




版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
THE
COMPETITIVERELATIONSHIP
BETWEENCLOUD
COMPUTING
ANDGENERATIVE
AICHRISTOPHECARUGATICloudcomputingproviders
andgenerative
artficialintelligence
(GenAI)
providers
nurturea
close,
interdependent
relationship:
GenAI
providers
needcloudproviders
totrain,
runanddeploy
theirGenAI
solutions,
whilecloudproviders
seeGenAI
providers
asa
businessdriver
togrow
theirmarket
shares
incloudandrelated
markets,
suchasproductivitysoftware
orsearch
engines.
ecloud/GenAI
relationship
takesvarious
forms,
includingexclusiveandstrategic
partnerships,
especiallybetweenlargecloudproviders
andGenAIproviders
across
allparts
ofthecloudmarket,
includinginfrastructure,
platformsandsoftware.Competitionbenefitsandrisks
are
likelytoresult
from
therelationships.
Competitionbenefitsarise
from
increased
competitionandinnovation
inthecloudandGenAI
sectors.Risksrelate
topotential
concentrations
arising
from
thepartnerships
betweencloudandGenAI
providers,
andfrom
anticompetitive
practices,
includingdiscriminationinthesupplyofITequipment
by
dominant
ITproviders,
interoperability
obstaclestoswitching,useofbusiness-userdata,
self-preferencing
ofcloudservices
over
third
parties,
tyingandpure
bundling.Merger
control
andantitrustlaws
canaddress
someofthecompetitionrisks,
whilelaws,
includingtheEuropean
Union’s
DigitalMarkets
Act
andData
Act,
candealwithcompetitionissuesindigitalmarkets
andthecloudsector.
Nevertheless
there
are
gaps.
eEuropean
Commissionshouldamendexisting
EU
instruments,
includingby
changing
thedefinitionofa
concentration
undermergercontrol,
andshouldspecifyinteroperabilityrequirements
forcloudproviders
undertheData
Act.
e
Commissionshouldalsocloselymonitordevelopments
inandoutsideEurope
through
market
investigations,
includingwithinternational
counterparts,
andshouldintervene
totackleimminent
competitionrisks
usingfastprocedural
tools,
suchasinterimmeasures.ChristopheCarugati(christophe.carugati@)isanAffiliateFellowatBruegelRecommended
citation:Carugati,
C.
(2023)‘ecompetitiverelationshipbetweencloudcomputingandgenerativeAI’,Working
Paper
19/2023,Bruegel1
IntroductionCloudcomputingandgenerativeartificialintelligence(GenAI)arecentraltotheeconomy'sdigitaltransformation1.Viathecloud,infrastructure,platformsandservicesareaccessibleonline,includingGenAIapplications,suchastheOpenAI-ownednaturalconversationallanguageapplicationChatGPT.GenAIapplicationsusenaturallanguageprocessing,suchastextsummarisation,togenerateoutputfrominputdata.Thecloudsectorwasworthanestimated€84.76billioninEuropein2022,withtheexpectationthatthiswilldoubleto€175.87billionby20272.ThesametrendofveryrapidgrowthcanbeseenintheGenAIsector.Itwasworthanestimated$6.33billioninEuropein2022andexpectssixfoldgrowthto$35.94billionin20273.CloudprovidersandGenAIprovidersrelyononeanother.GenAIprovidersneedcloudinfrastructurestotrain,runanddeploytheirapplications,whilecloudprovidersseeGenAIasabusinessdrivertogrowtheirmarketsizeinthecloudandotherrelatedmarkets,includingproductivitysoftware,searchengines,webbrowsers,e-commerceandadvertising.AsdemandsoarsforcloudcomputingandGenAI,competitionintensifies.CloudproviderscompetebyadaptingtheirinfrastructuresandservicestoGenAIneeds.Atthesametime,GenAIproviderscompetebyfrequentlyreleasingmodelsandapplications.Competitionrisksarealreadypresent.Competitionauthoritiesworldwidehavehighlightedatrendinthecloudsectortowardsconcentrationinthehandsofafewfirms,includingAmazon,MicrosoftandGoogle4.Thesefirmsaredeemed‘hyperscalers’bycompetitionauthoritiesandtheindustryastheyenjoylargeeconomiesofscaleinhostingcapabilitiesandservices,andeconomiesofscopeintheprovisionofanarrayofservices.Theycanalsoaffordhighfixedcostsbecauseoftheirinvestmentcapabilities.Whileauthoritieshavehighlightedcompetitionbetweencloudproviderstoattractcustomers,someofthepracticesidentifiedposecompetitionissuesbymakingitharderforcustomerstoswitchfromonecloudprovidertoanotherortousemultiplecloudproviders.Obstaclesincludingexitfeesorso-called‘egressfees’andinteroperabilityissues.ThehyperscalersareinvestingheavilyinGenAI.TheyaredevelopingexclusiveorstrategicpartnershipswithGenAIproviderstodeploytheirmodelsontheircloudinfrastructuresandservices.1
IntheEU,theEuropeanCommissionhassetatargetfor75percentofbusinessestousecloud
services,bigdataorAIby2030.2
Statista,‘CloudComputingMarketSizeinEuropeFrom2017to2030,bySegment,June2023’,/forecasts/1235161/europe-cloud-computing-market-size-by-segment.3
Statista,‘GenerativeAIEurope’,/outlook/tmo/artificial-intelligence/generative-ai/europe.4
CompetitionauthoritiesinJapan,SouthKorea,theNetherlands
andFrance(Autoritédelaconcurrence,2023)havealreadypublishedcloudsectorreports.CompetitionauthoritiesintheUnitedStatesandtheUnitedKingdomareattimeofwritinginvestigatingthecloudsector.1TheyalsoofferGenAImodelsandapplications.SomeevenintegratetheirGenAIapplicationsintotheirfamiliesofservices.ThesebehavioursmightreinforcethepositionsofthehyperscalersandraisecompetitionconcernsrelatedtodiscriminationinthesupplyofITequipmentbydominantITproviders,interoperabilityobstaclestoswitching,usageofbusiness-userdata,self-preferencingofcloudservicesoverthirdparties,tyingandpurebundling.CompetitionauthoritiesintheUnitedKingdomandPortugalhavealreadyissuedreportsoncompetitionandGenAI,andtheG7competitionauthoritieshavepledgedtocloselymonitorGenAI5.Inthiscontext,thisWorkingPaperexaminestherelationshipbetweencloudprovidersandGenAIproviders.Itdescribeshowtherelationshipworks,thenoutlinesthecompetitionbenefitsandrisksoftherelationship.Ithighlightshowcompetitionlawsandregulationscanrespondtocompetitionissues.Finally,itconcludeswithsomepolicyrecommendations.2
The
cloud/GenAI
relationshipCloudprovidersthatutilisesharedinfrastructure
supplythreemaincategoriesofsolution:infrastructureasaservice(IaaS),platformasaservice(PaaS)andsoftwareasaservice(SaaS)6.WeexamineinturneachoftheseandtheirusebyGenAIproviders.2.1
Infrastructure
(IaaS)Theinfrastructureofferedbycloudprovidersincludescomputingpowerresources,servers,storageandnetworks.SeveralcloudprovidersofferIaaS,includingAmazonWebServices(AWS),GoogleCloud,MicrosoftAzure,OVHcloud,OutscaleandScaleway.GenAIprovidersneedthisinfrastructuretotrain,runanddeploytheirmachine-learningmodels.Theserequiremassiveamountsofdataandcomputingresourcestogeneratenaturallanguageoutputfrominputdata,includingtext,imagesorvideo.GenAIprovidersscaleupordowntheirstorageandcomputingresources.Inparticular,theyneedaccesstolargestorageandcomputingresourceswhentrainingtheirmodelsforseveralweeks.Then,oncemodelsaremadeavailableonthemarket,theirneedsdependonhowextensivelyusersusethemodeltogenerateoutputs.CloudprovidersareindispensabletoGenAIproviderstoensurethisflexibility,withoutGenAIprovidersneedingtoinvestinhighfixed-costandhardlyscalableinfrastructure.5
CompetitionauthoritiesarealreadystudyingpotentialcompetitionrisksinGenAI.SeeCMA(2023),AutoridadedaConcorrencia(2023),G7competitionauthorities(2023),Carugati(2023c).6
InformationaboutthefunctioningofpubliccloudprovidersisbasedontheFrenchcloudsectorreport(Autoritédelaconcurrence,2023).Thispaperfocusesonlyonpubliccloudprovidersasopposedto
privatecloudproviders,whichusetheirowninfrastructures.Informationaboutmachine-learningmodelsisbasedonCarugati(2023c).2HyperscalerssupportGenAIprovidersfinanciallythroughexclusiveorstrategicpartnerships.Forinstance,Microsofthasinvestedmorethan$11billionsince2019inanexclusivepartnershipwithOpenAItosupportthedeploymentofOpenAItechnologiesexclusivelyviaitscloudservice,MicrosoftAzure7.TheexclusivepartnershipenablesthecloudpartnertoinvestininfrastructurespecificallydesignedfortheGenAIpartner.However,theexclusivepartnershipraisescompetitionrisks.TheGenAIpartnerandtheircustomerscannoteasilyswitchtoalternativecloudinfrastructures.OtherhyperscalersalsohavestrategicpartnershipswithGenAIprovidersfornon-exclusivehostingontheircloudinfrastructures.ThesestrategicpartnershipsallowGenAIproviderstobeavailableonalternativecloudprovidersandcloudproviderstohostseveralGenAIproviders,thusraisingfewercompetitionconcerns.Non-exclusivepartnershipsincludeGoogleCloudwithCoherein20218,AWSwithStabilityAIin20229,AWSwithHuggingFacein202310,AWSwithAnthropicin2023,11
andGoogleCloudwithAnthropicin202312.Aspartofsuchpartnerships,cloudprovidersmobilisetheirmassivecomputingresourcestorunanddeployGenAImodelsatscale.Forinstance,MicrosofthasdevelopedanddeployedaspecialisedsupercomputingsystemspecificallyandexclusivelydesignedtotrainOpenAImodels.Google13
andAmazon14
havedevelopedproprietaryAIsemiconductorstooptimisetheperformanceandcostsofAIworkloads.However,mostofthechipsnecessarytopowerAIapplicationsaresuppliedbyNvidia(estimatedglobalmarketshareof88percentin202215).ThepressreportedinSeptember2023atemporary7
OpenAIBlog,‘MicrosoftinvestsinandpartnerswithOpenAItosupportusbuildingbeneficialAGI’,22July2019,/blog/microsoft-invests-in-and-partners-with-openai;DinaBass,‘MicrosoftInvests$10BillioninChatGPTMakerOpenAI’,Bloomberg,23January2023,/news/articles/2023-01-23/microsoft-makes-multibillion-dollar-investment-in-openai#xj4y7vzkg.8
Cohere,‘CohereandGoogleCloudAnnounceMulti-YearTechnologyPartnership’,17November2021,/news-release/2021/11/17/2336632/0/en/Cohere-and-Google-Cloud-Announce-Multi-Year-Technology-Partnership.html.9
Amazon,‘StabilityAISelectsAWSasItsPreferredCloudProvidertoBuildArtificialIntelligencefortheFuture’,30November2022,/2022/11/stability-ai-selects-aws-as-its-preferred-cloud-provider-to-build-artificial-intelligence-for-the-future.10
AWS,‘AWSandHuggingFacecollaboratetomakeGenAImoreaccessibleandcostefficient’,21February2023,/blogs/machine-learning/aws-and-hugging-face-collaborate-to-make-generative-ai-more-accessible-and-cost-efficient/.11
Amazon,‘AmazonandAnthropicAnnounceStrategicCollaborationtoAdvance
GenAI’,25September2023,/2023/9/amazon-and-anthropic-announce-strategic-collaboration-to-advance-generative-ai.12
LizetteChapman,KatieRoofandJuliaLove,‘GoogleBets$2BilliononAIStartupAnthropic,InksCloudDeal’,Bloomberg,27October2023,/news/articles/2023-10-27/google-to-invest-2-billion-in-ai-startup-anthropic-wsj-says#xj4y7vzkg.13
Cloud,‘AccelerateAIDevelopmentwithGoogleCloudTPUs’,/tpu.14
AWS,‘AWSTrainium’,/machine-learning/trainium/.15
Wallstreetzen,
‘NvidiaCorpStatistics&Facts’,/stocks/us/nasdaq/nvda/statistics.3shortageofNvidiachipsbecauseofcapacityissuesatTaiwanSemiconductorManufacturingCompany(TSMC),Nvidia’smainsupplier16.TheshortageaddstopreviouschipshortagesarisingfromsupplyanddemandshocksfromthelowsupplyduringtheCOVID-19pandemic.andhighdemandfromAIapplications,whichimpactscloudproviders17.Inthiscontext,asnotedbelow,competitionauthoritiesaremonitoringNvidia’ssalesofchipsandhowitallocatessupplyamongcloudproviders,potentiallyprivilegingsomecustomersoverothers.Thus,inadditiontoincreasedconcentrationinthehandsofhyperscalersarisingfromthepartnerships,competitionrisksarisefromdiscriminationinthesupplyofITequipmentbydominantITproviders,andinteroperabilitybarrierswhenswitchingfromonecloudprovidertoanother.2.2
The
platform
level
(PaaS)CloudproviderssupplyingPaaSenabletheircustomerstoaccesssoftwareandtoolstodeveloptheirapplications.Thesetoolsincludedatabases,dataanalysis,toolsfordevelopers,AItools,servicesforinternetofthings(IoT)devices,ITcontainersandsecurity.PaaSisofferedbyseveralcloudproviders,includingAWS,GoogleCloud,MicrosoftAzure,OVHcloudandScaleway.GenAIprovidersofferAItoolstocloudproviderssothatcloudcustomerscanuseanddevelopAIapplications.Forinstance,OpenAIprovidesitsmachine-learningmodelsonMicrosoftAzurewithAzureOpenAIservice18.AWSprovidesaccesstovariousmodelsfromAI21Labs,Anthropic,Cohere,Meta,StabilityAIandAmazonthroughitsAWSBedrockservice19.GoogleCloudalsooffersaccesstovariousproprietary,open-sourceandthird-partymodels20.CloudprovidersandGenAIprovidersthuscompetefortheprovisionofAItools,withpotentialcompetitionrisks,includinglegitimateuseofthedataofbusinessusersforreasonssuchastheprovisionofdataanalysisservices,andalsopotentiallyillegitimatereasons,suchasdevelopmentofcompetingservicesandself-preferencing.2.3
Software
(SaaS)CloudproviderssupplyingSaaSenablecloudcustomerstoprovideapplicationstotheirusersviaanyconnecteddevice.ApplicationsincludeproductivitysoftwaresuchasMicrosoft365orGoogleWorkspace,orstreamingservicessuchasNetflix.GenAIprovidersenablecloudproviderstointegratetheirsolutionsintoproprietarySaaSapplications.Forinstance,MicrosoftintegratesOpenAIsolutionsintoitsproductivitysoftware,Microsoft35516
ChengTing-Fang,‘TSMCSees
AIChipOutputConstraintsLasting1.5Years’,NikkeiAsia,6September2023,/Business/Tech/Semiconductors/TSMC-sees-AI-chip-output-constraints-lasting-1.5-years.17
AlexWoodie,‘TheChipShortageSeemstobeImpactingAIWorkloadsintheCloud’,Datanami,
12March2021,/2021/03/12/the-chip-shortage-seems-to-be-impacting-ai-workloads-in-the-cloud/.18
MicrosoftAzure,‘AzureOpenAIService’,/en-us/products/ai-services/openai-service#Features.19
AWS,‘AmazonBedrock’,/bedrock/.20
GoogleCloud,‘GenAIonGoogleCloud’,/ai/generative-ai.4Copilot,itswebbrowser,MicrosoftEdge,itssearchengine,MicrosoftBing,anditsoperatingsystem(OS),MicrosoftWindowsCopilot.Googledoesthesamewithitsproductivitysoftware,GoogleWorkspaceanditssearchengine,GoogleSearch.SomeSaaSactasaplatformenablingthird-partydeveloperstodevelopapplicationsthatcomplementandinteractwiththeirservices.Forinstance,Microsoftenablesthird-partydevelopersincludingOpenTableandExpediatodevelopthird-partypluginsthatinteractwithOpenAIChatGPT,MicrosoftBing,MicrosoftDynamics365Copilot,Microsoft365CopilotandMicrosoftWindowsCopilot21.CloudprovidersandGenAIprovidersarethusverticallyintegrated,bringingpotentialcompetitionrisks,includingtyingandpurebundling.3
Competition
risksWhilethecloud/GenAIrelationshipbringswithitmanycompetitionbenefits,asdiscussedintheprevioussections,therelationshipalsocreatesrisksrelatedtopotentialconcentration(section3.1.)andanticompetitivepractices(section3.2.).3.1
Potential
concentration
issuesInchoosingcloudproviders,thereisariskthatGenAIprovidersusemostlyhyperscalers,increasingthetrendtowardconcentrationinthehandsofthehyperscalers.First,GenAIprovidersusetheinfrastructures,platformsandsoftwareofhyperscalers.Accordingly,GenAIprovidersandtheirconsumersbecomedefacto
cloudcustomersofthehyperscalers.Second,thepartnershipbetweencloudprovidersandGenAIprovidersmightreinforcethepositionofcloudprovidersinthecloudmarketandrelatedmarkets.Theextentofthereinforcementdependsonthenatureandconditionsofthepartnership,thusrequiringacase-by-caseanalysis.Bycontrast,somesmallercloudprovidersalsohavepartnershipswithGenAIproviders,suchasScalewaywithMetaandHuggingFaceatthenationallevel22.However,smallercloudprovidersmightnotbeabletocompetewithhyperscalersatgloballevelbecausetheymightlacktheinvestmentcapabilitiesandinfrastructurerequiredtotrain,runanddeployGenAImodelsandapplicationsgloballyatscale.Inotherwords,GenAIproviderscanbeavectorofgrowth,mainlyforhyperscalers.Nevertheless,thedegreeofconcentrationinthehandsofhyperscalersresultingfromGenAIwillhingeontheshareofrevenuesalesandcloudcustomersusingcloudresourcesforGenAI,whichisjustapartofoverallcloudactivities.21
FrankX.Shaw,‘MicrosoftBuildBringsAIToolstotheForefrontForDevelopers’,OfficialMicrosoftBlog,23May2023,/blog/2023/05/23/microsoft-build-brings-ai-tools-to-the-forefront-for-developers/.22
MetaBlog,‘MetaPartnerswithHuggingFace&ScalewaytoSupportOpenSource’,8November2023,/news/2023/11/meta-partners-with-hugging-face-scaleway-to-support-open-source/.5Thepotentialcompetitionrisksfromincreasedconcentrationinthecloudsectoraretwofold.First,whileconcentratedmarketscanbecompetitive,itbecomesharderforsmallerprovidersandnewcomerstocompeteinandtoenterconcentratedmarkets,especiallymarkets,likecloudmarkets,characterisedbystrongeconomiesofscaleandscope(Cremer,deMontjoye,andSchweitzer,2019).Concentratedmarketsarethuslesscontestable,withcustomershavingfewoptions.Second,concentratedmarketsaremorepronetoanticompetitivepracticesarisingfromabuseofdominanceandcollusionbecauseoftheiroligopolisticnatures,withcustomersfacinganimbalanceofpowerwiththeavailableoptions.Thesecompetitionrisksfromincreasedconcentrationaretheoriginofregulationsindigitalmarkets,includingtheEuropeanUnion’sDigitalMarketsAct(DMA,Regulation(EU)2022/1925;seesection4.2).3.2
Potential
anticompetitive
practicesTherelationshipbetweencloudprovidersandGenAIprovidersmightalsoleadtopotentialanticompetitivepracticesattheIaaS,PaaSandSaaSlevels.TherisksrelatetodiscriminationinthesupplyofITequipmentbydominantITproviders,interoperabilityobstaclestoswitching,usageofbusinessuserdata,self-preferencingofcloudservicesoverthirdparties,tyingandpurebundling.Table1indicatesthepotentialcompetitivebenefits,anticompetitiverisksandanticompetitiveeffectsofthesebusinesspractices.Discrimination
in
the
supply
of
IT
equipment
by
dominant
IT
providers.
DominantITprovidersallocatethesupplyofITequipment,likeGPUs,amongstcloudcustomers.Inthecontextofashortageofchips,theFrenchcompetitionauthoritystatedinitscloudsectorstudythathyperscalersbenefitfromprivilegedaccesstocertainITequipmentbecauseoftheirlargepurchasingvolumes(Autoritédelaconcurrence,2023).Inotherwords,dominantITprovidersmightdiscriminateamongstitscustomershowitallocatesitsupply.ThispotentialanticompetitivepracticealreadyraisedantitrustscrutinyinFrance,Europe,theUnitedStates,andChina23.Interoperability
obstacles
to
switching.
CloudprovidersmighthavetheabilityandincentivetolimittheinteroperabilityofGenAIsolutionswithothercloudproviders,withapotentialexclusionaryeffect.CloudproviderscanimposeorwithholdtechnicalrequirementsthatlimitGenAIsolutionsandtheircustomersfromswitchingfromonecloudprovidertoanother,orpreventthemfrommulti-homingwith23
Autoritédelaconcurrencepressreleaseof27September2023,‘TheGeneralRapporteuroftheAutoritédelaconcurrenceIndicatesthatanUnannouncedInspectionwasCarriedOutintheGraphicsCardsSector’,https://www.autoritedelaconcurrence.fr/en/press-release/general-rapporteur-autorite-de-la-concurrence-indicates-unannounced-inspection-was.Seealso,Form10-QNvidiaCorporation,21November2023,/financial-info/sec-filings/sec-filings-details/default.aspx?FilingId=17074143:“Ourpositioninmarketsrelating
toAI
has
ledtoincreasedinterestinourbusinessfromregulatorsworldwide,including
theEuropeanUnion,theUnitedStates,and
China.
Forexample,the
FrenchCompetitionAuthoritycollectedinformationfromus
regarding
ourbusinessandcompetitioninthegraphicscardand
cloudserviceprovidermarketas
partof
an
ongoing
inquiryintocompetitioninthose
markets.Wehave
alsoreceivedrequestsforinformationfromregulatorsintheEuropeanUnionandChina
regarding
oursalesof
GPUsand
oureffortstoallocate
supply,and
we
expecttoreceiveadditionalrequestsforinformationinthefuture”(p.42).6morethanonecloudprovider,so-called‘multi-cloud’.CloudprovidersmighthavetheincentivetodosotokeepGenAIprovidersandtheircustomersontheircloudservices.Usage
of
business
user’s
data.
CloudprovidersmighthavetheabilityandincentivetousedatafromGenAIprovidersandtheircustomers,whichmighthaveanexploitativeeffect.CloudprovidershavetheabilitytousethedataprovidedorgeneratedbyGenAIprovidersandtheircustomersastheystorethisdataontheircloudinfrastructure.Theymighthavetheincentivetodosoinordertoprovide,improveordeveloprelatedcloudservices,suchasdataanalysis,AItoolsorevenproprietaryGenAIsolutions.Asnotedabove,thisdatausagemightbeforlegitimatereasons,suchastheprovisionofdataanalysisservices.Oritmightbedoneforpotentiallyillegitimatereasons,suchasdevelopingcompetingservices,whichmightconstituteapotentialcontractbreachandanticompetitivepractice.However,thisexploitativestrategyisself-harmingasitunderminestrustbetweencloudprovidersandtheirGenAIprovidersandtheircustomers,whichmightleadthelattertowithdrawbusinessfromcloudproviders.Cloudprovidersmightthuslosesignificantbusinessopportunitiesandsufferreputationalharm.Accordingly,whiletheoreticallypossible,itisquestionablewhethercloudproviderswillengageinthisstrategyinpractice,astrust,businessopportunitiesandreputationsmightbemorevaluablethanthedevelopmentofcompetingservicesorproducts.Self-preferencing
of
cloud
services
over
those
of
third
parties.
Cloudprovidersmighthavetheabilityandincentivetopromotetheirowncloudservicesoverthirdparties,withpotentialexclusionaryeffectsonthirdparties.CloudprovidersareabletopromoteoverthirdpartiestheirproprietaryGenAItoolsandmodelsontheirplatforms.TheymighthavetheincentivetodosotoincreasethesalesoftheirproprietaryGenAIofferings.Tying.
Cloudprovidersmighthavetheabilityandincentivetoimplementtyingstrategies,whichmighthaveexclusionaryeffects.Cloudprovidersmighttietheprovisionofnon-dominantcloudservicestotheprovisionofrelateddominantservices,withGenAIasaconnectorbetweenthem.Forinstance,anon-dominantcloudproviderwithadominantpositioninproductivitysoftwareoroperatingsystems(OS)couldrequireorstronglyencouragesoftwaredeveloperstouseitsproprietaryorexclusivethird-partyGenAIsolutions,availableonlyonitsnon-dominantcloud,todevelopthird-partyGenAIapplicationsthatcomplementandinteractwithitsproprietarydominantproductivitysoftwareorOS.Cloudprovidershavetheincentivetodothistoincreasesalesoftheirnon-dominantcloudservices.Pure
bundling.
Cloudprovidersmighthavetheabilityandincentivetoimplementbundlingstrategies,withpotentialexclusionaryeffects.Cloudproviderscanprovidetheirproprietaryorthird-partyexclusiveGenAImodelsandapplicationstogetherwiththeirnon-dominantcloudservices.Theymightseektodothistoincreasesalesoftheirnon-dominantcloudservices.However,astheGenAIsectorisstilldevelopingwithintensecompetitionbetweenvariousGenAIproviders24,thispracticeisunlikelytohaveanegativeimpactoncloudprovidersastherearenotyetdominantGenAIproviders.Thisimplies24
AsofSeptember2023,theUKCompetitionandMarketsAuthority(CMA)counted160foundationmodelssince2018(CMA,2023).7thatGenAIcustomerscanusealternativeGenAIprovidersavailableondifferentcloudproviders.However,assumingaGenAIproviderbecomesdominant,thebundlingstrategymighthaveprocompetitivebenefitsandanticompetitiverisks.Table
1:
Summary
of
potential
anticompetitive
practices
in
cloud/GenAIPracticeProcompetitivebenefitsThepracticeenablesAnticompetitive
risksPotentialanticompetitive
effectPotentialDiscriminationintheThepracticemightsupplyofITequipment
cloudproviderswithpreventsmallercloud
anticompetitivebydominantITprovidersalready-scalablecloud
providersfrom
exclusionaryeffectofinfrastructuresandinvestmentcapabilitiestoadaptthemquicklyforGenAIproviders.accessingchipsinatimelywayfortheircloudinfrastructures.puttingsmallercloudprovidersatacompetitivedisadvantageagainsthyperscalers–smallerproviderscannotoffercloudservicestoGenAIprovidersbecausetheydonothavetherequiredITequipment.Thoughchipshortagesmightbetemporary,theycanhavelastingimpactsoncompetitionbecauseGenAIproviders,attractedtothehyperscalers’processingcapacities,mightcontinuetousethehyperscalers’infrastructuresaftertheendofshortages.PotentialInteroperabilityTheobstaclesprovide
Theobstaclesmightobstaclestoswitching
incentivestoinvestin
preventGenAIanticompetitiveinfrastructure,platformsandsoftware
customersfromspecificallyand/orprovidersandtheirexclusionaryeffectofexcludingalternativecloudproviders.switchingtoexclusivelydesignedal
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 糕点店财务管理与风险防范考核试卷
- 数据库应用操作技巧分享试题及答案
- 计算机四级考试小贴士试题及答案分享
- 行政组织中的创新思维与解决问题策略试题及答案
- 公司会所食堂管理制度
- 公路工程团队建设试题及答案
- 嵌入式系统中的数据通信技术试题及答案
- 展会公共人员管理制度
- 劳务公司分包管理制度
- 医院新进职工管理制度
- 职业技术学校中医康复技术专业人才培养方案
- 辽宁省名校联盟2025年高考模拟卷押题卷数学(三)
- 2024年四川巴中事业单位招聘考试真题答案解析
- 以好家风涵养好作风-新时代领导干部家风建设专题课件
- 2025年甘肃省武威第二十中学生物七年级下册新人教版期中模拟练习题(含答案)
- 银行客户经理培训课件
- 药品理化检验培训
- 腹部带蒂皮瓣护理
- 仓库7s管理制度培训
- 复式交分道岔检查课件
- 学校防恐防暴应急演练方案
评论
0/150
提交评论