SASIGNAL ATLASCross-industry intelligence / Research desk
SIGNAL ATLAS / RESEARCH DESK

Humanoid robots doing useful paid work at scale: whether disclosed industrial revenue ever appears behind the demonstrations

Scenario set · undefined · to 2030

Six alternative branches. These are subjective probabilities from the seed, not observed frequencies or investment recommendations.

Why this trend

Load-bearing assumption

That filed financial disclosure — SEC filings and exchange prospectuses — rather than demonstration video, deployment trackers or announced pilots, is the evidence that settles whether humanoids perform paid industrial work.

Preconditions

Not recorded.

base · 44% · band

Humanoids disappoint quietly while Asia compounds. Global industrial 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%, and incumbents retreat to high-precision, safety-critical and service revenue. 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. The category keeps its valuation premium far longer than its revenue justifies, then loses it without a single decisive event.

Mechanism

The price point sets the market: Unitree's G1 at $13,500 is a scientific-instrument and entertainment price, not a capital-equipment one, and less than 10% of Unitree's 2025 revenue came from industrial applications. Integration remains the dominant cost in real deployments, so buyers keep specifying wheeled, fixed-base and purpose-built machines that are cheaper to integrate. The decelerating element is Chinese listing policy: regulators have informally raised the bar for humanoid IPO candidates to require recurring revenue or progress toward profitability, narrowing the capital tap that funds the category.

Preconditions, early indicators and assumptions
Early Indicators
  • Indicator

    IFR 2027 and 2028 releases showing global installations above 600,000 with China above 60% of the total, while no humanoid company discloses annual industrial revenue above $100m

    Source

    S-13-01 IFR press releases; S-13-03 SEC EDGAR

    Would Be Visible By

    2028-10

  • Indicator

    Unitree FY2026 humanoid revenue below RMB 3.0bn with industrial share still under 20%

    Source

    S-13-04 Shanghai Stock Exchange STAR Market disclosures

    Would Be Visible By

    2027-05

  • Indicator

    Harmonic Drive Systems consolidated bookings showing no step-change attributable to humanoid joints

    Source

    S-13-10 Harmonic Drive Systems IR

    Would Be Visible By

    2027-06

Affected Industries
  • 13
  • 09
  • 03
  • 11
What Businesses Should Do

Evaluate automation on integration cost and payback at the current policy rate, not on form factor; require a named paying deployment with disclosed unit economics before treating any humanoid vendor as a supplier rather than a research partner.

Precedence Note

Distinguished from disruption by one observable only: a filed document showing humanoid revenue above $50m from industrial customers with a unit count. Until such a filing exists, announcements, pilots and letters of intent leave the outcome here.

Would Change Our Mind

Harmonic Drive or Nabtesco reporting a bookings step-change explicitly attributed to humanoid programmes.

upside · 11% · band

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, collaborative-robot volumes grow faster than industrial arms, and integrator margin migrates to software licences. Humanoids benefit indirectly, as one form factor among several, rather than being the cause.

Mechanism

The named constraint that releases is engineering labour, not robot hardware. ABB, FANUC, Yaskawa and KUKA are already integrating simulation across an installed base NVIDIA puts at over two million robots, and Isaac Lab 3.0, Newton 1.0 and Cosmos 3 shipped in March 2026. If simulation and manipulation models cut commissioning time materially, the payback calculation clears at a 3.50-3.75% policy rate for a much larger set of tasks, and demand comes from existing manufacturers rather than from anywhere new.

Preconditions, early indicators and assumptions
Early Indicators
  • Indicator

    A robot OEM or model vendor disclosing booked recurring software revenue above $250m attributable to manipulation models, and global installations above 750,000

    Source

    S-13-03 SEC EDGAR segment disclosure; S-13-01 IFR press releases

    Would Be Visible By

    2029-10

  • Indicator

    Teradyne reporting accelerating rather than steady collaborative-robot growth for three consecutive quarters

    Source

    S-13-21 Teradyne investor relations

    Would Be Visible By

    2027-10

  • Indicator

    A disclosed production deployment publishing task success rate and mean time between failures for a foundation-model-driven robot

    Source

    S-13-05 The Robot Report; S-13-03 SEC EDGAR

    Would Be Visible By

    2028-06

Affected Industries
  • 13
  • 09
  • 01
  • 03
What Businesses Should Do

Pilot on high-mix, short-run cells where integration cost has always been the blocker, and structure contracts so the software licence rather than the machine carries the price; measure commissioning hours per cell as the leading internal metric.

Precedence Note

Requires disclosed software revenue at scale plus a unit-volume break. Benchmark rankings, partner logos and simulation adoption on their own stay in base.

Would Change Our Mind

Global installations staying below 600,000 for a third consecutive year while simulation adoption is described as universal.

downside · 19% · band

Capital cost bites and China exports deflation. Rates stay high or rise, robotics-as-a-service providers discover they have become undercapitalised leasing companies, and Chinese overcapacity pushes price deflation into every market. Unit volumes grow while sector revenue stagnates, and several well-known Western robotics names fail or are absorbed below their last private mark.

Mechanism

The parameter that deteriorates is the cost of capital on the vendor's balance sheet rather than the buyer's. A 2026 survey found 29% of intralogistics operators funding robotics through RaaS against 36% pure capex; under RaaS the vendor finances the asset, so a hawkish Fed transmits directly into vendor solvency. Chinese domestic suppliers at 57% of their home market and roughly 85% of metal and machinery there compress hardware margin to commodity levels, which is the volume-up, value-down signature.

Preconditions, early indicators and assumptions
Early Indicators
  • Indicator

    Two or more mid-size Western robotics vendors restructuring or being acquired below their last private mark within four quarters, while IFR unit growth remains positive

    Source

    S-13-05 The Robot Report; S-13-01 IFR press releases

    Would Be Visible By

    2027-12

  • Indicator

    A RaaS provider disclosing a financing covenant breach or an impairment of leased fleet assets

    Source

    S-13-03 SEC EDGAR full-text search

    Would Be Visible By

    2027-12

  • Indicator

    European installations falling for a fourth consecutive year from the roughly 85,000 recorded for 2024

    Source

    S-13-01 IFR press releases

    Would Be Visible By

    2027-10

Affected Industries
  • 13
  • 09
  • 10
  • 11
What Businesses Should Do

Diligence the vendor's balance sheet as carefully as its robot when buying under RaaS, and require step-in or escrow rights on fleet software; price Chinese-supplied hardware into the capital plan rather than assuming incumbent pricing holds.

Precedence Note

Requires both legs: vendor distress and continued positive unit growth. Falling unit volumes alone indicate a demand problem and belong in base or failure, not here.

Would Change Our Mind

A 100 basis point reduction in the federal funds target combined with European installations returning to growth.

disruption · 10% · band

A real humanoid deployment appears. One company discloses an audited, multi-hundred-unit deployment doing paid work with unit economics that pencil, and the capital structure of the whole sector re-rates upward. Component capacity that looked speculative becomes scarce, and the incumbents' installed-base advantage is suddenly worth much less.

Mechanism

The substitute is a general-purpose machine that amortises across tasks, which changes the integration arithmetic rather than the hardware cost. Two independent parties are required for this to be a sector rather than a project; the corpus records component-layer commitments from Schaeffler, planning mass production of strain wave gearboxes for humanoids from 2027, and Harmonic Drive's move to the TSE Prime Market, as capacity being committed ahead of demonstrated demand. The incumbent asset that becomes worth less is the integrator relationship and the fixed-cell installed base.

Preconditions, early indicators and assumptions
Early Indicators
  • 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

    Source

    S-13-03 SEC EDGAR; S-13-04 Shanghai Stock Exchange

    Would Be Visible By

    2027-12

  • Indicator

    Harmonic Drive Systems consolidated bookings stepping up with humanoid programmes named as the cause

    Source

    S-13-10 Harmonic Drive Systems IR

    Would Be Visible By

    2027-12

  • Indicator

    A named industrial customer disclosing a multi-hundred-unit humanoid purchase order in its own filings

    Source

    S-13-03 SEC EDGAR full-text search

    Would Be Visible By

    2028-06

Affected Industries
  • 13
  • 09
  • 03
  • 11
  • 18
What Businesses Should Do

Hold a standing evaluation brief that can be activated on the first audited deployment disclosure rather than running continuous pilots on unproven units; secure reducer and magnet supply terms early, because the component layer is where scarcity would bind first.

Precedence Note

Requires a filed document with revenue, unit count and repeat orders from the same customer. Nothing resembling this exists as of 2026-09-15, and no press release, tracker or demonstration substitutes for it.

Would Change Our Mind

Two more years of filings in which every disclosed humanoid revenue line stays below $25m of industrial revenue.

regulatory · 11% · band

The chokepoint is used. Chinese rare-earth licensing is tightened against Japan and Europe rather than merely allocated, and robot production is constrained physically rather than commercially. On the alternative path, a serious injury involving a collaborative or mobile robot triggers an ISO or OSHA tightening that pushes mobile and learned-behaviour deployment back years. Either way regulation, not technology, becomes the binding variable, and the direction is genuinely two-sided because licensing can be loosened as abruptly as it was tightened.

Mechanism

The instrument is Chinese export licensing of NdFeB permanent magnets, roughly 80% of global output, which sit in the servo motor of essentially every industrial robot; Japan received zero covered rare-earth exports in July 2026 while South Korea became the largest destination. On the safety path the instrument is ISO/TC 299: current certification under ISO 10218-1/-2 and ISO/TS 15066 presumes deterministic, analysable behaviour in fixed or fenced workspaces, and no harmonised standard yet certifies a free-roaming biped or a learned policy.

Preconditions, early indicators and assumptions
Early Indicators
  • Indicator

    Japanese robot makers disclosing magnet-related production constraints in quarterly results

    Source

    S-13-08 FANUC IR; S-13-09 Yaskawa IR; S-13-10 Harmonic Drive IR

    Would Be Visible By

    2027-12

  • Indicator

    ISO/TC 299 opening a new work item on mobile manipulator or learned-behaviour safety

    Source

    S-13-20 ISO/TC 299 Robotics standards

    Would Be Visible By

    2028-06

  • Indicator

    China customs data showing covered rare-earth exports to Japan staying at zero for six consecutive months

    Source

    S-13-18 China customs rare-earth and magnet export data

    Would Be Visible By

    2027-03

Affected Industries
  • 13
  • 09
  • 10
  • 08
  • 03
What Businesses Should Do

Map magnet content and reducer sourcing to a named country of origin for every machine in the capital plan, and hold a qualified second source even at a cost premium; treat safety certification of learned behaviour as an unpriced schedule risk on any mobile-manipulator project.

Precedence Note

Regulatory outranks base and downside on occurrence of a disclosed production constraint or a new ISO work item, even if unit volumes look base-like.

Would Change Our Mind

Covered rare-earth exports to Japan resuming at pre-2026 levels for three consecutive months with no standards activity on learned behaviour.

failure · 5% · band

The second robot winter. Humanoid capital destruction becomes 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 while the Western venture ecosystem for robotics largely empties out.

Mechanism

The thesis that fails is that category-level valuation can be sustained by demonstrations. Figure carried a $39bn valuation, Unitree peaked near $66bn and XPeng's Dogotix raised $900m at $6.3bn, against Agility's $1.8m of 2025 revenue and a $140m operating loss. When the gap closes, the holders are venture and crossover investors and, through the STAR listing, Chinese retail shareholders; Unitree already fell 53% from its first-day high by 2026-09-09 with nothing changing operationally. The historical analogue is Rethink Robotics, which invented the cobot vocabulary, raised roughly $150m and shut down in 2018 while the category it created grew under Universal Robots — recovery of the category took about five years and the pioneer did not participate.

Preconditions, early indicators and assumptions
Early Indicators
  • 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

    Source

    S-13-05 The Robot Report; S-11-01 PitchBook-NVCA Venture Monitor

    Would Be Visible By

    2028-02

  • Indicator

    Unitree trading below its RMB 150.80 offer price

    Source

    S-13-04 Shanghai Stock Exchange

    Would Be Visible By

    2027-12

  • Indicator

    A SPAC transaction in the category terminating before completion

    Source

    S-13-03 SEC EDGAR

    Would Be Visible By

    2027-06

Affected Industries
  • 13
  • 11
  • 03
  • 09
What Businesses Should Do

Separate robotics supplier risk from humanoid narrative risk in vendor selection, because the two are being priced together and are not the same exposure; prefer vendors with audited revenue from deployed fleets over vendors with benchmark results.

Precedence Note

Failure outranks base on occurrence of a funding collapse plus an insolvency. Valuation deflation without a funding-market effect stays in base, because the corpus records the Unitree move as deflation from euphoria rather than collapse — it still trades at three times its offer price.

Would Change Our Mind

Robotics venture funding excluding humanoids growing year over year for two consecutive quarters through a humanoid drawdown.

Additional scenario notes

Probabilities Sum

1

What Must Be True To Grow
  • Capability: a humanoid performs paid work outside a demonstration, in someone else's hands, with a third party able to verify it
  • Economics: disclosed unit economics at a capital-equipment price point, not a $13,500 research and entertainment price point
  • Supply: NdFeB magnets and precision reducers available at rate, against roughly 80% Chinese magnet output and a Japanese reducer duopoly
  • Demand: repeat orders from the same industrial customer, which no filing anywhere currently shows
  • Permission: a harmonised safety standard that certifies a free-roaming biped or a learned policy, which no jurisdiction has yet attempted
  • Capital: financing available at a 3.50-3.75% policy rate with Chinese listing authorities now requiring recurring revenue
What Could Stop It
  • Input chokepoint: Chinese rare-earth export licensing, with Japan at zero covered exports in July 2026
  • Physical constraint: precision reducers, a Japanese duopoly with process know-how that does not transfer from a datasheet
  • Capital cost: long-payback automation failing to clear at current rates, now sitting on vendor balance sheets through RaaS
  • Substitution: wheeled, fixed-base and purpose-built machines that are cheaper to integrate for the same task
  • Attention withdrawal: the promoter class moving on, which the corpus records as the systematic pattern across the sector's fifty overhyped trends
Uncertain Assumptions
  • Text

    Private humanoid companies — Figure, Apptronik, 1X, AGIBOT, UBTech, Neura — are not concealing material industrial revenue

    Confidence

    medium

    Load Bearing

    false

    If Wrong

    The true installed base is larger than the filed record shows and the base case understates the category; the corpus position is that the number is unknown rather than known to be small

  • Text

    IFR's headline installation figures remain available in the free press releases rather than moving entirely behind the paywall

    Confidence

    medium

    Load Bearing

    false

    If Wrong

    The unit-volume indicators in base, upside and downside become unresolvable without a licence, and the sector loses its only population-level measure

  • Text

    Filed financial disclosure rather than demonstration is the evidence that settles the question

    Confidence

    high

    Load Bearing

    true

    If Wrong

    The whole set is measuring the wrong thing, and a category could reach industrial scale through private Chinese deployment that never appears in any filing this framework reads

  • Text

    Manipulation generalisation does not follow a scaling curve like language did

    Confidence

    low

    Load Bearing

    false

    If Wrong

    Upside and disruption both become far more likely and the base case's low-thousands-of-units humanoid outcome is wrong by an order of magnitude

Authored

2026-09-15

More from this sector