2022年人工智能行業(yè)研究：主要報(bào)告

March2023

ArtificialIntelligenceSectorStudy

ResearchreportfortheDepartment

forScience,Innovation&

Technology(DSIT)

Contents

IExecutiveSummary 2

1.Introduction 5

1.1.Methodology&Sources 5

1.2.Approach 5

1.3.InterpretationofData 6

1.4.Acknowledgements 7

2.UKArtificialIntelligenceSectorProfile 6

2.1.DefiningtheUKArtificialIntelligenceSector 6

2.2.NumberofUKAICompanies 8

3.LocationofUKAICompanies 17

3.1.AIActivitybyUKRegion 17

3.2.RegionalAIActivitybySector 18

3.3.InternationalActivity 19

4.EconomicContributionofUKAICompanies 22

4.1.EstimatedRevenue 22

4.2.EstimatedEmployment 26

4.3.EstimatedGrossValueAdded 30

4.4.SummaryofEconomicContribution 31

5.InvestmentinUKAICompanies 33

5.1.InvestmenttoDate 33

5.2.InvestmentMarketDynamics 41

6.FutureAISectorDevelopment 44

6.1.RecentSectorDevelopments 44

6.2.PotentialFutureSupport 45

6.3.SectorChallenges&Opportunities 46

6.4.FurtherSectorAnalysis,Monitoring&Evaluation 49

IExecutiveSummary

ThegovernmentcommissionedPerspectiveEconomics,glass.ai,IpsosandacademicexpertstoundertakearesearchstudytobetterunderstandtheprofileoftheUKAISectoranditscontributiontotheUKeconomy.Basedonacombinationofextensivecollectionandanalysisofsecondarydataandstrategicqualitativeresearchincludingasurveyof250UKAIbusinesses,and22in-depthinterviewswithAIbusinessesandstrategicstakeholders,thisreportprovidesabaselinesetofdataonthesizeandscaleoftheUK’sAIsector,intendedtosupportgovernment’songoingdevelopmentandmonitoringofkeyAIpolicies.

I.1HeadlineSectorMetrics

Thestudyhasidentifiedatotalof3,170UKAIcompaniesthatgenerated￡10,6bninAIrelatedrevenues,employedmorethan50,000peopleinAIrelatedroles,generated￡3.7bninGrossValueAddedandhavesecured￡18.8bninprivateinvestmentsince2016.

FigureI.1–SectorHeadlines

I.2KeyFindings

ThereportprovidesfurtherbreakdownsofthesemetricsacrossUKregions,andaccordingtopredictedAIbusinessmodelsandtechnologicalcapabilities.Someofthemostsalientfindingsemergingfromthisbaselineresearchinclude:

?Atotalof3,170activeAIcompanieshavebeenidentifiedthroughthestudy.

?Ofthe3,170activecompaniesidentifiedthroughthestudy60%arededicatedAIbusinessesand40%arediversifiedi.e.,haveAIactivityaspartofabroaderdiversifiedproductorserviceoffer.

?Comparedtosimilarstudiesintootheremergingtechnologysectors,agreaterproportionofdiversifiedAIcompanieshavebeenidentified,highlightingthebroadscopefordevelopmentofAItechnologyapplicationsbyestablishedtechnologycompaniesacrosssectors.

?Onaverage269newAIcompanieshavebeenregisteredeachyearsince2011,withapeakinnewcompanyregistrationsinthesameyearastheAISectorDeal(2018,n=429).

?Together,thedataoncompanysizeandbusinessmodelsuggestthatdedicatedAIcompaniesarebothsmallerandmoredependentonAIproductsforrevenue.DiversifiedAIcompaniesaretypicallylargerandlikelytogenerateagreaterproportionofrevenuesfromlesscapital-intensiveprovisionofAIrelatedservices.

?London,theSouthEastandtheEastofEnglandaccountfor75%ofregisteredAIofficeaddresses,andalsofor74%oftradingaddresses.JustunderonethirdofAIcompanieswitharegisteredaddressoutsideofLondon,theSouthEastandtheEastofEnglandstillhaveatradingpresenceinthoseregions,highlightingtheapparentsignificanceofthoseregionstodevelopmentoftheUKAIsectortodate.

?Whileabsolutenumbersaresmaller,thestudyhasidentifiedmorenotableproportionsofwiderregionalAIactivityinautomotive,industrialautomation&machinery;energy,utilitiesandrenewables;health,wellbeingandmedicalpractice,andagriculturaltechnology.

?Inthemostrecentfinancialyear,annualrevenuesgeneratedspecificallyfromAIrelatedactivitybyUKAIcompaniestotalledanestimated￡10.6billion,splitapproximately50/50betweendedicatedanddiversifiedcompanies.

?AcrossbothdedicatedanddiversifiedAIcompanies,studyestimatessuggestthatthereare50,040FullTimeEquivalents(FTEs)employedinAIrelatedroles,53%ofwhicharewithindedicatedAIcompanies.

?Basedonacombinationofofficialcompanydata,surveyresponsesandassociatedmodelling,AIcompaniesareestimatedtocontribute￡3.7bninGVAtotheUKeconomy.ForlargecompaniestheGVA-to-turnoverratiois0.6:1(i.e.,forevery￡1ofrevenue,largeAIcompaniesgenerate60pindirectGVA).GVA-to-turnoverratiosamongSMEsaremuchlower(0.2:1formediumsizedcompaniesandnegativeforsmallandmicrobusinesses),whichreflectsthecapitalintensive,highR&Dnatureofdeeptechnologydevelopment.

?Since2016,AIcompanieshavesecuredatotalof￡18.8bninprivateinvestment.2021wasarecordyearforAIinvestment,withover￡5bnraisedacross768deals,representing

anaveragedealsizeof￡6.7m.Further,AIinvestmentincreasedalmostfive-foldbetween2019and2021.

?In2022dedicatedAIcompaniessecuredahigheraveragedealvaluethandiversifiedcompaniesforthefirsttime.However,dataonAIinvestmentbystageofevolutionmayalsobesignallingsometighteningofinvestmentavailabletoSeedandVentureStagecompaniesand,giventhesignificanceofprivateinvestmentforAItechnologydevelopmentevidencedbydataonrevenuesandGVA,thiscouldposearisktorealisingthepotentialwithinearly-stageAIcompanies.

?ThestudyhighlightedanotableopportunityforcompaniesoperatingintheAIimplementationspacetobuildteamsofAIimplementationexpertsthatcansupportAIadoptionopportunitiesacrosssectors.Thisadoptionopportunityissupportedbyinvestmentdata,whichhighlightsthatin2022investmentsweremadein52uniqueindustrysectors,comparedtoinvestmentsacrossjust35differentsectorsin2016.

1.Introduction

PerspectiveEconomics,incollaborationwithIpsos,glass.ai,andProfessorsRobProcter(UniversityofWarwick)andRogerWoods(Queen’sUniversityBelfast)werecommissionedinAugust2022todeliveranassessmentoftheUK’sartificialintelligence(AI)sector.

Theaimofthestudyistobetterunderstandthescale,profileandeconomiccontributionofUK’sAISector,andtoprovideabaselinesetofdatathatcansupportgovernment’songoingdevelopmentandmonitoringofkeyAIpolicies.

AItechnologieshavebeenindevelopmentfordecades,howevertheirtransformativepotentialisbeingincreasinglyrealisedthroughdevelopment,applicationandpublicdebateregardingevermoresophisticatedmachinelearningsoftware.ThisreportisthereforetimelygiventheimportanceofgovernmentpolicyregardingtheethicalandregulatoryparameterswithinwhichAItechnologiesaredevelopedandappliedintheUK.

1.1.Methodology&Sources

Thestudyhasbeendesignedtoprovideinsightintothefollowingsetofcoreresearchquestions:

?HowmuchdoestheUK’sAISectorcontributetotheUKeconomy,includingrevenue,employment,GrossValueAdded(GVA),exportsandR&Dspending?

?WhatisthecompositionoftheUK’sAIsector,intermsofbusinesssize,location,andproductoffering?

?Whathavebeenthedriversofgrowthinthemarket,andwhatarethekeyupcomingchallenges?

Itisanticipatedthattheresearchwillbereplicatedinsubsequentyearsandassuch,themethodologyfordatacollectionandanalysisiswhollytransparentandrepeatable.

1.2.Approach

Thestudyusesamixedmethodsapproach,combiningacademia,policyandinvestmentspheres.Keymethodologicalstepsaresummarisedbelow,withfullerdetailprovidedinappendicestothereport.

Stage1–Collationofinitialdatainputs:along-listofAIcompaniesdeemedtobepotentiallywithinthescopeofthestudywasidentifiedfromnumeroussources,predominantlyviawebintelligencegeneratedbyGlass.ai’sweb-readingcapabilities.JustunderonethirdofcompanieswerealsoidentifiedviaothersourcesincludingbutnotlimitedtoBureauvanDijk’sFAME,Beauhurst,Crunchbase,LightcastandFDIMarkets.

Stage2–Initialclassificationandfiltering:AsetofkeywordsandcategorieswereidentifiedthroughacombinationofautomatedclassificationusingGlass.ailanguagemodelsandworkshopsessionswithrepresentativesfromacademia,industry,governmentandthecore

studyteam.Thelong-listofpotentiallyin-scopefirmswasrefinedandfilteredtoprovideashortlistof3,170in-scopeAIcompanies.

Stage3–Surveydesignandadministration:adetailedbusinesssurveywasdesignedwithinputfromthestudysteeringgroup,includingrepresentativesfromDSITandacademicandcommercialresearchexpertise.Thesurveywasadministeredviamultiplechannels,includingviatelephone,e-mailandweb-hosting.Atotalof250responseswerereceived.

Stage4–Dataaugmentation:aseriesofmanualdataqualitycheckswereconductedacrosskeymetrics(revenue,employment,location,classification)byboththecorestudyteamandDSITanalysts.Companydatawasthenaugmentedusingmultipledatasources,providingaconsistentsetofkeymetricsforeachUKAIbusiness.

Figure1.1–Shortlisting&AugmentationOverview

Source:PerspectiveEconomics

Stage5–Regional&sub-sectoralanalysis:moregranulardataonthetradinglocationsofin-scopeAIcompanieswasgatheredthroughweb-intelligenceandproprietarydatasources,enablingamoredetailedanalysisofthetradingpresenceofUKAIcompanieslocally,andinternationally.

Stage6–Sectormodelling:Theshort-listedAIcompanysetwasusedtoproduceanalysesofthenumber,scaleandlocationofUKAIcompanies,incorporations,investment,R&Dexpenditureandexports.

Stage7–Qualitativeinterviews&casestudies:in-depthfollow-upinterviewswereconductedwith10AIcompaniesthatrespondedtothesurvey.Findingswerecombinedwiththosefrom10in-depthsemi-structuredstrategicstakeholderinterviewstoaddressqualitativeresearchquestionsregardingstrengths,weaknesses,opportunities,challengesandriskstotheUKAIsector.

Stage8–Analysis&reporting:findingsfromthequantitativeandqualitativeresearchweresynthesisedthroughsteeringgroupdiscussionsandqualitativeanalysissessionsandtriangulatedtoinformthisbaselinereport.

1.3.InterpretationofData

ArtificialIntelligenceactivityintheUKisnotdefinedbyaformalStandardIndustrialClassification(SIC)code1.ThisstudythereforeusesexperimentalmethodstoidentifyandquantifyAIactivityacrosstraditionaleconomicsectors.Theapproachandmethodologyare

1SICcodesarethecurrentsystemofclassifyingbusinessestablishmentsandotherstatisticalunitsbytypeofeconomicactivityinwhichtheyareengaged.

consistentwiththoseemployedtodeliveranalysesoftheUKcybersecuritysectorannuallysince20182.Thedatausedtoinformthestudyincludes:

?IdentificationofAIfirmsaccordingtoanagreedtaxonomyusingAIdrivenlanguagemodelsappliedacrosswebsites,news,socialmedia,academicandofficialsources.

?EnrichmentofwebdatausingopenandproprietarydatasourcesincludingCompaniesHouse(companyname,registrationnumber,locations,incorporationdate),BureauvanDijkFAME(revenue,employment,profitability,remuneration,R&Dspend)andBeauhurst(externalgrants,fundraisings,acceleratorattendance,M&Aactivity).

Acrossthisreport,percentagesfromthequantitativedatamaynotaddto100%duetoroundingand/ortheoptiontoselectmultipleresponsestocertainsurveyquestions.ItisalsoimportanttonotethatthesurveydataisbasedonasampleofAIcompaniesandarethereforesubjecttosamplingtolerances.Theoverallmarginoferrorforthesampleof250AIcompanies(withinapopulationof3,170companies)isbetweenc.3andc.6percentagepointsata95%confidencelevel.Thelowerendofthisrange(3percentagepoints)isusedforsurveyestimatescloserto10%or90%.Thehigherend(6percentagepoints)isusedforsurveyestimatesaround50%.Datafromthe22qualitativeconsultationsisintendedtobeillustrativeofthekeythemesaffectingAIactivityintheUKgenerally,ratherthanastatisticallyrepresentativeviewofAIsectorbusinessesorinvestors.

1.4.Acknowledgements

TheauthorswouldliketothanktheDSITteamfortheirsupportacrossthestudy.DSITandthereportauthorswouldalsoliketothankallthosewhocontributedtotheresearch,includingthosewhotookpartinin-depthstrategicstakeholderinterviews,respondedtothebusinesssurvey,orotherwiseofferedintelligenceandinsightstothestudy.

Note:Thisreportusesexperimentalmethodstodefine,scopeandmeasurethescale

oftheUK’sAIsector.Wethereforewelcomecommentsandfeedbackregardingthe

methodologyorfindingsherein,throughcontacting

digital-analysis-team@.uk

.

2DSIT(2022)CyberSecuritySectoralAnalysis2022,accessibleat[.uk/government/publications/cyber-security-sectoral-analysis-2022]

2.UKArtificialIntelligenceSectorProfile

TheNationalAIStrategydescribesArtificialIntelligence(AI)asthe“fastestgrowingdeeptechnologyintheworld,withhugepotentialtorewritetherulesofentireindustries,drivesubstantialeconomicgrowthandtransformallareasoflife”3.Recognisingchallenges,limitationsandquestionablevalueoftryingtotightlydefineAI,theAIregulationpolicypaper–Establishingapro-innovationapproachtoregulatingAI4–describesAIas“ageneral-purposetechnologylikeelectricity,theinternetandthecombustionengine.”ItdefinesthecorecharacteristicsofAIasthe‘a(chǎn)daptiveness’and‘a(chǎn)utonomy’ofthetechnologyi.e.,thatAItechnologycanoperateonthebasisofinstructionswhichhavebeenlearntratherthanprogrammed,andthatcanbeautonomouslyappliedwithindynamicandfast-movingenvironments.

2.1.DefiningtheUKArtificialIntelligenceSector

TheanalysescontainedinthisreportarebasedonacommerciallyorientedtaxonomyofAIactivityintheUK.The‘commerciallyoriented’distinctionismadegiventhecommercialnatureofthelanguageusedtoinformthisstudy(drawnfromwebandtrade-baseddescriptionsofcompanyactivity),vis-à-vismoretechnicalterminologythatiscurrentlybeingusedinparallelactivitytobetterunderstandresearch-relatedtechnologicalAIdevelopments.Asdiscussedfurtheroverleaf,thestudysegmentscompaniesaccordingtoanagreedtaxonomy,includingadelineationbetween‘dedicated’and‘diversified’AIcompanies.Table2.1providesanillustrationofsomeofthemostprominentdedicatedanddiversifiedAIcompaniesidentified.

Table2.1–KeyAISectorContributors–Dedicated&Diversified

Dedicated

Diversified

1DeepMind

1FacebookUK

2

LimeJump

2

IBMUK

3LoopMe

3Microsoft

4

Peak

4

GoogleUK

5Ivefi.ai

5Accenture

6

Lendable

6

Amazon

7Deloitte

7EquippedAI

8

Improbable

8

Vodafone

9Cognizant

9Exscientia

10

Tractable

10

BT

Source:Glass.ai,PerspectiveEconomics

3DSIT(2021)NationalAIStrategy,DepartmentforScienceInnovation&Technology.

4.uk/government/publications/establishing-a-pro-innovation-approach-to-regulating-ai/establishing-a-pro-innovation-approach-to-regulating-ai-policy-statement

ThetaxonomyusedtodescribeAIactivityinthisstudyisillustratedinFigure2.1.SalientpointstonoteregardingthetaxonomyarediscussedbelowFigure2.1,andthefulltaxonomyisalsoavailabletoviewintheappendicestothisreport.

Figure2.1–UKAITaxonomy

Source:PerspectiveEconomics

Foreaseofreference,salientpointsregardingthesectortaxonomyinclude:

?Pre-requisitesforinclusion:TobeincludedinthestudycompaniesmustberegisteredandhaveanactivepresenceintheUK.

?DedicatedvsDiversifiedAIcompanies:atthehighestlevel,thetaxonomysegmentsthebusinesspopulationaccordingtowhethertheyareadedicatedAIcompany,orwhetherAIactivitymakesupasmallerproportionofamuchbroadercommercialbusinessoffering.DedicatedAIcompaniesareconsideredtobebusinessesthatprovideaproprietaryAItechnicalservice,product,platformorhardwareastheirprimaryrevenuesource.

?AIBusinessModel:atalowerlevelthetaxonomysegmentsbetweencreatorsofAIinfrastructure5,developersofAIproducts6andAIserviceproviders7.AdoptersofAIproductsorservicesdevelopedbyothersareconsideredtobeoutsidethescopeofthisstudytoavoiddoublecountingandtohelpensurethattheanalysisispredominantlyfocussedonvalueaddedtotheUKeconomybyAIsectoractivity.

5Includinghardware,frameworks,software,librariesandplatforms.

6Companiesproducingbespoke,valueaddingAIsolutionsmarketedandsoldasproducts.

7CompaniesofferingskillsandexpertisetosupporttheadoptionofAIproducts.

?AICapabilities:theanalysescontainedinthereportsegmentAIsectoractivityaccordingtothemaintechnologicalcapabilitythatunderpinsbusinessmodels.WhilemanyofthecompaniesidentifiedemploymultipleAIcapabilities,languagemodelswereadjustedtoidentifyboththeforemostAIcapability,aswellasallothercapabilitiesmentioned.MachineLearningisagenerictermthatunderpinsallothercapabilities.Itisincludedasacategoryherebecauseinmanyinstancesdescriptivecompanyinformation(thebasisofclassification)doesnotfurtherspecifytechnicalcapabilities.

?Industries:tosupportcomparativeanalyseswithSICbasedeconomicdataeachcompanyisalsoassignedtoasingleindustrywhichisderivedfromandcanbemappedbacktoSICCodes.

Inaddition,eachin-scopecompanyhasbeenclassifiedintoindustrysectorsusingGlass.ai’sproprietarytopicontologies.ThemostprominentindustrysectorsreferredtoinSection3arelistedbelowandasummaryofcompaniesassignedtobothGlass.aisectorsandStandardIndustrialClassification(SIC)codesareavailableintheappendices.

?ComputerSoftware

?InformationTechnologyandServices

?Biotechnology,LifeSciences&Pharma

?FinancialServices

?ProfessionalServices

?WiderHealth&MedicalPractice

?R&DandScientific

?Automotive,IndustrialAutomation&Machinery

?Energy,Utilities&Renewables

?AgriculturalTechnology

2.2.NumberofUKAICompanies

BasedonacombinationofAIdrivenwebintelligence,andcollationofcompanydatafromnumerousopenandproprietarysourcesincludingCompaniesHouse,BureauvanDijk,BeauhurstandLightcast,weestimatethattherearecurrently3,170activecompaniesintheUKprovidingAIinfrastructures,productsandservices.Aspreviouslystated,thisfocussesspecificallyonvalue-addedbytheAIsectoranddoesnotthereforeincludethewidervalueaddedbyadoptionofAItechnologiesacrossothersectors.

2.2.1.RegisteredCompaniesbySize

Ninety-sixpercentofthecompanies

identifiedareSMEs;60%ofall

companiesaremicrobusinesses

(Figure2.2).

Consultationwithstrategic

stakeholdersfromacrossindustry,

academiaandpolicyspheres

pointedtothepresenceofa

significantnumberoflarge

technologyfirmsasakeystrengthof

theUK’sAIecosystem,deemedto

beatleastinpartduetotheUK’s

reputationforhighqualityscientific

researchandinnovation.This

assertionissupportedbya

comparisonofthesizeofcompaniesin

theAIsectorvis-à-visthebroaderUKbusinesspopulation8(Table2.1).ThetablebelowevidencesthattheAIsectorhasagreaterconcentrationoflarge,mediumandsmallbusinessesthanthegeneralUKBusinesspopulation.

Table2.1–AISizeProfileComparison

Size

UKBusiness

Population

Estimates(2022)

Percentage

AISectoral

Analysis

Percentage

Large(250+

employees)

7,675

<1%

132

4%

Medium(50-249)

35,940

3%

262

8%

Small(10-49)

217,240

15%

887

28%

Micro(1-9)

1,187,045

82%

1,889

60%

AllBusinesseswithatleast1employee

1,447,900

100%

3,170

100%

Source:ONS,Glass.ai

8UKBusinessPopulationEstimates(2022):Availableat:

.uk/government/statistics/business-population-

estimates-2022

2.2.2.Dedicated&DiversifiedAICompanies

Ofthe3,170activecompaniesFigure2.3–DedicatedandDiversifiedAICompanies

identifiedthroughthestudy60%are

dedicatedAIbusinessesand40%are

diversified(i.e.,haveAIactivityas

partofabroaderdiversifiedproduct

orserviceoffer,Figure2.3).

Incomparisontoothersimilarstudies

theproportionofdiversified

companieswithintheAIsectoris

higher.Thisisindicativeofthe

comparativelybroadscopeforAI

technologyapplicationsacross

sectors,andpointstoanintense

focusondevelopmentofAI

technologyamongbothdedicated

companies(e.g.,DeepMind,Source:Glass.ai,PerspectiveEconomics(n=3,170)

Improbable,BenevolentAI)and

established,diversifiedtechnologycompanieswithmuchbroaderserviceoffers(e.g.,Amazon,Google,Microsoft,IBM)9.

Figure2.4overleafshowsthatmostlargeAIcompaniesarediversified(89%,n=118),whereasthemajorityofmicro-AIcompaniesarededicated,meaningthatAIiscoretotheirbusinessmodel(68%,n=1,288).

9Itisworthnotingherethat,giventhebreadthandvaryingscaleofAIactivity,itisnotpossibletodelineatededicatedanddiversifiedAIfirmssolelyonthebasisoftheproportionofAIrelatedrevenueoremploymentwithincompanies.CompanieswithrelativelysmallAIteamscanbededicatedAIcompaniesandbythesametoken,companieswithlargeAIteamscanbediversified.Thereforeinstead,thestudyusedacombinationofdataonAIrelatedemploymentandadetailedmanualreviewofcompanydescriptionsasthebasisoffinaldecisionsonwhetherornotacompanyfallsintothededicatedordiversifiedcategory.

Figure2.4–AICompanySize

Source:Glass.ai,PerspectiveEconomics(n=3,170)

2.2.3.AICompanyRegistrations

AnalysisofincorporationdatesacrossthepopulationofAIcompaniesshowssignificantgrowthinAIcompanyregistrationssince2011.Onaverage,269newAIcompanieshavebeenregisteredeachyearsince2011,withapeakinnewcompanyregistrationsinthesameyearastheAISectorDeal(2018,n=429)andsmallernumbersofnewcompanyregistrationssince(Figure2.5overleaf)10.

10Analysisexcludes2022duetodatagapsassociatedwiththenormallaginavailabilityofcompanydata.

Figure2.5–AICompanyRegistrations

Source:PerspectiveEconomics,Glass.ai,CompaniesHouse(1998–2021|n=3,030

companiesincorporatedsince1998)

2.2.4.PredictedAIBusinessModel

ThetaxonomycanbeusedtobetterunderstandtheprofileoftheAIsectoraccordingtothebroadfocusofAIactivity(i.e.,infrastructure,productsorservices)andatalower-level,categorisationofthecorecapabilityofin-scopecompanies.Figure2.6overleafpresentsthetwomaintaxonomylevelsasanexcerptforeaseofreference.Analysesthatfollowfocusonthebusinessmodelsandcapabilitiesofcompaniesincludedinthefinaldataset.EachcompanyisassignedtoasinglebusinessmodelandcapabilitybasedonthehighestprobablecategorisationusingthelanguagemodelsdevelopedbyGlass.ai.

Figure2.6–AIBusinessModels&Capabilities

Source:Glass.ai,TaxonomyWorkshopOutputs

Acrosstheentirepopulation82%ofcompaniesfallwithinthebusinessmodelcategoriesofAIproductsandinfrastructures(72%and11%respectively),withtheremaining18%engagedpredominantlyinprovidingAI-relatedservices11.AgreaterproportionofdedicatedAIcompaniesprimarilyproduceAIrelatedproducts(75%ofdedicatedcompaniescomparedto66%ofdiversifiedcompanies).Together,thedataoncompanysizeandbusinessmodelsuggestthatdedicatedAIcompaniesarebothsmallerandmoredependentonthesuccessoftheAIproductsthe