Automation · Siasun · Inovance · Unitree Robotics (https://www.unitree.com) · Boston Dynamics (https://bostondynamics.com) · Tesla (https://www.tesla.com).

Notable startups. Agility Robotics (https://www.agilityrobotics.com) · Figure AI (https://www.figure.ai) · Apptronik · 1X Technologies (https://www.1x.tech) · AGIBOT/Zhiyuan · UBTech · XPeng Dogotix · Neura Robotics · Skild AI (https://www.skild.ai) · Physical Intelligence (https://www.physicalintelligence.company) · Generalist AI · FieldAI · Locus Robotics (https://locusrobotics.com) · Verity (https://verity.net) · CMR Surgical (https://cmrsurgical.com) · Cornerstone Robotics · Autonomous Solutions Inc. (https://asirobots.com) · Carbon Robotics · Monarch Tractor · Monumental (https://www.monumental.co) · UltraSense · Loomia · Aescape.

Active investors. Churchill Capital Corp XI (https://www.churchillcapitalcorp.com), the SPAC vehicle for Agility Robotics. Midea Group, as the strategic owner of KUKA. Hyundai Motor Group, as owner of Boston Dynamics. Teradyne, as owner of Universal Robots and MiR. Chinese state-guided industrial funds, which are the dominant but least transparent capital source in the largest market. Note the macro context: over 70% of Q2 2026 venture capital went to AI-focused companies, and robotics funding totals are badly distorted by a handful of humanoid rounds.

Platforms and standards bodies. ISO/TC 299 Robotics (https://www.iso.org/committee/5915511.html), owner of ISO 10218-1/-2 and ISO/TS 15066 · IEC, for 62443 industrial cybersecurity · ROS 2 / Open Robotics, the de facto middleware · NVIDIA Isaac and Omniverse, an emergent de facto simulation platform · UL and TÜV as certification bodies.

Regulators. US FDA CDRH (https://www.fda.gov/medical-devices) — the pacing regulator for surgical robotics · OSHA and the ANSI/RIA R15.06 regime in the US · the EU Machinery Regulation and the AI Act, whose transparency obligations took effect 2026-08-02 while high-risk employment-context obligations slipped to 2027-12-02 · China's MIIT (https://www.miit.gov.cn) · the Shanghai Stock Exchange STAR Market (https://english.sse.com.cn), which has become a de facto gatekeeper on humanoid capital · China's MOFCOM, through rare-earth export licensing.

Research institutions. ARM Institute, the DoD-funded Manufacturing USA robotics institute (https://arminstitute.org) · Carnegie Mellon Robotics Institute · Fraunhofer IPA · AIST (Japan) · KIST and KIRO (Korea) · the Chinese Academy of Sciences' Shenyang Institute of Automation, the origin of Siasun. (The last five are named as sector landmarks; their 2026 activity was not verified in this research.)

Trade organisations. International Federation of Robotics (https://ifr.org) — the sector's statistical backbone · A3, Association for Advancing Automation (https://www.automate.org) · Japan Robot Association (https://www.jara.jp/e/) · China Robot Industry Alliance · MHI, for material handling.

Consumer and civil-society groups. Thin, and that is itself notable. There is no significant organised consumer constituency in robotics because the sector is overwhelmingly B2B; the closest analogues are organised labour (the IAM, UAW and Teamsters on automation clauses in contracts), patient-safety organisations in surgical robotics, and academic AI-ethics groups working on labour displacement. A sector that will reshape employment has almost no civil-society counterparty, which is a genuine structural gap rather than an oversight by this research.

5. Products, business models, technologies, customers

Major products. Articulated arms by payload class; cobots; goods-to-person storage systems and AMR fleets; surgical platforms plus their instruments; autonomous tractors and retrofit autonomy kits; humanoid platforms; simulation and orchestration software; and the component stack of reducers, actuators, encoders and sensors.

How money is made today. Capital equipment sale plus integration services, with recurring revenue from service contracts, spares and — in surgical — consumable instruments. Intuitive's model is the sector's most profitable precisely because the instrument stream is larger and stickier than the system sale.

How that is changing. Three shifts, in descending order of how well evidenced they are.

1. Opex substitution for capex. With the federal funds target at 3.50-3.75% and three FOMC members dissenting in favour of a hike, long-payback automation no longer clears a capital budget easily. A 2026 intralogistics survey found 29% of operators funding robotics through RaaS against 36% pure capex. Practitioners are blunt that this is a financing and service business, not a pricing wrapper — it moves the cost of capital and the reliability risk onto the vendor's balance sheet, which most robotics startups cannot fund. Construction robotics is reaching for the same structure from a different direction, with Monumental selling completed walls rather than robots.
2. Software and model licensing. NVIDIA now licenses Isaac GR00T N1.7 commercially. Universal Robots' Gen 7 ships an AI-ready tool flange with integrated vision and force-torque sensing, ROS 2 support and open APIs, explicitly to host third-party models. The intent is to convert integration labour into licensed software. Whether buyers will pay for that is not yet evidenced by any disclosed revenue.
3. Outcome pricing. Pay-per-pick, walls-as-a-service, per-acre. Early, and concentrated where the output is easy to count.

Technologies that matter. Precision reduction gearing, which nothing has displaced in forty years. Force-torque and tactile sensing, now emerging as the constraint on dexterity rather than model quality — UltraSense is pursuing subsurface ultrasound for durability, and Unichem bought Loomia to enter robot skin. Power architecture: moving from 12V to 48V cuts current fourfold and wiring losses sixteenfold but introduces deceleration transients the motor-control electronics must survive. Simulation and sim-to-real transfer. Manipulation foundation models. Permanent-magnet servo motors, and therefore NdFeB supply.

Customer segments and what they buy on. Automotive OEMs buy on cycle time, line-level integration and supplier continuity — and are no longer the growth engine. General manufacturers buy on payback period and installed-base compatibility. Warehouse and 3PL operators buy on labour cost and labour availability, not capability. Hospitals buy on clinical evidence, surgeon preference and total cost including instruments, and buy in long capital cycles that make competitive displacement slow even after clearance. Large corporate farms buy on labour scarcity and operational window, not on cost per acre. Research labs, universities and entertainment buyers are the actual customer base for humanoids today — the fact that explains Unitree's shipment numbers and its revenue mix simultaneously.

6. Geography

Production concentrates in East Asia and Germany. Japan is the largest robot-producing nation and the home of FANUC, Yaskawa, Kawasaki, Denso, Harmonic Drive and Nabtesco. Germany hosts KUKA, Schaeffler and a deep integrator base. China now produces the majority of the robots it installs. Switzerland and Sweden anchor ABB. Denmark is the cobot cluster around Universal Robots in Odense.

Demand concentrates overwhelmingly in Asia. In 2024, 74% of installations were in Asia, 16% in Europe and 9% in the Americas. China alone installed 295,000 units. IFR's preliminary read is that China's 2025 installations were roughly ten times the US figure of 38,000.

Density tells a different story from volume. On 2024 data: Korea 1,220 robots per 10,000 manufacturing employees, Singapore 818, Germany 449, Japan 446, Sweden 377, the US 307, Chinese Taipei 302, and China just 166 against a global average of 132. Western Europe hit a record 267. China installs more than half the world's robots and remains far below the OECD leaders on density, because its manufacturing workforce is so large — which is the strongest available argument that Chinese volume has years left to run.

Capital concentrates in the US and China. US capital dominates robot foundation models and the highest private humanoid valuations; Chinese capital dominates humanoid production volume and now has a public-market venue for it.

Regulation concentrates differently again. The FDA sets the surgical robotics clock globally, because US clearance is the commercial gate. The EU sets the machinery-safety and AI-transparency regime. China sets the component supply through export licensing, which is the most powerful single regulatory instrument touching this sector.

Non-US market in detail — China, with a regional source. The Shanghai Stock Exchange's own disclosure around Unitree's listing (english.sse.com.cn, 2026-08-10) is the single most informative primary document located in this research. It establishes: IPO proceeds of about RMB 6.10bn from 40.45m shares at RMB 150.80, representing 10% of post-offering capital, with offline subscriptions oversubscribed 2,760.67 times; 2025 revenue of RMB 1.70bn against RMB 392.77m in 2024 and RMB 159.13m in 2023; statutory net profit of RMB 278.21m and adjusted net profit of RMB 590.75m; overseas revenue of RMB 731.66m, 43.65% of main-business revenue; and cumulative 2023-2025 shipments of 33,294 quadrupeds and 5,632 humanoids, of which over 5,500 humanoids shipped in 2025 alone. A 73-day review from application acceptance on 2026-03-20 to listing-committee approval on 2026-06-01 was described as a record pace. Separately, IFR's reading of China's 15th Five-Year Plan (2026-2030) is that AI-powered robotics is now a core national objective with thousands of subordinate sectoral and regional plans mandated to align — while noting that the plan places humanoid commercialisation toward the plan's conclusion, not its start.

Japan's specific and underpriced exposure. Japan is the second-largest installer (44,500 units in 2024), has a density of 446, and hosts the world's dominant precision-reducer suppliers. China controls roughly 80% of global NdFeB permanent-magnet output, the input for the servo motors in essentially every robot. Chinese magnet exports hit records in H1 2026 while being allocated politically, and Japan received zero covered rare-earth exports in July 2026, with South Korea becoming the largest destination. A licensing decision in Beijing can constrain FANUC, Yaskawa, Harmonic Drive and Nabtesco output without touching a single robot tariff — and it lands precisely as those firms are already losing Chinese domestic share.

7. Historical trend patterns

Robotics has a long, well-documented record of premature enthusiasm, and knowing it is the main defence against 2026's.

The 1980s Japanese factory-automation wave. "Lights-out manufacturing" was declared imminent. It arrived in narrow, structured, high-volume processes — machining cells, semiconductor fabs — and nowhere else. The pattern established then still holds: robots win where the environment can be engineered to suit them, and lose where they must adapt to an environment engineered for people.

Service robots, roughly 2000-2010. Honda's ASIMO (2000) was the canonical humanoid demonstration, walking, climbing stairs and conducting an orchestra for well over a decade of public appearances. It never performed commercial work. Honda retired the programme. ASIMO is the single most instructive precedent for 2026: a technically genuine, beautifully demonstrated humanoid, continuously improved for eighteen years, with a lifetime commercial deployment of zero.

Rethink Robotics, 2008-2018. Baxter and Sawyer created the cobot category's vocabulary — safe, cheap, no fencing, trainable by demonstration — raised roughly $150m, and shut down in 2018. The category it invented then grew substantially, under Universal Robots. The lesson is that being right about the category and wrong about the timing or the cost curve is fatal to the pioneer and profitable for the follower.

The 2014-2016 "robot apocalypse" employment panic. Widely forecast mass displacement did not materialise on the predicted timetable; robot density rose while manufacturing employment in several high-density economies also rose. The displacement story was directionally real and wildly wrong on pace.

Warehouse automation, 2012-2022. Amazon's Kiva acquisition (2012) was the genuine inflection — this is the one that worked. It ended in a classic overbuild: pandemic demand was extrapolated, capacity was committed, and 2022-23 brought a sharp digestion phase. The segment has now re-accelerated, which is what a real trend looks like after a real correction.

Surgical robotics, 2000-2026. Intuitive is the sector's one unambiguous durable success: a two-decade monopoly built on clinical evidence and instrument lock-in. It is also a cautionary tale about forecasting competition — challengers have been "two years away" repeatedly since the mid-2010s. The 2026 difference is that the clearances now exist on the record.

Agricultural autonomy, 2016-2026. Slow, unglamorous, under-covered, and steadily working. It has never had a hype wave, which may be why it has never had a crash.

The specific false positives this sector has produced. ASIMO and humanoid service robots. Rethink's cobots. "Lights-out" factories. Foxconn's 2011 announcement of one million robots. Mass-market consumer home robots beyond the vacuum. Exoskeletons for industrial work. Delivery sidewalk robots at scale. Every one of these was supported at the time by impressive video and credible engineers. None was ever supported by disclosed unit economics, which is the discriminator this dossier applies to humanoids in §10.

8. What is changing now

As of 2026-09-15, five things are genuinely different from a year ago.

Humanoid robotics acquired a disclosure regime, and the disclosures are bad. Until August 2026 the category had no audited numbers. Then Unitree listed on the STAR Market and filed a prospectus; Agility Robotics filed a Form S-4 with the SEC on 2026-09-04 to merge with Churchill Capital Corp XI. The resulting facts are the most important in this dossier and are set out in §10. Chinese regulators have since reportedly tightened informal listing criteria for humanoid candidates, requiring recurring revenue, progress toward profitability, or significant technological innovation — the capital tap is being narrowed by the same authority that opened it.

The industrial robot market's centre of gravity completed its move. Chinese domestic vendors at 57% of their home market, China at 54% of global installations, and a global unit market flat near 542,000 for a fourth year while Europe fell 8% and every major European market declined. The growth is China's; the ex-China market is contracting.

The US turned, and not because of cars. 38,000 installations in 2025, up 11%, with automotive flat at 13,500 and food up about 30%. The macro backdrop reinforces the automotive part: the EPA rescinded vehicle GHG standards effective 2026-04-20, the US EV tax credit expired, and 2026 is the first year of declining annual US BEV sales — the retooling wave that drove robot demand from 2018 is over.

AI compute reached robotics — through the toolchain, not through autonomy. NVIDIA's March 2026 releases (Cosmos 3, Isaac GR00T N1.7 commercially licensed, GR00T N2 preview, Isaac Lab 3.0, Newton 1.0) landed with ABB, FANUC, Yaskawa and KUKA integrating simulation across a 2-million-robot installed base. That part is real and is changing engineering practice. The autonomous-generalist-manipulation part is claimed, benchmarked by the vendor, and not yet visible in any disclosed deployment metric.

The component layer became the interesting one. Schaeffler committing to mass-produce strain wave gearboxes for humanoids from 2027; Harmonic Drive moving to the TSE Prime Market; acquisitions in tactile skin; a serious technical argument that humanoid capability is gated by packaging, power architecture and per-joint sensing rather than by models. Meanwhile China's rare-earth export licensing — with Japan cut to zero covered exports in July 2026 — sits directly upstream of every servo motor in the sector.

Against the macro brief: this is a rising-rate, sticky-inflation environment in which capital cost is a live constraint, which is why the RaaS shift is a financing event rather than a marketing one. The exit window reopened violently, which is why two humanoid companies reached public markets in a single quarter. And per sector 11's correction, aggregate capital-markets euphoria has not produced cash — a caution that applies with unusual force to a category whose leading US player books $1.8m of revenue.

9. The five lists

Five most important current trends

1. China's domestic robot makers displacing foreign incumbents in the world's largest market (T-13-01).
2. Global installations plateaued near 542,000 while China absorbs all the growth (T-13-02).
3. Warehouse automation re-accelerating after the 2021-22 overbuild (T-13-04).
4. Intuitive's surgical monopoly meeting credible multi-vendor competition (T-13-05).
5. Rare-earth export licensing as a direct constraint on Japanese robot makers (T-13-15).

Five fastest-growing signals

1. Humanoid companies reaching public markets via STAR listings and SPACs (T-13-11) — and the regulatory pushback against it.
2. Robot foundation models entering narrow, supervised commercial pilots (T-13-09).
3. The humanoid component supply chain industrialising ahead of humanoid demand (T-13-10).
4. Agricultural autonomy reaching fleet scale (T-13-12).
5. Tactile sensing emerging as the binding constraint on dexterity (T-13-14).

Five trends most likely to affect businesses

1. Robotics-as-a-service as the financing mechanism under high rates (T-13-06) — this determines whether a buyer can automate at all.
2. Warehouse automation re-acceleration (T-13-04).
3. Automotive ceasing to be the robot demand engine (T-13-17) — a supplier-mix problem for every integrator.
4. China's domestic displacement (T-13-01) — a revenue problem for every incumbent vendor.
5. Rare-earth licensing (T-13-15) — a production-continuity problem with no vendor-side mitigation.

Five trends most likely to affect consumers

1. Surgical robotics competition (T-13-05) — the only trend here that touches ordinary people directly and materially, through procedure availability and cost. Overlaps with the business list.
2. Agricultural autonomy (T-13-12), through food production cost and farm labour.
3. Warehouse automation (T-13-04), through delivery cost and warehouse working conditions. Overlaps with the business list.
4. Humanoid robots as a narrative (T-13-19) — which shapes public expectations, policy debate and labour anxiety far more than any deployed unit does.
5. Construction robotics (T-13-13), through housing cost, if it works.

Where these overlap, and why. Warehouse automation and surgical robotics appear on both the business and consumer lists because they are the two segments with real deployed scale and a direct end-user consequence. Humanoids appear on the consumer list purely as a narrative force, which is an honest description of their current effect: they change what people expect and what politicians discuss, and they do almost no work. The fastest-growing list and the most-important list share nothing, which is the central finding of this dossier — the fastest-moving things in this sector are not yet the most important ones.

10. Overhyped / overlooked / cooling / reversing

Most overhyped: humanoid robots doing useful paid work at scale

The evidence that hype outruns substance, entirely from prospectuses, SEC filings and the sector's own trade body:

• Unitree, the global volume leader, shipped 5,632 humanoids cumulatively 2023-2025, over 5,500 of them in 2025. Humanoid revenue was RMB 868m, 51.78% of total revenue. And less than 10% of its 2025 revenue came from industrial applications. The G1 sells at $13,500 — a research, education and entertainment price point, not a capital-equipment one. The world's largest humanoid maker is, commercially, a scientific-instrument and entertainment company.
• Agility Robotics, the leading US deployment story, disclosed $1.8m of 2025 revenue against a $140m operating loss in its S-4. That is the most-cited humanoid logistics deployment in the world, and it generates less revenue than a single mid-sized systems integrator.
• Tesla had zero Optimus robots performing useful work as of January 2026, against a promise of 10,000 units in 2025. Musk declined to give a 2026 target, calling output "quite slow" and "impossible to predict."
• IFR itself — the robot manufacturers' own trade association, with every incentive to talk the category up — states that mass adoption as universal factory helpers or in households "will not happen within the near- and medium-term future."
• HSBC, pre-IPO, warned that "the surge in shipments for robot makers could be illusionary" and that the shipment upcycle was unlikely to be sustained over one to two years without AI improvements.

Against that: Figure AI carried a $39bn valuation as of September 2025, Unitree peaked near $66bn, and XPeng's Dogotix raised $900m at $6.3bn. The gap between aggregate category valuation and verifiable industrial revenue is the largest attention-to-substance gap encountered in this research.

The crucial definitional point. Search results for humanoid deployment counts are dominated by "deployment trackers" that count announced pilots, letters of intent, demonstration units and units produced as "deployments." The filings count revenue. Almost the entire public disagreement about humanoid progress is this definitional gap, not a factual dispute. Treat every humanoid video as marketing unless it comes with an independently verified paying deployment — and as of 2026-09-15, essentially none does.

Secondary overhype: "generalist manipulation is solved." The claims in circulation — top benchmark placement, new-task success "more than twice as often" as competitors, learning from a single demonstration — are vendor-stated and were not independently replicated here. No disclosed production deployment reports task success rate, cycle time or MTBF for a foundation-model-driven robot. The falsifier is simple: global installations are flat at ~542,000 and integration is still the dominant cost. If generalist manipulation were production-ready, those numbers would already be moving.

Most overlooked

1. Rare-earth export licensing as a robotics story. It is covered intensively as a semiconductor and defence story and almost never as a robot story, yet NdFeB magnets are ~80% Chinese and sit in every servo motor. Japan — home to the world's dominant precision-reducer and a large share of its robot production — received zero covered rare-earth exports in July 2026. Attention has missed it because it requires joining a trade-policy dataset to a robotics supply chain, and the two literatures do not overlap.
2. Precision reducers as the real chokepoint. A Japanese duopoly with process know-how that does not transfer from a datasheet, sitting upstream of every robot and every humanoid. Harmonic Drive's consolidated bookings are a better leading indicator of genuine humanoid volume than any humanoid company's announcements — and almost nobody watches them.
3. Agricultural and off-road autonomy. It works, it is deploying at fleet scale, and it attracts a fraction of humanoid attention. Missed because it is unglamorous, happens on private land in rural areas, and has no consumer-facing story.
4. Warehouse automation's audited revenue. Symbotic booked $720.8m in a single quarter, up 21.7%, against the entire humanoid category's low-single-digit-millions of verifiable industrial revenue — and receives a small fraction of the coverage.
5. The absence of a civil-society counterparty. A sector that will reshape employment has essentially no organised public-interest constituency. That absence shapes which regulations get written and which do not.

Cooling, with the indicator that turned

• Humanoid valuations. Unitree fell from a RMB 1,100 first-day high to RMB 513.93 on 2026-09-09 — down 53% from the high and 39% from the first-day close, erasing roughly $35bn from peak valuation, while nothing changed operationally. The turned indicator is a liquid public price, which private humanoid marks must now be tested against. Caveat honestly: it still trades at three times its offer price, so this is deflation from euphoria, not collapse.
• Automotive as the robot demand engine. US automotive installations were 13,500 units, 1% down, while total US installations rose 11%. The turned indicator is that composition split, reinforced by the EV tax credit expiry and the 2026-04-20 GHG standards rescission.
• European industrial robot demand. Europe -8% to about 85,000 units, with Germany -5%, Italy -16%, France -24% and the UK -35%. The turned indicator is that the 2023 European high of 92,393 units proved to be a peak, not a trend.

Trends that may reverse, and the mechanism

• Chinese domestic displacement could stall if the domestic price war makes the won share unprofitable and consolidation removes the marginal competitors. Mechanism: margin, not technology.
• The humanoid deflation could reverse violently upward if a single credible, audited, multi-hundred-unit paying industrial deployment is disclosed. The category is not disproven; it is undisclosed. One real contract with real numbers would re-rate everything.
• The US recovery could reverse on capital cost. One year of growth off a two-year trough is not a trend, and the sector's payback calculations are directly exposed to a Fed that had three members dissenting in favour of a hike in July 2026.
• Rare-earth constraint could reverse in either direction — allocation is discretionary and can be loosened as abruptly as it was tightened.

11. Risks and major uncertainties

Sector-specific risks. Component single-sourcing, particularly NdFeB magnets and precision reducers, where no short-run substitute exists. Safety-certification exposure: a serious injury involving a collaborative or mobile robot working near people would tighten the regime across every segment simultaneously. Customer concentration, acutely at Symbotic. Capital cost, which determines whether long-payback automation clears at all and which now sits on vendors' balance sheets via RaaS. Chinese price competition compressing the hardware layer to commodity margins. Humanoid capital destruction, which is now large enough to damage sentiment for adjacent, better-founded robotics businesses. Regulatory-deadline slippage of the kind the macro brief documents for the EU AI Act, which makes compliance-driven demand forecasts unreliable.

Genuine unknowns — "we don't know but could find out." The actual global installed base of humanoids in paying industrial work: Figure, Apptronik, 1X, AGIBOT and UBTech disclose nothing, so the number is unknown rather than known to be small, and their eventual filings would settle it. Symbotic's precise Walmart concentration, which is in the 10-K this research could not parse. FANUC's and Yaskawa's 2026 order intake, which exists and was not retrievable. Japanese and Chinese production statistics from JARA and CRIA, which are published but were not accessible within the search budget. Whether Census BTOS has a robotics-use supplement in its microdata. Task success rates and MTBF for foundation-model-driven robots, which the vendors have and do not publish.

Genuine unknowns — "nobody can know." Whether manipulation generalisation follows a scaling curve like language did, or plateaus on contact-rich physical tasks. Whether humanoid form factor is ever economically justified against wheeled or fixed-base alternatives for real work, as opposed to being justified by the availability of human-shaped environments. Whether Chinese rare-earth licensing is a durable instrument of statecraft or a negotiating position. Where safety regulation lands for non-deterministic learned behaviour, since no jurisdiction has attempted it. And the pace question that has defeated every prior robotics forecast: the direction of these trends is mostly knowable; the timing has never once been.

12. Scenarios to 2030

Base — "Asia compounds, humanoids disappoint quietly." Global installations resume slow growth toward IFR's 700,000 by 2028, driven almost entirely by China, India and Southeast Asia. Chinese domestic share passes 65%. Incumbents retreat to high-precision, safety-critical and service revenue. Warehouse automation compounds at double digits. Surgical becomes a genuine three-way market by 2029. Humanoids settle into research, education, entertainment and a handful of real logistics niches at low-thousands of annual units, with most of today's entrants absorbed or wound down. Falsifiable early indicator: IFR's 2027 and 2028 releases show global installations above 600,000 with China above 60% of the total, while no humanoid company discloses annual industrial revenue above $100m.

Upside — "the integration cost collapses." Foundation models genuinely cut deployment engineering, and the addressable set of tasks expands to short-run and high-mix work that has never been automatable. Installations break decisively above IFR's forecast; cobot volumes grow faster than industrial arms; integrator margin migrates to software licences. Falsifiable early indicator: a robot OEM or model vendor discloses booked, recurring software revenue above $250m attributable to manipulation models, and global installations exceed 750,000 before 2029.

Downside — "capital cost bites and China exports deflation." Rates stay high or rise, RaaS providers discover they have become undercapitalised leasing companies, and Chinese overcapacity exports price deflation into every market. Unit volumes grow while sector revenue stagnates. Several well-known Western robotics names fail or are absorbed. Falsifiable early indicator: two or more mid-size Western robotics vendors restructure or are acquired below their last private mark within four quarters, while IFR unit growth remains positive — the volume-up, value-down signature.

Disruption — "a real humanoid deployment." One company discloses an audited, multi-hundred-unit deployment doing paid work with disclosed unit economics that pencil. The entire capital structure of the sector re-rates upward, component capacity that looked speculative becomes scarce, and the incumbents' installed-base advantage is suddenly worth much less. Falsifiable early indicator: an SEC filing or exchange prospectus showing humanoid revenue above $50m from industrial customers with a stated unit count and repeat orders from the same customer. Nothing resembling this exists as of 2026-09-15.

Regulatory — "the chokepoint is used." Chinese rare-earth licensing is tightened against Japan and Europe rather than merely allocated, and robot production is constrained physically. Alternatively, a serious safety incident triggers an ISO/OSHA tightening that pushes mobile and collaborative deployment back years. Either path makes regulation, not technology, the sector's binding variable. Falsifiable early indicator: Japanese robot makers disclose magnet-related production constraints in quarterly results, or ISO/TC 299 opens a new work item on mobile manipulator or learned-behaviour safety.

Failure — "the second robot winter." Humanoid capital destruction is large and public enough to poison sentiment for robotics generally. Funding dries up for well-founded businesses alongside speculative ones, exactly as the post-2000 collapse did to legitimate internet infrastructure. Installations keep growing in China and the Western venture ecosystem for robotics largely empties out. Falsifiable early indicator: two consecutive quarters in which disclosed global robotics venture funding excluding humanoids falls more than 40% year over year, combined with at least one high-profile humanoid insolvency or fire-sale.

13. Data gaps and limitations

1. The shared search budget was exhausted mid-research (200/200 WebSearch calls across six agents). Roughly eleven searches were available to this sector against a planned twenty-two. Per the macro brief's standing instruction, research continued via WebFetch on primary sources — which is how IFR's five 2026 press releases, the Shanghai Stock Exchange's Unitree disclosure, SEC EDGAR full-text search and the SEC XBRL API were obtained. Discovery capability was nonetheless materially reduced: topics reached only through targeted search — construction robotics beyond one company, exoskeletons, Korean and Indian vendors, ABB's reported robotics separation, cobot market sizing — are thinner than they should be. This sector should be prioritised for re-run with a full budget.
2. IFR World Robotics is paywalled and this is the sector's central licensing need. The free press releases give headline installations, stock and density; they withhold end-use and application breakdowns, vendor-level detail, the full national time series and the service-robot data entirely. The 2026 edition, covering 2025 data, publishes around late September 2026 — days after this research date, so the most current full-year data in this dossier is 2024 with partial 2025 national figures. Phase 2 must budget for this licence; there is no substitute.
3. No population-level measure of robot adoption exists for the US. Census BTOS publishes an AI-use series but no robotics-use series was located, so IFR's unit data cannot be cross-checked against firm-level adoption. This is a structural blind spot for the whole sector, not a retrieval failure.
4. Japanese and Chinese primary statistics were not retrieved. JARA production data and CRIA/MIIT figures are published but were not accessible within the reduced budget, and most are Chinese- or Japanese-language only. China is 54% of global installations and reportedly 97% of humanoid shipments, and it is being read here substantially through a German trade body's English summary. This is the largest non-English gap. (The non-English-origin requirement is met by the Shanghai Stock Exchange source, which carries Unitree's audited prospectus figures.)
5. Company IR pages failed repeatedly. FANUC's IR library and finance pages returned 404; Yaskawa's, Harmonic Drive's and Symbotic's IR sites rendered only as link indexes; Intuitive's news-release index returned 404 and its quarterly figures had to be taken from trade press; ABB's media pages did not surface the robotics-separation news and that item is recorded as unverified. Symbotic's revenue was ultimately obtained cleanly from the SEC XBRL API, which is the recommended route. CNBC returned 403 to fetching, so the reported +542% Unitree first-day figure is cited from search metadata only.
6. Agility Robotics' financial figures are Tier B reporting a Tier A document. EDGAR confirms the S-4 exists and was filed 2026-09-04 (CIKs 0002074973 and 0001727116); the $1.8m revenue and $140m operating loss figures come from The Robot Report's reading of it. The S-4 itself was not parsed — EDGAR's forms filter did not work on the full-text endpoint and browse-edgar is blocked by robots.txt. Phase 2 should pull the S-4 directly.
7. Private humanoid companies disclose nothing. Figure, Apptronik, 1X, AGIBOT, UBTech and Neura publish no shipment or revenue figures. The true global humanoid installed base is therefore unknown, not known to be small. This dossier's position is that the burden of proof sits with the claimants, but the honest statement is that the number cannot be verified in either direction.
8. Content farms dominate this sector's search results to an unusual degree. Queries for humanoid deployment counts returned pages almost exclusively from undated, unauthored "tracker" and "market report" sites citing unsourced figures. None of them is used anywhere in this dossier. This is the same pattern sector 09 recorded, and it is worse here, because humanoids are the most SEO-contested robotics topic.
9. Several 2026 items rest on a single trade-publication index page, including Schaeffler's gearbox plans, the Unichem/Loomia acquisition, XPeng Dogotix's round, Medtronic's reported $700m Cornerstone Robotics investment and the ARM Institute's $90m awards. All are marked low confidence or unverified in the entity and trend records.
10. Conflicting figures recorded rather than resolved. Unitree's first-day gain (CNBC +542% versus Fortune +460%, an intraday-peak versus closing-price difference); Unitree's 2025 profit (RMB 278.21m statutory versus RMB 590.75m adjusted, with Fortune reporting ~RMB 600m unlabelled); and China's share of global installations (51% on 2023 data, 54% on 2024). All are in the contradictions arrays.
11. Agricultural robotics evidence is vendor-authored. The ASI/U.S. Sugar account is bylined by ASI's own CEO and is classified Tier C. No independent verification of fleet size, cost per acre or uptime was located, and the trend is scored accordingly (evidence_quality 1).

14. Ranking scorecard

15. Sources

1. "Record of 4 Million Robots in Factories Worldwide" (World Robotics 2024), International Federation of Robotics, https://ifr.org/ifr-press-releases/news/record-of-4-million-robots-working-in-factories-worldwide, 2024-09-24, Tier A
2. "Global Robot Demand in Factories Doubles over 10 Years" (World Robotics 2025), International Federation of Robotics, https://ifr.org/ifr-press-releases/news/global-robot-demand-in-factories-doubles-over-10-years, 2025-09-25, Tier A
3. "US Robot Industry Returns to Double Digit Growth", International Federation of Robotics, https://ifr.org/ifr-press-releases/news/us-robot-industry-returns-to-double-digit-growth, 2026-06-18, Tier A
4. "Robot Density Surges in Europe, Asia, and Americas", International Federation of Robotics, https://ifr.org/ifr-press-releases/news/robot-density-surges-in-europe-asia-and-americas, 2026-04-08, Tier A
5. "China Makes AI-powered Robots Core of National Strategy", International Federation of Robotics, https://ifr.org/ifr-press-releases/news/china-makes-ai-powered-robots-core-of-national-strategy, 2026-05-05, Tier A
6. "Jane Heffner is New President of International Federation of Robotics" and "IERA Award 2026 Goes to Flying Warehouse Robots by Verity from Switzerland", International Federation of Robotics, https://ifr.org/ifr-press-releases, 2026-07-02 and 2026-06-16, Tier A
7. "Unitree IPO draws spotlight to China's fast-growing humanoid robot sector", Shanghai Stock Exchange (carrying China Daily), https://english.sse.com.cn/news/newsrelease/voice/c/c_20260811_10828578.shtml, 2026-08-10, Tier A (non-English-origin regional source; carries Unitree's audited prospectus figures)
8. "Unitree shares down 53% from IPO debut", The Robot Report, https://www.therobotreport.com/unitree-shares-down-53-from-ipo-debut/, 2026-09-09, Tier B
9. "Unitree, famous for its dancing robots, surges by 460% on its trading debut", Fortune, https://fortune.com/2026/08/19/unitree-china-dancing-robots-ipo-trading-surge-valuation/, 2026-08-19, Tier B
10. Churchill Capital Corp XI / Agility Robotics Form S-4, DRS, DRS/A, 425 and 8-K filings, US Securities and Exchange Commission EDGAR full-text search, https://efts.sec.gov/LATEST/search-index?q=%22Agility+Robotics%22, S-4 filed 2026-09-04, Tier A
11. "Agility Robotics reports $1.8M revenue ahead of humanoid SPAC" and related humanoid category items (Schaeffler gearboxes 2026-08-19; Unichem/Loomia 2026-08-14; XPeng Dogotix 2026-08-21; UltraSense 2026-09-12), The Robot Report, https://www.therobotreport.com/category/humanoids/, 2026-09-07 to 2026-09-14, Tier B
12. "Tesla pushes Optimus V3 reveal later this year - again", Electrek (Fred Lambert), https://electrek.co/2026/04/22/tesla-optimus-production-fremont-model-sx-line/, 2026-04-23, Tier B
13. "NVIDIA and Global Robotics Leaders Take Physical AI to the Real World", NVIDIA, https://nvidianews.nvidia.com/news/nvidia-and-global-robotics-leaders-take-physical-ai-to-the-real-world, 2026-03-16, Tier A (primary but promotional; all claims from it labelled marketing)
14. "5 robotic surgery trends to watch in 2026", MedTech Dive (Susan Kelly), https://www.medtechdive.com/news/5-robotic-surgery-trends-to-watch-2026/810577/, 2026-01-27, Tier B
15. "Intuitive says da Vinci systems 'exceeded expectations,' it wants more", MassDevice, https://www.massdevice.com/intuitive-says-davinci-systems-exceeded-expectations-it-wants-more/, 2026-01-14, Tier B
16. Symbotic Inc. XBRL company facts, revenue concept (Form 10-Q, fiscal Q3 2026, CIK 0001837240), US Securities and Exchange Commission, https://data.sec.gov/api/xbrl/companyconcept/CIK0001837240/us-gaap/RevenueFromContractWithCustomerExcludingAssessedTax.json, filed 2026-08-05, Tier A
17. "Universal Robots launches its seventh generation robot platform at IMTS", The Robot Report, https://www.therobotreport.com/universal-robots-launches-its-seventh-generation-robot-platform-at-imts/, 2026-09-14, Tier B
18. "How autonomous fleets are changing what's possible in American agriculture", The Robot Report, bylined by Mel Torrie, CEO of ASI, https://www.therobotreport.com/how-autonomous-fleets-are-changing-whats-possible-american-agriculture/, 2026-09-13, Tier C (vendor-authored; treated as marketing)
19. "AI can't outrun a humanoid's hardware", The Robot Report, by Andreas P. Friedrich, Allegro MicroSystems, https://www.therobotreport.com/ai-cant-outrun-a-humanoids-hardware/, 2026-09-08, Tier B (supplier-affiliated author; technical content)
20. "Why RaaS needs more than a subscription model", The Robot Report, https://www.therobotreport.com/why-raas-needs-more-than-a-subscription-model/, 2026-09-14, Tier B
21. "One brick at a time: How Monumental uses robotics to build walls", The Robot Report, https://www.therobotreport.com/one-brick-at-a-time-how-monumental-uses-robotics-to-build-walls/, 2026-09-11, Tier B
22. Harmonic Drive Systems Inc. Investor Relations (TSE Prime Market reclassification, 2026-02-27), https://www.hds.co.jp/english/ir/, 2026-02-27, Tier A
23. "2026 Intralogistics Robotics Survey: Robotics moves into the mainstream", Modern Materials Handling / Peerless Research Group with MHI and The Robotics Group, https://www.mmh.com/article/2026_intralogistics_robotics_survey_robotics_moves_into_the_mainstream, 2026-06-01, Tier B (self-selected panel, n=166; carried forward from sector 09)
24. "Rare Earth Export Restrictions One Year Later", CSIS (Gracelin Baskaran, Meredith Schwartz), https://www.csis.org/analysis/rare-earth-export-restrictions-one-year-later, 2026-04-27, Tier B (carried forward from sector 09)
25. "China's Global Exports of Rare Earth Elements and Rare Earth Permanent Magnets", Silverado Policy Accelerator, https://silverado.org/data-dashboards/china-global-exports-rare-earth-elements-rare-earth-permanent-magnets/, 2026-08-01, Tier B (carried forward from sector 09)
26. Business Trends and Outlook Survey, US Census Bureau, https://www.census.gov/programs-surveys/btos.html, page revised 2026-08-12, Tier A (cited for the absence of a robotics-use series)
27. Macro Context Brief and addendum, Phase 1 research programme, internal, 2026-09-15 — FOMC rate and dissents, EU AI Act timetable, EPA GHG rescission, EV credit expiry, VC concentration and the DPI/cash correction from sector 11.

More research in this sector · Explore trend records