2026人形機(jī)器人：新市場(chǎng)的交匯時(shí)刻（英）

STUDY

Humanoidrobots2026

Theconvergencemomentforanewmarket

MANAGEMENTSUMMARY

Humanoidrobots2026

Theconvergencemomentforanewmarket

Humanoidrobotsaremovingfromsciencefiction

toindustriaIreaIity.AdvancesinAIandrobotics

hardwarearemakingitincreasingIyfeasibIetobuiIdmachinescapabIeofoperatinginhuman–designedenvironments.Atthesametime,Iaborshortagesarestrengtheningtheneedfornewformsofautomation.Thismarksaconvergencemomentwhere

technoIogicaIcapabiIitymeetsmarketdemand,bringinghumanoidrobotscIosertoreaI–worId

depIoymentacrossindustryandbeyond.

IfcurrenttrajectorieshoId,theeconomicimpIicationscouIdbesubstantiaI.Aspartofthebroadermove

towardphysicaIAI,humanoidrobotsareIikeIyto

evoIveintoamuIti–triIIion–doIIarindustry-accordingtoourprojections,representingamarketofuptoUSD750biIIionby2035anduptoUSD4triIIionby2050,

comparabIeinscaIetotheautomotiveindustrytoday.ForindustriaI,automotiveandeIectronicscompanies,humanoidrobotsthusrepresentasignificantgrowthopportunityand,atprojectedoperatingcostsof

justtwodoIIarsanhour,apotentiaIIeverformajorefficiencygains.

WhatdistinguishesthecurrentwaveofdeveIopmentfrompreviousautomationcycIesishowrobots

areIearningandadapting.Advancesacrosskey

technoIogiessuchasadvancedactuators,edge

computingandgenerativeAI/vision–Ianguage

modeIs(VLMs)providethefoundationforinterpretingcompIexandunpredictabIeenvironments.With

working–agepopuIationsprojectedtodecIineby

upto22percentinsomeregionsby2050,industries

aIsofaceastructuraIIaborgapthattraditionaI

automationcannotfuIIyaddress.Humanoidrobots

offeradistinctiveadvantage:theycanoperatewithinprocessesandinfrastructuredesignedforhuman

workers,performingdiversetaskswithoutexpensiveproductandfaciIityredesign.

PerhapsthestrongestvaIidationofthistrajectoryisthebreadthofindustryengagementaIready

emerging.EarIyprototypesarebeingdepIoyedin

manufacturingfaciIities,whiIeIogisticsoperators

aretestingwarehouseappIications.Combinedwithequityinvestmentsofapprox.USD10biIIionand

partnershipswithIeadingsemiconductorpIayers,humanoidroboticsismovingbeyondspecuIationtowardcommerciaIreaIity.ThekeyquestionisnoIongerwhetherhumanoidrobotswiIIemergeasaviabIetechnoIogy,buthowquickIytheywiIIscaIe-andwhichcompaniespositionthemseIvesearIyenoughtocapturetheopportunity.

2IRoIandBerger

Contents

P4

P8

P14

P16

P21

CoverAIgenerated

P26

1/Atrillion-dollarmarket-withuncertaintiming

Marketgrowthscenariosandtheeconomicsofhumanoidrobots

2/Technologicalreadiness

Hardwareismaturing,butsoftwareandecosystemgapsremain

3/Twoecosystems,twoscalingcurves

China'sdeployment-ledstrategyandtheWest'sAI-drivenapproach

4/Wherevaluewillemergefirst

Laborshortages,productivitygainsandfirstdeploymentopportunities

5/Strategicimplicationsforindustryplayers

Howcompaniesshouldpositionthemselvesintheemergingvaluechain

6/Conclusion

Humanoidrobots2026|3

4IRoIandBerger

1/Atrillion-dollarmarket-withuncertaintiming

Marketgrowthscenariosandtheeconomicsofhumanoidrobots

Funding

Approx.USD10biIIon

inventurecapitaIand

strategicinvestment

gIobaIIy

Techecosysteminvolvement

Leadingsemiconductor

andtechnoIogy

companiespartnering

withhumanoidrobot

deveIopers

Cross-industrysynergies

AppIicationsacross

industries,with

automotiveemerging

asanchorproducer

andbuyer

Humanoidrobots

Theconvergencemoment

Technologymaturity

Hardwarenearing

readiness-remaining

chaIIengeIargeIy

softwareaddressed

throughAI

Solutiontoa

structuralproblem

Laborshortagesand

agingworkforcesdrive

sustaineddemand

forautomation

Marketpotential

Long–termmarket

potentiaIexceeding

USD1triIIion

T

hehumanoidroboticsmarkettodayremainsinitsprototypingphase,withdeploymentsmeasuredindozensratherthanthousands.Yetprojectionsindicateexponentialgrowth.Basedonourmarketmodeling,weexpecttheindustrytoreachUSD300billionattheOEMlevelby2035underbaselinescenarios,risingtoUSD750billioninmoreoptimistictrajectories.Overtime,thispointstoatrillion-dollarmarket,evenifthetimingoflarge-scaleadoptionremainsuncertain.Theseestimatesarebasedonmodeleduniteconomics,projecteddemandpatternsandanalogiestosupplychainramp-upsinotherindustries.

Theeconomicfundamentalsincreasinglysupportthistrajectory.AtprojectedcostsofUSD20,000-30,000peradvancedhumanoidrobotincludingalltrainingcosts,theoperationaleconomicsbecometransformative.WithanhourlyoperatingrateofapproximatelyUSD2-afractionofhumanlaborcostsindevelopedmarkets-theserobotscandelivercompellingreturnsinindustrialenvironmentswhilealsobecomingaccessibletoprivateconsumersforhouseholdtasks.Thisdual-marketpotentialaddresseslaborshortagesacrossmanufacturing,logisticsandserviceswhileopeningentirelynewconsumermarkets.

bA

ANotallhumanoidsarecreatedequal

Advancedandentry-levelrobotsdifferincapabilities,dimensionsandprice

Pricerangein2035

Heightandweight

Features

Advanced

USD20,000-30,000

165-180cm,65-80kg

Powerful,full-sizedsystems

designedforcomplextasks

acrossindustries,offeringhighadaptability,strengthand

advancedAIcapabilities

Entry-level

USD8,000-10,000

120-140cm,30-40kg

Compactandaffordable

systemsfocusedonbasic

functions,withlightweight

designandalimitedtaskrangesuitedtosimpleautomation

andserviceapplications

Source:Marketinterviews,deskresearch

6IRoIandBerger

THEEMERGINGHUMANOIDROBOTICS

INDUSTRYECOSYSTEM

Theopportunityextendsfarbeyondfinishedrobots-thesurroundingvaluechainisalsoexpandingrapidly.By2035,bodyactuatorsalonerepresentamarketworthbetweenUSD26billionandUSD79billion,withhandactuatorsaddingafurtherUSD9billiontoUSD26billion.Computeandconnectivitysystems,perceptionsystems,structural

components,energyandchargingsystems,andothersubsystemscontributemorethanUSD35billion.Theassemblyandsupplychainsegment-encompassingassemblycosts,overhead,energyandlabor-couldreachbetweenUSD45billionandUSD113billion,creatingsubstantialopportunitiesforequipmentmanufacturersandserviceproviders.B,C

BThenextautomotiveindustry?

GIobaIhumanoidroboticsmarketsizebysystem/component,2035[USDbn]

310

Humanoidrobot(OEMIeveI)

300

750

AssembIy&suppIychain

Motion-actuator

Hand/end–effectorsystem

Motion-other

Energy&charging

SkeIeton&structuraIcomponents

Compute&

connectivity

Perceptionsystem

OthercomponentsOptimisticBaseIine

45113

2679

9

7

6

26

21

18

1442

617

39

Marketsize-humanoidrobots[OEMIeveI]System/componentMarketsize2035[USDbn]

No.ofunits

peryear[m]

3

30

60

120

200

Optimisticscenario

4,000

2,400

750

1,200

<1

90

No.ofunits

peryear[m]

1

10

30

60

100

BaseIine

scenario

2,000

<1

40

300

600

1,200

2025

2030

2035

2040

2045

2050

Source:RoIandBergerHumanoidRobotsmarketmodeI,marketinterviews,deskresearch

Humanoidrobots2026I7

Lookingbeyond2050,thepotentialcouldapproachthescaleoftheautomotiveindustry.TheglobalautomotivemarketgeneratesroughlyUSD2.5trillionannuallyinvehiclesalesalone,withsubstantialadditionalvalueinparts,servicesandmanufacturinginfrastructure.Ashumanoidrobotsachievemassdeploymentacrossindustrialandconsumermarkets,similareconomicscouldemerge.Underoptimisticscenarios,thiscouldmeanOEM-

levelrevenuesexceedingUSD4,000billionannually,withcomponent,serviceandmanufacturingequipmentmarketstogethercreatingatotaladdressablemarketapproachingautomotive-scaleproportions.Realizingthispotentialwilldependonsustainedtechnologicalprogress,buttheeconomicdriversandmarketstructuresuggesthumanoidroboticscouldbecomeoneofthedefiningindustrialsectorsofthemid-21stcentury.

C3xtheprice

Componentcostbreakdownofadvancedvs.entry–IeveIhumanoidrobots

Singlehumanoidrobot[USD]Totalmarketsize[USDbn]

Categories

Motion-actuator

Advanced4,000

Entry-level1,265

Advanced60

Entry-level19

Total

79

Hand/end–

effectorsystem

1,400

333

21

5

26

Motion-other

1,072

342

16

5

21

Energy&charging

845

380

l13

6

18

SkeIeton&structuraIcomponents

1,950

820

.29

12

.42

Compute&

connectivity

800

345

12

5

17

Perceptionsystem

465

166

7

2

i9

Others

(e.g.connectors)

440

230

7

3

10

Total11,0003,90016558223

Source:RoIandBergerHumanoidRobotsmarketmodeI,marketinterviews,deskresearch

8|RolandBerger

2/Technologicalreadiness

Hardwareismaturing,butsoftwareandecosystemgapsremain

O

verthepastyears,humanoidrobotshavemadesubstantialprogress.Currentprototypesalreadydemonstratefunctionalmobilityanddexterity,allowingrobotstoexecutetasksincontrolledandsemi-structured

environments.Coresubsystems-compute,sensing,actuationandpower-arevalidatedatapilotlevel.

However,technicalfeasibilitydoesnotyettranslateintoscalableindustrialreadiness.Hardwaredevelopment

DMindthegap

HardwarematurityassessmentofcurrentR&Dprototypes

Masscommercializationreadynow

Compute&PerceptionEnergy&Skeleton&

connectivitysystemchargingstructuralcomp.

Maturity

Description

?Processingandcommunicationarchitecture

?Enablesreal-timedataprocessing

?Vision:cameras,3Dsensors

?Touch:e-skin

?Motion:IMU,accelerators

?Li-ioncells+BMS

?Thermalmanagement

?Machined/castaluminum/steelparts

?PEEK1materialsforstrength/weight

Currentstatus

?Sufficientprocessingpower

?EdgeAIenables

on-deviceinference

?5-10mscontrolloops

?Visionhardwareismature

?3Dstructuredlightfornextgeneration

?E-skinevolutionongoing

?Currentruntime

2-8hourspercharge

?Targetfor2028:

16hours

?Automotivebatterydirectlytransferable

?PEEK1provenin

aerospaceandmedical

?Modulardesign

enablesflexiblecomponent

replacement

Keychallenges

?Regulation(e.g.ICTS)

?Need100+sensors/

?Achieve16hours

?PEEKis5-10xmore

createsdeviating

standards:ChinaandWesterncountries

?Computevs.battery

?Heatdissipationincompactform

handforhuman-leveltouch

?Processmultimodaldatastreamsinrealtime

runningtimewithout

weightpenalty

?Fastcharging

(30-60min.)

degradesbattery

lifespan

expensivethannormalindustrialplastics

?Long-termdurabilityisunprovenatscale

?Optimizestrength-to-weightratio

LowHigh

1Polyetheretherketone:ahigh-performance,semicrystallinethermoplasticknownforextremetemperatureresistance(upto250°C),superiormechanicalstrengthandexceptionalchemicalresistance

Source:Marketinterviews,deskresearch,pastRolandBergerprojects

Humanoidrobots2026|9

hasreachedanadvancedpre-commercialstage:systemsoperatereliablyindemonstrationsandearlypilots,butcostefficiency,long-termdurability,scalabilityacrossusecasesandsupplychainrobustnessremainunderdevelopment.

Theremaininggapisnolongerabout"canitwork?"-butaboutwhetheritcanoperatereliably,affordablyandatscale.D

Masscommercializationin1-3years

Motionsystem-actuator

Motionsystem-other

Hand/end-

effectorsystem

Other

components

Maturity

Description

?Motors&reducers

?Encoders,drives

?Multi-DOFrobotic

?Wiringharnesses

?Multiplerotary&

andtorquesensors

handsandgrippers

?Displays,audio

linearactuators

?Bearings

?Underactuated/fully

?Connectors,

(25-35)perrobot

actuatedfingers

fasteners

?Tendonforhand

?Tactilesensing

Currentstatus

?Transitiontoaxialfluxmotors+cycloidal

reducers

?Costdeclining

drastically(upto50%)

?Responsetime<5-10mslatency

?Force-torque

feedbackenablessafeinteractions

?Bearingandencoderarematuretechnology

?Earlycommercialdexteroushands(low-midvolume)

?Limitedrobustnessfor

continuousindustrialuse

?Perceptionandcontrolstillinearlystages

?Automotive-gradeconnectors

?Displayandaudioarematureand

cost-effectivetechnology

Keychallenges

?Axialfluxmotorsand

cycloidalreducersneed1-3yearstomature

?50-90%costreductionrequired

?Balancepower&safety

?Coordinate30-50DOFinrealtime

?Longbearingwearunproven(e.g.in

continuousbipedalmovement)

?Reducevibration/noise

?Human-likedexterityatindustrialcost

?Improvedurability

?Reduceactuatorcount

?Integratetactile

sensingwithouthighercomplexity

?Route30-50+cablesthroughmovingjoints

?Electromagnetic

interferenceshielding

?Designfor

serviceability

LowHigh

Source:Marketinterviews,deskresearch,pastRolandBergerprojects

10|RolandBerger

Industryconsensuspointstoinitialhardwaredesignstabilizationaround2028-29,withsupplychainmaturationexpectedtofollowthereafter.Asaresult,commercializationwillunfoldgraduallyratherthansimultaneouslyacrossallsubsystems.Threerecurringconstraintsmustbeovercometobridgethegapbetweentoday'sprototypesandcommerciallyviablesystems.

1

Thecost-performanceimperative:

Commercialdeploymentrequiressubstantialcostreductions-estimatedat50-90percentforcriticalsubsystemssuchasactuators-whilesimultaneouslymaintainingorimprovingsafetyandperformancecharacteristics.

2

Thedurabilitygap:

Amajorchallengeremainsthedurabilityofcomplexsystemsindemandingproductionenvironments.Forexample,advancedrobotichandscurrentlyhavealifespanoflessthanoneyearinvolumeapplications,necessitatingfrequentandcostlyreplacements.

3

Ongoingtechnologytransitions:

Severalsubsystemsareprogressingthroughgenerationalshifts,suchasthemovetowardaxialfluxmotorsandcycloidalreducers.Whilethesetransitionsmayimproveperformance,theycouldextendadoptiontimelinesbyapproximatelyonetothreeyearsasnewdesignsarevalidatedandstandardized.

Actuators-thecorevaluedriver

Actuatorsrepresentthesinglemostcriticalsubsysteminhumanoidrobots.Theydeterminetorquedensity,dynamicperformance,energyefficiencyandultimatelycoststructure.Currentsystemsrelyprimarilyonelectricmotorscombinedwithharmonicorcycloidalgearboxes.Theindustrytrendismovingtowardfullyintegratedactuatormodulescombiningmotor,gearbox,driveelectronics,torquesensingandthermalmanagementintoacompactunit.

Keytechnologicalleversincludehighertorquedensitymotors,low-backlashhigh-efficiencyreducers,alongsideintegratedforceandtorquesensingandimprovedthermalmanagement.Cost-optimizedmanufacturingatscalewillbeessentialforcommercialization.

Softwareandecosystemmaturity

Whilehardwareplatformsareapproachingfunctionaladequacy,thebroaderecosystem-includingsoftwarearchitectures,datainfrastructure,supplychainindustrializationandregulatoryframeworks-remainsmateriallylessmature.Accordingtoexpertassessmentsandinterviewswithindustrystakeholders,ecosystemreadinesscurrentlytrailshardwaredevelopmentbyanestimatedthreetofiveyears.Physicalsystemsoperateinpilotenvironmentswithincreasingreliability,buttheenablingconditionsrequiredforrepeatable,large-scaledeploymentarestillintheprocessofevolving.E

Humanoidrobots2026|11

ESmartrobotsneedsmartersoftware

Softwareandecosystemmaturityassessment

formasscommercializationSoftwaretrailinghardwareby3-5years

Ecosystemnotyetready

?Safetystandardsforhuman-humanoidrobotcollaboration

?Exportcontrols

(ICTS),AIgovernance(EUAIAct),product

compliance(CE,FCC)

?Standardstimeline:2-3yearsminimumforISOratification,industry

adoption,certificationinfrastructure

?FragmentedregulationacrossUS,EUandChina

?Unclearwhobears

responsibilityforAI-drivenphysicalerrors

DescriptionCurrentstatusKeychallenges

VLM1

?Coresystem

combiningperceptionandreasoning

?Enableszero-shotlearning2

?GenAIcompressingdevelopmentcycle

?Controlledenvironment

tasksapproachinghumanlevel

?Open-endedenvironmentsstillneed5-10years

?Difficultywithcontext-basedreasoning

?Reliabletransferof

virtualtrainingto

physicalenvironment

?Localcomputingpowerrequirementishigh

(200+TOPS)

Trainingdata

?Multimodaldatasets:vision,tactile,

proprioception,forcefeedback

?Capturediverse

environments,

tasks,failuremodes,recoverystrategies

?Publicdatascarcity:LLMtextcorporaexist;robot

manipulationdatadoesn'texistatscale

?Limitedopen-sourcedatasetsforroboticsAItraining

?LeadinghumanoidOEMsarecollectingproprietarydata

?DatagenerationbottleneckpreventsLLM-alike

large-scaletraining

?Needlabeled,multi-angleandmultimodalrecordingsexponentiallymore

expensivethantext

?Databecomesa

competitiveadvantage

andproprietaryassetof

leadingOEMs,leadingtoafragmentedecosystem

Supply

chain

?Globalsupplier

networksacrossUS,

EuropeandChinawithregionalspecialization

?Integrationofexistingautomotive,electronicsandroboticssupply

chains

?Nomatureend-to-end

supplychainyet:USfocusedonsoftware;Chinafocusedonindustrialization

?Dualsupplychain

regulatedcomponents

(high-cost)sourcedfrom

Westerncountries;standardcomponents(low-cost)

sourcedfromChina

?MostTier2suppliersarestillinthetestingphases:1-2

yearsminimumtoprogresstomassproductionstage

?Suppliersarehesitanttoinvestincapacity

?ICTS/exportcontrolsforce2-3xcostpremiumsfor

criticalcomponents

Regulation

?Noharmonizedglobal

standards:US/EU/Chinapursuingdivergent

D

regulatorypaths

?ICTSimpact:13critical

componentsfaceexclusionfromChinesesuppliers

?Safetycertification

undefined:existing

standardsdonotapplytohumanoids

Maturity:LowHigh

1Vision-languagemodel,2Amachine-learningcapabilitywhereamodelcancorrectlyhandletasksitwasneverexplicitlytrainedon,withoutseeinglabeledexamplesbeforehand

Source:Marketinterviews,deskresearch,pastRolandBergerprojects

12|RolandBerger

TheprimarybottleneckhasshiftedfrommechanicalengineeringtoAIarchitectureanddatastrategy.Leadingdevelopersaretransitioningtowardvision-languagemodelsandend-to-endlearningsystemsthatdirectlyconnectperceptiontoactuation.Inspiredbyautonomousdrivingarchitectures,theseapproachesreducemanualprogrammingandenableadaptivetaskexecution.

HierarchicalAIstacksareemergingasthedominantdesign:ahigh-levelreasoninglayer(vision-languageandfoundationmodels)enablestaskplanningandcontextualunderstanding,whilealow-levelcontrollayertranslatesintentintoprecisemotorcommandscloselycoupledtotherobot'skinematics.Thisarchitecturesupportsgradualexpansionfromsingle-tasktrainingtowardbroadergeneralization-aprerequisiteforcross-industrydeployment.

However,theshifttolearning-basedsystemsintroducesseveralstructuraldependenciesthatwillshapehowtheecosystemevolves:

Dataasthecoreconstraint

UnlikegenerativeAIsystems,humanoidrobotsrequiresynchronizedsensor-to-actuatordatafromreal-worldenvironments.Suchdataisproprietaryandcostlytogenerate,andremainsscarceinreal-worldoperatingenvironments.Syntheticdata,teleoperation,industrialpartnershipsandfleetlearningarethereforeessentialcomponentsofcompetitivestrategy.

Simulationasanaccelerator,notasubstitute

Physics-basedsimulationenvironments(worldmodels)enablescalabletrainingandreduceearly-stagedatarequirements.Yetthesim-to-realgappersists,limitingthefeasibilityofpurelyvirtualtraining.Real-worldvalidationthereforeremainsindispensable.

Computeinfrastructureasabarriertoentry

TraininghumanoidfoundationmodelsdemandssubstantialAIhardwareclustersanddistributedtrainingpipelines,alongsideoptimizedinferencesystems.Thisincreasescapitalintensityandfavorsplayerscapableofverticallyintegratinghardware,softwareandAIinfrastructure.

Beyondsoftwareanddata,thebroaderindustrialecosystemwillalsoshapethepaceofdeployment.Supplychainmaturitywilldeterminehowquicklyhumanoidrobotstransitionfromlow-volumepilotstoeconomicallyviablemassdeployment.Today'ssystemsarelargelyassembledfromcustom-developed,high-cost,low-volumecomponents.

Scalingrequiresashiftfromengineering-drivensourcingtowardplatform-based,automotive-stylesupplyecosystems.Currentdevelopersdependheavilyonspecializedsuppliersforactuators,precisiongearboxes,sensorsandcontrolelectronics.Thesesuppliersoperateatlimitedscale,withextendedleadtimesforcriticalcomponentssuchasharmonicdrivesandhigh-torquemotors,creatingbottleneckseveninpilotprograms.

Regulatoryframeworksrepresentanadditionaldeploymentconstraint.Existingsafetystandardsweredevelopedfortraditionalautomationsystemsoperatingwithinfixed,enclosedzones,aswellasforcollaborativerobots("cobots")withdefinedoperatingenvelopesandpredictabletoolconfigurations.Humanoidrobots,bycontrast,functionindynamic,human-centricenvironments,canchangeposition,heightandtools,andmanipulateawiderangeofobjects.Thisvariabilitymakesitsignificantlymorecomplextodefineconsistentriskprofilesandrendersexistingsafetyconceptsinsufficient.

Futureregulatoryframeworkswillthereforeneedtoaddressmovementspeed,forcelimits,objectinteraction,reactiontimesandAI-drivendecisionprocesses.

Establishingsuchstandardsrequiresextensivetesting,validationprotocolsandempiricalsafetydata-particularlyinregulation-intensivemarkets.Atpresent,noharmonizedglobalframeworkexists:companiesmustnavigateafragmentedlandscapeofmachinerydirectives,workplacesafetyrules,productliabilityregimesandemergingAIgovernancerequirements.Thisfragmentationincreasescompliancecomplexityandextendstime-to-market.

Humanoidrobotsare

movingfromscience

fictiontoreality-thekey

gapisinsoftwareand

dataforAImodels.

Humanoidrobots2026|13

14|RolandBerger

3/Twoecosystems,twoscalingcurves

China'sdeployment-ledstrategyandtheWest'sAI-drivenapproach

T

hehumanoidroboticsmarketisnotevolvingasasingleglobalrace.Astechnologicalreadinessimproves,twoecosystemsareemergingwithdistinctscalinglogics:NorthAmericaandEMEA(Europe,theMiddleEastandAfrica),pushingAI-first"generalist"robotarchitectures;andChina,industrializingfasterwithamanufacturinganddeploymentflywheel.Thesecontrastingapproachesshapehowquicklyrobotsreachreal-worlddeploymentandwherecompetitiveadvantagesarelikelytoemerge.F

WESTERNECOSYSTEM:AI-FIRST,CAPITAL-RICH,SCALE-POOR

WesternleadersareincreasinglypositioningthemselvesasAIandsoftwarecompanies,bettingthatcompetitiveadvantagewillcomefromfoundationmodelsandvision-languagesystems,supportedbyproprietarydatasetsthatenablerobustautonomyinunstructuredenvironments.Thecapitalbasesupportsthisview:NorthAmericahasnearlythesamefunding(USD3.8billion)despitehavingfewerstartupOEMsthantheChineseecosystem.

Theconstraintislessmechanicaldesignandmore"dataplusdeployment":real-worldtrainingdata,validationcyclesandsafetycases.Inthecurrentsnapshot,Westernproductionremainslargelyinthepilotphase,slowing iterationanddelayingsoftwarematuration.

CHINESEECOSYSTEM:DEPLOYMENT-FIRST,SCALE-DRIVENLEARNING

Chinaispursuingavolume-ledstrategy:deployrobotsintodefined,controlledworkflowssuchasentertainmentandlogistics,iteraterapidlyanddrivethecostcurvedownthroughmanufacturingscale.Thatapproachisvisibleinoutput:morethan15,000unitsin2025–atleast30timesNorthAmerica'svolumeandover150timesthatofEMEA.

PolicysupportandIP(intellectualproperty)leadershipreinforcestheindustrializationpush,withaclearroadmaptowardecosystembuild-outandscaleddeployment.China

alsoleadspatentinginhumanoidrobotics,signalingsustainedinvestmentincorecapabilities–notonlyassemblycapacity.

STRATEGICIMPLICATIONS

Thesecontrastingstrategiesarecreatingtwodistinct industrialflywheelsandthreestrategicpositionsintheemergingmarket.Chinacurrentlybenefitsfromascaleadvantage,buildingapowerfuldataandcostflywheel throughrapiddeployments.NorthAmerica,bycontrast,commandsdeepcapitalpoolsandstrongAIcapabilitiesbutwillneedastepchangeinproductionscaletoclosethe"dataplusdeployment"gap.EMEAoccupiesamoreconstrainedposition,withasmallerstartupbaseand limitedfunding,combinedwithminimalprojected2025output.Thisincreasestheriskoflong-termdependenceoneitherChinesehardwareplatformsorUSAIstacks.

Geopoliticsislikelytoreinforcethesedifferences.Exportcontrols,procurementrulesand"trustedsupplychain"requirementsarepushingtheindustrytowardparalleltechnologystacksandregionallyanchoredsupplychains–effectivelycreatingseparatetechnologymarketswithlimitedcross-borderinteroperability.

Humanoidrobots2026|15

FDifferentregions,differentstrengths

Startups,funding,productionandtechnologicalmaturitybyregion

North

AmericaEMEAChina

RoW

Total

No.ofstartupOEMs[#]

25

22

39

20

106

No.ofauto/tech/

industrialOEMs[#]

2

2

13

6

23

StartupOEM

funding[USDbn]

3.8

0.8

4.1

0.3

9

Production2025[#]

~500

~100

~15,000

~300

16,000

Technologicalmaturity

Source:Deskresearchandpublicinformation

16|RolandBerger

4/Wherevaluewillemergefirst

Laborshortages,productivitygainsandfirstdeploymentopportunities

R

egardlessofthe