追蹤美國(guó)人工智能并購(gòu)案

CenterforSecurityandEmergingTechnology|1

ExecutiveSummary

Thecommercialartificialintelligenceindustryisevolvingrapidly,andthecompetitiondynamicsinthisburgeoningsectorwillimpacttherate,diversity,anddirectionofAIinnovationintheyearsahead.MaintainingU.S.technologicalleadershipintheyearsaheadwillrequirepolicymakerstopromotecompetitionintheAIsectorandpreventincumbentfirmsfromwieldingtheirmarketpowerinharmfulways.

OneimportantcomponentofthiseffortwillbemonitoringmergersandacquisitionsactivityintheAIsector.M&Aallowscompaniestogainaccesstotalent,technologies,andotherresourcesthatmayotherwisebeoutoftheirreachortoodifficulttodevelopin-house.Thesetransactionscanallowfirmstomaintaintheirtechnologicaledge,gaineconomiesofscale,andexpandtheirbusiness,allofwhichcandrivegrowthand

promotethehealthyfunctioningofamarketeconomy.Ontheflipside,however,M&Acanalsoenablecompaniestoentrenchtheireconomicpower,reduceincumbentfirms’incentivestoinvestininnovation,andhampertheabilityofnewdisruptivefirmsto

enterthemarket.

ThisbriefseekstoshedlightonmajortrendsinM&AactivityintheU.S.AIsectorbetween2014and2023.Ouranalysisisbasedonadatasetof4,354M&A

transactionsgatheredthroughPitchBook,athird-partyproviderofcorporatefinancialinformation.Wefound:

1.AnnualM&AtransactionsinvolvingAIcompaniesmorethandoubledoverthelastdecade,from225in2014to494in2023.However,M&Atransactionshavedeclinedsincetheirpeakin2021(828).

2.TheproportionoftotalM&Atransactionsinwhichnon-AIcompaniesacquiredAIcompaniesgrewfrom10percentin2014to45percentin2023.Still,themajorityoftheseacquisitionswereconductedbyothercompaniesinthe

technologyindustryratherthanfirmsinothersectors.

3.LargeincumbenttechnologycompaniesrankamongthetopacquirersofAI

companies,includingApple(28transactions),Alphabet(23),Microsoft(18),andMeta(16).However,theoverallAIM&Aactivityremainsfairlydiffuse,with

1,446uniqueacquirersengaginginAIM&Atransactionsoverthepastdecade.

4.InU.S.cross-borderAIacquisitions,Americanfirmshavepurchased503foreignAIcompanies,whileforeignfirmshavebought271AmericanAIcompanies.U.S.firmsmostfrequentlyacquiredAIfirmsbasedintheUnitedKingdomand

CenterforSecurityandEmergingTechnology|2

Canada.FirmsintheUnitedKingdomandCanadawerealsothemostfrequentforeignacquirersofU.S.AIcompanies.

ThecommercialAIsectorisstillinitsinfancy,andthedynamicsofmarketcompetitiontodaycanhavemajoreffectsonthetrajectoryofAIinnovationtomorrow.ContinuingtomonitorthelandscapeofM&Atransactionsathomeandabroad—particularlythose

involvingincumbenttechnologycompanies—willbecrucialtopromotinganinnovativeanddynamicAIecosystemintheyearsahead.Additionally,itwillbeimportantfor

competitionauthoritiestoupdatetheirprocessesandprocedurestoappropriatelyscrutinizealternativebusinessarrangementssuchas“partnerships”between

incumbentfirmsandAIstartups,whichfunctionsimilarlytotraditionalacquisitionsbutappeartoavoidregulatoryframeworks.

CenterforSecurityandEmergingTechnology|3

TableofContents

ExecutiveSummary 1

TableofContents 3

Introduction 4

Methodology 6

Findings 8

TopAcquirersofAICompanies 12

U.S.Cross-BorderAIAcquisitions 16

KeyTakeaways 19

Authors 21

Acknowledgments 21

Appendix 22

Appendix1:SelectionofAI-RelevantAcquiringPartiesandCompanies 22

Appendix2:TypesofAIMergersandAcquisitions 23

Appendix3:CountofAnnualAIAcquisitionsbyDealType,2014–2023 24

Appendix4:IndustrySectorsofNon-AIAcquirersofAICompanies 24

Endnotes 25

CenterforSecurityandEmergingTechnology|4

Introduction

Inthefallof2010,threeLondon-basedtechnologistsandentrepreneursfoundeda

companywiththeambitiousgoalofbuildingtheworld’sfirstartificialgeneral

intelligence(AGI).Atthetime,AIsystemswerehighlyspecialized,trainedtoperformdiscretetaskswithinnarrowlydefinedapplicationareas.Byemployingrecentadvancesindeepneuralnetworksandreinforcementlearning,thetriobelievedtheycouldteachmachinestothinkmorelikehumans,gatheringknowledgefromdisparatesourcesandapplyingitacrossawiderangeofdomains

.1

ThefoundersenvisionedAIsystemsthatwouldonedayhelptosolvesomeoftheworld’smostpressingscientificandsocietalproblems.

2

Withinafewyears,thecompany,namedDeepMind,madesomemajorstrides.Mostnotably,itsresearchersdevelopedanAIsystemthattaughtitselftoplaysevenclassicAtarigames.Inafewofthese,includingPongandBreakout,theAIevenoutperformedhumanexperts

.3

ThoughDeepMindhadyettoreleaseaproductormonetizeitsAI-

poweredgamingsoftware,itraisedmorethan$26millionbythespringof2012

.4

In2014,seeingthepromiseinitsfledglingAIsystems,Google(nowAlphabet)acquiredthestartupforathen-whopping$650million

.5

Overthenextdecade,DeepMindhelpedcementAlphabet’sreputationasanAI

pioneer,buildingsystemsthatcouldpredictthestructureofproteinsandbestthe

world’stopplayersofchess,Go,andshogi

.6

AstheracetodevelopgenerativeAI

heatedupinSiliconValley,AlphabetmergedDeepMindwithitsotherin-houseAI

developmentteam,Brain,toformGoogleDeepMind

.7

Today,thisconsolidatedentityistheepicenterofthecompany’sAIefforts,producingGemini,Gemma,andothermodelsthathavemadeAlphabetatopcompetitorintherapidlyexpandingAIindustry.Itis

impossibletoknowwhatmighthavehappenedhadAlphabetnotgonethroughwithitsacquisitionofDeepMind,butitisdifficulttoimaginethetechgiantbeingbetteroffhaditnotacquiredtheBritishstartupadecadeago.

Alphabet’ssuccessfollowingtheacquisitionofDeepMindunderscoresthecriticalrolethatmergersandacquisitionscanplayinshapingthetrajectoryofanindustry.M&A

allowscompaniestogainaccesstotalent,technologies,andotherresourcesthat

otherwisemaybeoutoftheirreach.Thesetransactionscanallowfirmstomaintain

theirtechnologicaledge,gaineconomiesofscale,andexpandtheirbusiness,allof

whichcandrivegrowthandpromotethehealthyfunctioningofamarketeconomy

.8

Ontheflipside,however,M&Acanenablecompaniestoentrenchtheireconomicpower,reduceincumbentfirms’incentivestoinvestininnovation,andhampertheabilityof

newdisruptivefirmstoenterthemarket

.9

CenterforSecurityandEmergingTechnology|5

M&Atransactionscanalsoplayaroleinshapingthelandscapeofinternational

technologycompetition.TheUKgovernmenthasnotedthatGoogle’sacquisitionof

London-basedDeepMindmayhavesetbackthecountry’sdomesticAIsector,andithassubsequentlyincreaseditsscrutinyofnationalsecurity-relevantbusiness

transactions

.10

TheCommitteeonForeignInvestmentintheUnitedStates(CFIUS)hassimilarlyintensifieditsreviewsofcross-bordertransactionsinvolvingcountriesaimingtoacquiretechnologiesrelevanttonationalsecurity,includingAI,biotechnology,andquantumcomputing

.11

AsAIsystemsbecomeincreasinglyintegratedacrosstheglobaleconomy,analyzing

M&AactivityintheAIindustrywillbecrucialformaintainingU.S.technological

leadership,monitoringAIadoptionacrosssectors,andpromotingacompetitive,

diversified,andresilientAIecosystem

.12

Thefederalgovernment’srecentinvestigationsintoAIindustrypartnershipsandrevisionsofmergerguidelineshaverenewed

discussionsabouttherolethatM&Aplaysinshapingemergingtechnologysectors

.13

WehopetoinformthosediscussionsbyexaminingbroadtrendsinM&Aactivity

involvingAIcompaniesoverthelastdecade.

WebeginbyexploringhowthenumberofannualAIM&Atransactionshaschangedovertime,aswellasthetypesofcompaniesinvolvedinthosedeals.WethenexaminethetopacquirersofAIcompaniesandcross-borderM&Atransactionsinvolvingthe

U.S.andforeignAImarkets.WeconcludewithbroadtakeawaysforpolicymakerslookingtopromoteadynamicandcompetitiveAImarketintheyearsahead.

Methodology

ThisresearchaimstoshedlightontheAI-relatedM&AactivityinvolvingU.S.-

headquarteredbuyersortargetsbetween2014and2023.OuranalysisleveragesdatafromPitchBook,whichprovidescorporateandfinancialinformationonpublicly-tradedandprivately-heldcompanies

.14

UsingPitchBookdata,wedeterminethelocationof

buyers’andacquiredcompanies’headquarters,maptherelationshipsbetween

corporatesubsidiariesandtheirparententities,andidentifythebuyersandtargetsinvolvedintheAIsector.WeincludeM&Atransactionsintheanalysisifeitherthebuyerorthetargetare:

1.AssignedtoPitchBook’s“ArtificialIntelligence&MachineLearning”verticalorincludesanAI-relatedkeywordintheirdescription(seeAppendix1formoredetails)and;

2.HeadquarteredintheUnitedStates

Toensuredealsareassignedappropriately,subsidiariesreceivedtheheadquarterslocationsoftheirparententities.IfthesubsidiarywasnotAI-relevantbuttheparentcompanywas,thentheparent’sdesignationwasused.Lastly,weexcludeM&Adealsinvolvingmorethanoneacquirer,aswellasdealsinwhichlessthan100%ofthe

targetcompanywasacquired

.15

Wemanuallyreviewedarandomsample(n=200)oftheresultingsetofdealstoensurerelevancetoouranalyticalgoals

.16

Usingthismethodology,weconstructedadatasetthatincludes4,586U.S.AIM&Atransactionsconductedoverthepasttenyears

.17

ThisdatasetencompassesthreetypesofM&Atransactions,whichwerefertocollectivelyas“AIacquisitions”

:*

1.AIBuyer/AITarget:Transactionsinwhichtheacquirer(oritsparententity)isdesignatedasAI-relevant,andthecompanybeingacquiredisalsodesignatedasAI-relevant.

2.AIBuyer/Non-AITarget:Transactionsinwhichtheacquirer(oritsparententity)isdesignatedasAI-relevant,butthecompanybeingacquiredisnotdesignatedasAI-relevant.

*SeeAppendix2formoreinformationonthethreetypesofM&Atransactions.

CenterforSecurityandEmergingTechnology|6

CenterforSecurityandEmergingTechnology|7

3.Non-AIBuyer/AITarget:Transactionsinwhichtheacquirer(oritsparententity)isnotdesignatedasAI-relevant,butthecompanybeingacquiredisdesignatedasAI-relevant.

OuranalysisreliesonPitchBook,whichcollectsandverifiesindustryassignments,

headquarterdata,andparent-subsidiaryrelationshipsfromavarietyofsources.

However,likealldatasources,PitchBookisimperfectandhasanumberoflimitations.

First,PitchBookisunlikelytocaptureallrelevantAIbuyersandtargetsgiventhattheAIindustryisevolvingquicklyasnewtechnologiesareintroduced,newcompaniesadoptAIsystems,andnewinvestorsenterthemarket.Similarly,theremaybeM&AdealsthatwerenotpubliclyreportedonandthusnotincludedwithinPitchBook.Furthermore,the“Verticals”assignmentandrecordedkeywordsaresubjecttoPitchBook’scollection

process,whichmayfavorcompaniesthatemphasizetheirAIcapabilities.

WevalidatedandmanuallyreviewedrandomsamplestoensuretheAI-relevanceofpartiesandthedetailsofthedealsarereliable.Despitetheselimitations,thisreportprovidesusefulsnapshotinformationabouttheU.S.AIM&Alandscapetofurther

informthecompetitivenesspolicyinthenearfuture.

CenterforSecurityandEmergingTechnology|8

Findings

OuranalysisfoundthatoverallM&AactivityinvolvingU.S.AIcompanieshastrendedupwardoverthelastdecade.ThenumberofannualAIM&Atransactionsmorethan

doubled,risingfrom247in2014to523in2023.AsFigure1shows,thenumberof

annualM&AtransactionsinvolvingU.S.AIcompaniesgrewsteadilybetween2014and2019,plateauedin2020,andthenspikedsharplyin2021,with871totaldeals

markingan83percentincreasefromtheprioryear.Whilethenumberoftransactionshasfallensincethen,overallM&Aactivityin2023wasstillsomewhathigherthan

beforethe2021spike.

Figure1.AnnualAIM&ATransactionsInvolvingU.S.Companies,2014-2023

Source:CSETanalysisofdataprovidedbyPitchBookData,Inc.

ThereareavarietyoffactorsthatcouldhavecontributedtothesetrendsinAIM&A

transactions.ThesteadygrowthinthenumberofM&Adealsinthelatterhalfofthe

2010scoincidedwithanexpansionoftheoverallU.S.AIsector

.18

AsshowninFigure2,thenumberofnewAIcompaniesformedintheUnitedStateseachyeargrew75

percentbetween2014and2017,andcontinuedhoveringatthiselevatedlevelbefore

spikingin2023amidthegenerativeAIboom

.*

GlobalinvestmentintheAIindustryalsogrewsignificantlyduringthissameperiod,thoughwewereunabletodeterminethe

exactnumberofnewglobalfirmsbasedonourdata

.19

ItisperhapsunsurprisingthatweobservedanincreaseinAIM&Atransactionsasmorecompaniesenteredthesectorandthetechnologyunderlyingtheirproductsimproved.

Additionally,thespikeandsubsequentdrop-offinAIacquisitionsbetween2020and2023generallytracksbroadereconomy-widetrendsinM&Aduringthatsameperiod.AnalystsatBostonConsultingGroupfoundvirtuallyeveryindustryexperienceda

notableincreaseinthenumberofM&Atransactionsbetween2020and2021,followedbyaslightdeclinein2022

.20

WhilethetrendwasmorepronouncedinouranalysisofAIacquisitions,itisnotablethattheAIsectorfollowedsimilarpatternsasother

industries.Thereareanumberofpotentialexplanationsfortheseeconomy-wide

fluctuationsinM&A,includingchanginginterestratesandeconomicdisruptionsrelatedtotheCOVID-19pandemic,thoughdeterminingtheexactcausesisbeyondthescopeofouranalysis.

Figure2.AnnualU.S.AICompanyFormation,2014-2023

Source:CSETanalysisofdataprovidedbyPitchBookData,Inc.

*ItisimportanttorecognizethatFigure2doesnotreflecttheoverallsizeoftheU.S.AImarket(e.g.,itdoesnotincludecompanyexits).Evenso,giventheprevalenceofU.S.firmsintheAIsector,U.S.

companyformationisausefulproxyforoverallmarketsize.

CenterforSecurityandEmergingTechnology|9

CenterforSecurityandEmergingTechnology|10

BeyondtheincreaseintotalAIM&Aactivity,wealsoobservedthatthetypesofcompaniesparticipatingintransactionschangedovertime.Asdiscussedinthe

previoussection,ouranalysisencompassesthreedifferenttypesofAIM&A

transactions:AIacquirersbuyingAIcompanies,non-AIacquirersbuyingAIcompanies,andAIacquirersbuyingnon-AIcompanies.Differentfactorsarelikelytomotivateeachtypeoftransaction.Aswithotherintra-industryM&Atransactions,AIcompaniesmayacquireotherAIcompaniestogainaccesstonewtalent,eliminatecompetitors,expandexistingAIproductsandservices,orforavarietyofotherreasons.Non-AIcompaniesmayacquireAIcompaniesasawaytoquicklyadoptAIcapabilitieswithoutbuilding

themfromscratch.AIcompaniesmaypurchasenon-AIcompaniesinordertoaugmentordiversifytheirexistingAIofferings,andpotentiallytoaccessnewdistribution

channelsforthoseproducts

.*

OneexampleisAlphabet’s(anAIcompany)acquisitionofthehealthtechfirmFitBit(anon-AIcompany)in2021,whichallowedAlphabettooffermoreAIproductstailoredtohealthcare.AIfirmsmayalsopurchasenon-AIfirmsto

supportnon-AIlinesofbusiness.Uber’s(anAIcompany)acquisitionofPostmates(anon-AIcompany),whichexpandedtheridesharecompany’sfooddeliverybusiness,likelyfallsintothiscategory

.21

Ouranalysisfoundthatoverthelastdecade,themostcommontypeofAIacquisitioninvolvedAIcompaniesbuyingnon-AIcompanies.Thiscategoryaccountedfor56

percentoftheM&Atransactionsinourdataset(2,576).Roughly32percentofthetotaltransactionsinvolvednon-AIbuyersacquiringAIcompanies(1,464),andjust12

percententailedAIcompaniesbuyingotherAIcompanies(546).

However,wealsoobservedthattheannualdistributionofdealtypeschanged

dramaticallyovertime.AsshowninFigure3,in2014onlyabout10percentofM&Atransactionsinourdatasetinvolvednon-AIcompaniesbuyingAIcompanies,butby2023thiscategoryaccountedfornearly45percentoftransactions.Duringthatsameperiod,theshareoftransactionsinwhichAIcompaniesacquiredAIcompaniesgrew

*ManyAIcompaniesarelargeconglomeratesthatoperateinmultiplesectors.Additionally,firmsmay

alsoacquirecompaniesforreasonsthathavenothingtodowithgainingnewtechnologicalcapabilities.Industryincumbentsmayacquirestartupstoeliminatepotentialcompetitors,forexample,orprivate

equityinvestorsmayacquirefirmsinordertorestructureandflipthemforaprofitorfacilitatean

industryroll-up.Ourdatasetofferslittleinformationontherationalebehindspecificacquisitions—the

discussionhereisintendedtohighlightbroadtrendsandpotentialcapabilities-basedreasonswhy

differenttypesoffirmsmayengageinAIacquisitions.Formore,see:JamesChen,“Roll-upMerger:

Overview,BenefitsandExamples,”Investopedia,May19,2024,

/roll-up-

merger-definition-4683958.

slightlyfrom4percentto13percent,andtheshareinvolvingAIcompaniesbuyingnon-AIcompaniesshrunkfrom86percentto42percent

.*

Figure3.ShareofAnnualAIM&ATransactionsbyDealType,2014-2023

Source:CSETanalysisofdataprovidedbyPitchBookData,Inc.

Therapidriseintheproportionofdealsinvolvingnon-AIcompaniesacquiringAI

companiessuggeststhatfirmsacrosstheeconomyareincreasinglylookingtointegrateAIintotheiroperationsandcommercialofferings,andmanyareturningtoM&Ato

acquirethosecapabilitiesratherthandevelopthemin-house.Still,thistrendshouldnotbeoverstated.Overhalfofthenon-AIcompaniesthatacquireAIfirmsareinvolvedintheinformationtechnology(IT)sector,andwhilewedoobserveacquisitionsbyfirmsinothersectorssuchasfinance,healthcare,andconsumergoods,theyaccountfora

relativelysmallshareofoveralltransactions(seeAppendix4formoreinformation).This

*ThetotalnumberoftransactionsbetweenAIacquirersandnon-AIcompaniesgrewmodestlyduringthisperiod,justatamuchslowerratethantransactionsintheothertwocategories(seeAppendix2).

CenterforSecurityandEmergingTechnology|11

doesnotnecessarilymeanthatfirmsintheseothersectorsarefailingtoadoptAI,butratherthattheymaymoreoftenaccessAItoolsthroughexternalITvendors

.*

TherestofouranalyseswillfocusexclusivelyonacquisitionsofAIcompaniesand

excludetransactionsinwhichAIcompaniesacquirednon-AIcompanies.Zoominginon

acquisitionsofAIcompaniesallowsustomoreclearlyilluminatetrendsinthenascentAIindustryandtheirimplicationsformarketcompetitionandthegeopolitical

landscape.

TopAcquirersofAICompanies

OuranalysisrevealedthatmanyfirmshaveacquiredAIcompaniesoverthelast

decade,butonlyasmallnumberhaveengagedinasignificantamountofM&Aactivity.

Overall,wefoundthat1,446uniquecompaniesconducted2,010acquisitionsofAI

firmsbetween2014and2023.AsshowninFigure4,thevastmajority(82percent)oftheseacquirersengagedinasingleAIacquisition.Ofthe262companiesthat

conductedmultipleacquisitionsofAIfirms,only48engagedinmorethanthreetransactionsandjust12companiesparticipatedin10ormoretransactions.

Figure4.DistributionofAcquirersbyNumberofAIAcquisitions

Source:CSETanalysisofdataprovidedbyPitchBookData,Inc.

*Basedonourdataset,wewereunabletoanalyzethespecificAItechniquesusedandthedifferenttypesofproductsofferedbyvariouscompanies.

CenterforSecurityandEmergingTechnology|12

Fromacompetitionpolicyperspective,therearetwobroadtakeawaysfromthese

findings.First,theAIecosystemappearstoberelativelydiffuseinsofaraswesee

acquisitionsdistributedacrossalargenumberofcompanies.Atahighlevel,wedonotclearlyobservelargeacquirersattemptingto“roll-up”theAIindustrythroughthe

aggressiveM&Aactivitythathasplaguedthedefenseindustrialbaseandothersectorsoftheeconomy,andevenamidongoingacquisitions,thenumberofcompanies

participatingintheAImarkethasincreasedovertime(seeFigure2)

.22

Whilethese

findingssuggesttheoverallmarketforAIsystemsremainscompetitive,itisimportanttorememberthatAIisageneral-purposetechnologywithmanydifferentapplications,andcompetitiondynamicsmayvarywidelyacrossdifferentsubsetsoftheAIsector.Additionally,itisworthrecognizingthatincertaincasesasingleacquisition,suchasGoogle’spurchaseofDeepMind,canhaveanoutsizedimpactonthedynamicsofa

particularproductmarket.

Second,ouranalysisshowsthatwhiletheoverallAIM&Alandscapeisgenerally

diffuse,thereareasmallnumberoffirmswithademonstratedinterestinacquiringalargenumberofAIfirms.Thesemostprolificacquirersrepresentasmallshareof

overalltransactions,butasapparentnodesofeconomicandtechnologicalpower,theyhavethepotentialtoalterthedynamicsoftheAIecosystem

.23

AsshowninTable1,fourU.S.bigtechcompanies—Apple,Microsoft,Alphabet,and

Meta—rankamongthemostprolificacquirersofAIcompaniesoverthepastdecade

.*

Thefactthatthesecompaniestopthelistisperhapsunsurprising.Assomeofthemostvaluablecorporationsintheworld,thesefirmsmaintainlargetrovesofcapitalthatcanbeusedtoexpandtheirbusinesses,enteradjacentmarkets,andacquireotherfirms’

technologiesandtalent

.24

Allfourareconsideredleaderswithintoday’sAIindustry,anditisreasonabletoattributetheirsuccessatleastsomewhattotheircomparatively

aggressiveapproachtoAIM&Aoverthelastdecade.

*NotablyabsentfromthelistoftopacquirersisAmazon,whichconductedonly6acquisitionsofAIfirmsbetween2014and2023.

CenterforSecurityandEmergingTechnology|13

CenterforSecurityandEmergingTechnology|14

Table1:Acquirerswith10orMoreAcquisitionsofAICompanies,2014–2023

Acquirer

NumberofAIM&ADeals

Apple

30

Microsoft

21

Alphabet

18

Intel

17

CiscoSystems

14

MetaPlatforms

14

InternationalBusinessMachines(IBM)

14

Accenture

11

ServiceNow

11

DataRobot

10

Salesforce

10

Snap

10

Source:CSETanalysisofdataprovidedbyPitchBookData,Inc.

Note:U.S.bigtechcompaniesarebolded.

Whileacquisitionsbylargeincumbentsarenotinherentlyproblematic,theyoften

warrantadditionalscrutinyfrompolicymakers.AsnotedintheU.S.government’slatestmergerguidelines,incumbentfirmsmayusesuchacquisitionstoeliminatecompetitive

threatsandretaintheirdominantposition,potentiallyslowingtechnological

developmentintheprocess

.25

ShouldbigtechfirmsusemarketpowertopreventnewfirmsfromgainingafootholdintheburgeoningAImarket,eitherthroughM&Aorsomeothermeans,theycouldlimittherate,diversity,anddirectionofAIinnovation

.26

Indeed,manyofthecompanieslistedinTable1havealreadyfacedscrutinyfromauthoritiesintheUnitedStatesandabroadforengaginginharmfulacquisitionsandotheranti-

competitiveconduct

.27

Wewereunabletodeterminethemotivationandcompetitiveeffectsofthebigtechplatforms’pastacquisitionsofAIcompaniesbasedonourdataset.However,our

analysisdidrevealonenotabletrendinthefirms’approachtoM&A:theAIcompaniesacquiredbyApple,Microsoft,Alphabet,andMetatendedtobe“younger”thanthoseacquiredbyotherfirms.Specifically,bigtechplatformsacquiredAIcompaniesan

averageof4.8yearsafterthecompanywasfounded,whileotherfirmsacquiredAI

CenterforSecurityandEmergingTechnology|15

companiesanaverageof7.4yearsaftertheywerefounded.Somemayinterpretthisfindingassupportingtheclaimthatthebigtechcompanieslooktoacquireyoungfirmsbeforetheyposeacompetitivethreat

.28

AnalternativeexplanationwouldbethatbigtechfirmshaveabetterunderstandingofAIsystemsthanothercompanies,and

thereforetendtoacquirepromisingtechnologiesandstartupsearlierintheirlifecycle.Still,theeffectsofsuchdealsoncompetitionandinnovationwouldlikelybesimilarregardlessofthemotivesdrivingtheacquisition.

Regulatorshaverecentlystartedinvestigatingthebehavioroftheincumbent

technologyplatformsintheAImarket,includingtheinvestmentsthatMicrosoft,

Alphabet,andAmazonhavemadeinindependentAIlabssuchasOpenAIand

Anthropic

.29

Thoughfewdetailsonthese“partnerships”havebeenpubliclydisclosed,theyappeartoofferincumbentfirmsawaytopullstartupsundertheircorporate

umbrellawhileavoidingregulatoryscrutiny.InJanuary2023,forinstance,MicrosoftandOpenAIstruckadealinwhichthetechgiantwouldallocatethelab$10billionworthofcashandcloudcreditsinexchangeforanexclusivelicensetouseOpenAI’smodelsandasubstantialstakeinthestartup’sprofits

.30

ThedealprovidedMicrosoftwithmanyofthesamebenefitsitwouldhavegainedfromadirectacquisitionof

OpenAI,butbecauseitwasnottechnicallyanM&Adeal,itavoidedtriggeringagovernmentmergerreview.Thesepartnershipshavebecomeincreasingly

commonplaceacrosstheAIindustryinrecentyears,anditwillbecrucialforthefederalpolicymakerstoupdatetheirprocessesandprocedurestoappropriatelyscrutinizesuch“acquisition-like”investmentsaswellasmoretraditionalM&Adeals

.31

Thevertical

integrationofthebigtechcompaniesacrosstheAIsupplychainoffersthemotherpotentialopportunitiestoreduceoreliminatecompetitivethreats,thoughthese

channelsarebeyondthescopeofthisreport

.*

*Theverticalintegrationofthelargetechcompaniesanditseffectsisbeyondthescopeofthispaper,butitwillbeexploredinforthcomingCSETresearch:JackCorrigan,“AIMarketPowers”(CenterforSecurityandEmergingTechnology,