AIasGameChanger|Authors&ContactpersonAIasGameChanger|LeadAugustinFriedelSoftwareDefinedVehiclesAugustin.Friedel@LeadMatthiasBorchArtificialMatthias.Borch@ContactStephanBaierArtificialStephan.Baier@

AuthorMarcusWillandMobilityMarcus.Willand@AuthorDr.NilsSchaupensteinerTransformationAdvisoryNils.Schaupensteiner@AuthorPatrickRuhlandTransformationAdvisoryPatrick.Ruhland@Thestudy“AIasGameChanger“anditssummarywerepublishedby:MHPGesellschaftfürManagement-undIT-BeratungmbHAllrightsreserved!Noreproduction,microfilming,storage,orprocessinginelectronicmediapermittedwithouttheconsentofthepublisher.Thecontentsofthispublicationareintendedtoinformourcustomersandbusinesspartners.correspondtothestateofknowledgeoftheauthorsatthetimeofpublication.resolveanyissues,pleaserefertothesourceslistedinthepublicationorcontactthedesignatedcontactpersons.Opinionarticlesreflecttheviewsoftheindividualauthors.Roundingdifferencesmayoccurinthegraphics.3ContentsContents 4offigures 612KeyFindings 8WelcometoChange! 10RevolutionandAutomotiveMarketPotential 11InvestmentinCompaniesWithanAIFocus 15PilotProjectsandImplementation 19AIModels,Levels,andUseCases 23TheGameChanger:WhatCanBeAchievedWithAI 26AutomobileManufacturersWithLowAIInvestment 29AIModels:MakeorBuy? 29AIApplicationsAlongtheAutomotiveChain 31AIOperationinVehiclesandintheCloud 35AIMonetizationinVehicles 39AddedofAIApplicationsinCompanies 40WhattheCustomerWants:TheUserPerspective 47UseandUnderstandingofAIApplications 49AdvantagesandDisadvantages–GenerallyandinVehicles 49PurchasingDecision,andWillingnesstoPay 514AIasGameChanger|AIasGameChanger|ContentsSuccessFactorsandStrategicApproach 55StrategyandGoalPlanning 56ThinkfromthePerspectiveofthenotthe56OrganizationalAnchoringandOwnership 58LocalDifferencesrequirelocalSetup 59ReducingComplexity 59UseandMonetizationofData 60ChecklistforsuccessfulImplementation 61Challenges,Responsibility,andRisks 63CostsofandOperation 64DataandDigitalizationasaBasis 65BusinessModelsandCasesforB2CandB2B 65EthicsandResponsibility 67NewRisksandRegulatoryChallenges 69AIApplicationsintheAutomotiveIndustry:7RecommendationsforAction 7110.FurtherInformations 75LiteratureandSources 76ContactInternational 78AboutMHP 795ofFigure1:Technologysupercycles–artificialintelligenceasthenextrelevantplatformshift(Coatue,2024) 12Figure2:AImarketsizeintheautomotivesector(PrecedenceResearch,2024) 12Figure3:investmentsinAIcompaniesfoundedsince2001,inUSDbillion2024) 16Figure4:InvestmentinAIstacklayers(Coatue,2024) 17Figure5:CompanieswithteamandbudgetforAI(Capgemini,2023) 21Figure6:InterconnectedAIconcepts 24Figure7:VisualizationofAIasapyramid 25Figure8:ClassificationofAIterms 27Figure9:TheperformanceofAImodelscomparedtohumancapabilitiesintheMMLUtest(iAsk,2024) 28Figure10:SchematicdiagramofthetrainingofAIfoundationmodelsforvehicles 30Figure11:UseofAIalongthevaluechain 32Figure12:SignificantimprovementsoffunctionsandfeaturesthroughAI 33Figure13:InterestinAIfunctionscomparedinternationally 34Figure14:Roleofon-premise,cloud,andvehicleforAImodels 35Figure15:Levelsofasoftware-definedvehicle(SDV)(Willand,Friedel,&Schaupensteiner,2023) 36Figure16:DifferentmodelsforADASandADapplicationsandfunctions 37Figure17:potentialatdifferentstagesofthevaluechain(Capgemini,2023) 40Figure18:UseofAI-basedsolutionsbyregion 41Figure19:KeydriversbehindtheuseofAIinproduction 426AIasGameChangerAIasGameChanger|TableoffiguresFigure20:Decisiveissue–fewerusersofsoftwareduetoAIorfreesoftware(Coatue,2024) 43Figure21:PossibleusesofAIinsoftwaredevelopment(Wee2024) 44Figure22:UnderstandingofAIincars 48Figure23:AdvantagesofusingAIincars 49Figure24:TheperceivedadvantagesanddisadvantagesofusingAI 50Figure25:AIincars:purchasemotivationorblocker? 51Figure26:instakeholderswithregardtotheimplementationofAIinvehicles 52Figure27:WillingnesstopayforAIfunctions 52Abb.28:AssessmentofthefutureAIcompetenceofcarmanufacturersbyregion 53Figure29:Customerandusecasefirst,andthenAIapplicationsandmodels 57Figure30:Dimensionsforvalidatingtechnicalfeasibility 57Figure31:costsforAImodelsareincreasing(StanfordUniversity,2024) 64Figure32:Dataavailabilityandqualitybyregion 65Figure33:Customers’willingnesstopayisunclear;costsariseforimplementationandoperation 66Figure34:ClassificationofAIusecasecategoriesandpossiblebusinessmodels 67Figure35:RisksassociatedwiththeuseofAI 68Figure36:PrinciplesandpenaltiesoftheEUAIAct 701:ThedevelopmentofAImodelsdividedintodifferenttimephases 27712KeyFindingsThewidespreaduseofAIispredictedtobethenextrelevantplatformshiftaftercloudtransformation–originalequipmentmanufacturers(OEMs)needtostepuptheiractivities.MorethanOnlyofrespondentsseetime-savingasthebiggestbenefitofAIapplications.SkepticismaboutAIapplicationsisgreaterintheUSthaninEuropeorChina.

ofrespondentsinChinastatethattherisksofAIoutweighthebenefits;thisfigureisaround25inEuropeandtheUS.ThemostfrequentlymentioneddisadvantagesofAIarefearoflossofcontrol,lossofdataandpersonalprivacy,andsecurityrisks.8CustomersworldwidewanttouseAIincars,butrarelypayforit. KIInChina,AIfunctionsmostlyhaveapositiveinfluenceoncarpurchasingdecisions–onlyofrespondentswouldnotbuyavehiclebasedonAIfunctions.

ChinesecarmanufacturersareregardedasleadersinAIinnovation.Infiveyears’time,JapaneseOEMsbeattheforefront,followedbyChineseandGermanOEMs.AIisnotonlyrevolutionizingthein-vehiclecustomerexperience–theentirevaluechainisexperiencingdisruptivechange.carmanufacturersarethemosttrustedwhenitcomestotheuseofAI,faraheadofstateinstitutionsnewcarmanufacturers.

SuccessfulimplementationofAIapplicationsisnotpossiblewithoutpriordigitalizationandaccesstospecificdatasources.AIasGameChanger|AIasGameChanger|12KeyFindings9WelcometoChange!Dearreaders,Artificialintelligencewillbethenextplatformshiftthatrevolutionizesallindustrialsectors.StakeholdersintheautomotivevaluechainhaverealizedthatAIischallengingmanytradi-tionalprocessesandwaysofthinking.TheintroductionofthePC,thestationaryInternetandthenthemobileInternet,andCloud/SaaSpreviouslyhadasimilarlydisruptiveimpact.Newbusinessmodelsandprofitpoolsareemerging,whileatthesametimetherearenu-merouschallengestobetackledwithregardtotechnology,partnerships,andethicalissues.Inthisstudy,wetracethegroundbreakingdevelopmentsinAIsofarandexaminetheop-portunitiesandriskswithintheautomotiveindustry.Accompanyusthroughpresentfuturescenarios–withspecificrecommendationsforactionforyourownstrategywhenitcomestoimplementingAIapplicationsinproductionandinvehicles.Whetherthenewtechnologiesmeettheexpectationsofdriversisdeterminedrightthereinthecockpit.why,inChapter8,weoutlinetheuserperspectivebasedonourowncurrentdata.OurinternationalsurveyprovidesinformationaboutwhichproductsservicesfromautomotivecompaniescouldfulfillAIneedsandwhatthewillingnesstolookslike.Thatmakesthisstudyessentialreadingfordecision-makers,CIOs,andapplica-tiondevelopers.InvestorsinAItechnologiesandAIteamsneedaconsistent,long-termcost-benefitWethereforeexaminethedirect/indirectmonetizationofin-carAIandlookatnewbusinessmodelsbasedonAIanddigitalization.Ultimately,asissooftenthecase,itbecomesclearthatthejourneyintonewtechnologicalterritoryisbestundertakenwithexperiencedtravelguides.Gettheknow-howyouneed–andalwaysbecurious!ENABLINGYOUTOSHAPEABETTERTOMORROWBestregards,Dr.JanWehingerClusterLeadSoftwareDefinedVehiclesMHPManagement-undIT-BeratungGmbHLudwigsburg,September202410本報告來源于三個皮匠報告站（）,由用戶Id:349461下載,文檔Id:183532,下載日期:2024-12-04AIasGameChanger|AIasGameChanger|01.RevolutionandAutomotiveMarketPotential01.RevolutionandAutomotivePotential11EveryonerecognizesthatAIisthenextplatformshiftDesktopInternet(Web1.0)DesktopInternet(Web1.0)MobileInternet(Web2.0)Cloud/SaaSGenerativeAINetworkingPCMainframe1960–1980

1980s

1990s

2000s

2010s

2015–2020

2022–...Figure1:Technologysupercycles–artificialintelligenceasthenextrelevantplatformshift(Coatue,2024)AI-BasedsystemsforautomotiveindustrytoreachUS$35.7billionby203335.726.626.6...inbillionUS$20.05.87.32023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033Figure2:AImarketsizeintheautomotivesector(PrecedenceResearch,2024)12ItishighlylikelythatthebigtechnologycompaniessuchasGoogle,Meta,andMicrosoft–whichgainedinimportancewiththelastplatformshifts(supercy-cles)–willalsodominatetheAIage.Alongtheautomotivevaluechain,stakeholderssometimesaccusedofhavingrespondedtotheplatformshiftstoolateorwithanineffectivestrategy.Inouropinion,therelevanceofconnectivityandsolutionswasrecognizedtoolateandimplementationcouldhavebeenTheindustryisattheningoftheAIplatformshiftandthereisstilltheportunitytorespondearlywithatargetedstrategy.CompanieslikeApplehaveshownthatitisnotneces-

OneisthatartificialintelligenceincreasinglyreplacepeopleandjobsmayCurrently,AIapplicationsareregardedmoreasaplementratherthanareplacement.AcademicsasKarimLakhanifromHarvardBusinessSchoolbelievethatAIwillnotreplacehumans.OnepossiblescenarioisthatpeoplewhouseAIwillhaveasignificanttageoverworkerswhodonotuseit.RegardingthequestionofwhetherAIwillimproveeconomy,asurveyshowsamixedpicture.34percentofrespondentsbelievethattheuseofwillimprovetheeconomicsituationintheircountryinthenextthreetofiveyears.Thishopeisabove“AIWon’tReplaceHumans—ButHumansWithAIWillReplaceHumansWithoutAI.”(HBR,2023)sarytobethefirstinnovator.WithastrongAIstrategy,acompanycanalsoexploitpotentialasafastThemarketforartificialintelligenceintheautomotiveindustryhasshownremarkablegrowthinrecentItiscurrentlyestimatedtobearoundUSD3.9billionin2024andisexpectedtogrowtoUSD15billionby2030.SomemarketanalysesanticipatethatAIsalestheautomotivesectorwillrisetooverUSD35billionin2033.Growthfrom2024to2033correspondstorateof28percent.Estimatesinothermarketreportsmaybeslightlyerorbutallshowthesametrend.Thisthatextensiveeconomicopportunitiesarebeingedalongthevaluechainformanufacturers,suppliers,andserviceproviders.

incountriessuchasThailand,India,andSouthAfrica.Atthelowerendoftherankingarecountriesinclud-ingBelgium,Japan,theUS,andFrance(Ipsos,2023).Overall,thereareincreasingsignsthattherearemoreopportunitiesthanrisks.Thetargeteduseofarti-ficialintelligencewillsignificantlyaffectourinthecomingdecades.AIboostsefficiencyandcounterthenegativeeffectsofskillsshortages,graphicchanges,andhighlocationcosts.Itisnowuptotheautomotiveindustrytotakeboldandatelyfastaction–andfollowastrategicallyintelligentapproach.AIasGameChangerAIasGameChanger|01.RevolutionandAutomotiveMarketPotential1314AIasGameChanger|AIasGameChanger|02.InvestmentinCompaniesWithanAIFocus02.InvestmentinCompaniesWithanAIFocus15Magnetforinvestment:investmentinAIcompaniesfoundedsince2001inbillionsofUSdollars39.6Bn.US$39.6Bn.US$234.1Bn.US$41.7bn.SiliconValleyAlookatthedistributionofAIinvestmentshowsthedominanceofthoseregionsthatalsodominatedthemarketinthelastplatformshifts(seeCoatue,2024;Figure1).Itcanbeassumedthattheautomotivein-dustrywillcontinuetobedependentonhyperscalersandtechnologycompanies.Collaborationsregardingsoftware,cloudapplications,andtheuseofAIareex-pectedtoincrease.AnanalysisshowsthatalargeshareoftheinvestmentinAIcompaniescomesfromtheUS.Acloser(Coatue,2024)showsthatonlyapprox.3percenttheventurecapitaldealshaveaclearlinktoAI,butthat15percentoftheinvestedcapitalflowsintostart-ups.Fromthisimbalance,itcanbeconcluded

thatthemarketseesrelativelyhighvaluationscorrespondinglyhighinvestmentrounds.Thefinanc-ingroundsshowthatmostoftheinvestmentsin2024wentintocompaniesthatdevelopAImodelssuchasMistral,andClaude.AtotalofUSD14lionwasinvestedinAImodelsinthefirsthalfoftheThisequatesto62percent.In2024,asmallerproportionofthecapitalinvestedAIcompanieswentintofirmsthatdevelopsemicon-ductorsforAIapplications.Roboticsapplications,ashumanoidrobots,garneredapprox.USD2billionincapital,whichcorrespondstoaround9percentofthetotal.101.2bn.US$101.2bn.US$16AmongthelargestinvestorsintheAIfieldaremajortechnologycompaniesincludingAmazon,NVIDIA,andAlphabetcompany).In2023,thesecompaniesinvestedUSD25billionandwerethusresponsiblefor8ofinvestment.Carmanufacturers’investmentsincompaniesdealwithartificialintelligencearemoremodest.aresomeexamples:InvestmentsbyNIOCapitalMomenta:Start-upwithafocusonautonomousingandonthedevelopmentoftechnologiesforronmentalperceptionandhigh-precisionmappingPony.ai:Companyfocusingonautonomousdriving;itformspartnershipstodevelopmobilitysolutionsBlackSesameTechnologies:CompanyspecializinginAIchipsandsystems

InvestmentsbyBMWiVenturesAIasGameChanger|02.InvestmentinCompaniesWithanAIFocusAlitheon:SpecializesinopticalAItechnologyforjectidentificationandauthenticationwithAIasGameChanger|02.InvestmentinCompaniesWithanAIFocusRecogni:Focusesonhigh-performanceAIwithlowpowerconsumptionforautonomousvehiclesAutoBrains:DevelopsAIsolutionsfortheautomotiveindustry,particularlyinthefieldofautonomousdriv-ingtechnologiesInvestmentsbyPorscheSensigo:DeveloperofanAI-supportedplatformforoptimizingvehiclediagnosticsandrepairprocessesWaabi:CanadiandeveloperofAI-basedsolutionsforself-drivingtrucksAppliedIntuition:Providessoftwaresolutionsforthedevelopmentofdriverassistancesystemsandauton-omousdrivingCresta:Specializesinreal-timeintelligenceforcustom-erinteractionsandcommunicationsolutionsWhereareAIVCdollarsgoing?Funding 1008060400

62%AI

20%AIApps

9%AIOps/AICloud

9%AI

<1%AISemisFigure4:InvestmentinAIstacklayers(Coatue,2024)1718AIasGameChanger|03.PilotProjectsandImplementationAIasGameChanger|03.PilotProjectsandImplementation03.PilotandImplementation19Withoutcomprehensivepriordigitalization,theimplementationofAIapplicationswillbeaninsurmountablechallenge.CarmanufacturersandsuppliersshouldallocatebudgetsforAIandbuildupexpertisepromptly. 20Intheautomotiveindustry,amixediswithtotheacceptanceandimplementationAIapplicationsalongthevaluechain.Thelevelofplementationislowamongsuppliersanddealersinservices.Automobilemadefurtherintermsofimplementation,ispotentialforLookingattheautomotiveindustryasawhole,4percentofcompanieshavebeguntoimplementapplicationsatselectedlocations.Thatisaroundasmuchasinthepharmaceuticalindustry.InthefigureisfourtimesSome28percentcompaniesintheautomotivevaluechainareonAIpilotprojects,andthevastmajority(68percent)arestillatexplorationstage(CapgeminiResearchstitute,2023).

Only30percentofthecompaniesintheautomotivesectorhaveadedicatedteamandanextrabudgettheintroductionandimplementationofAIparison,therateis62percentinretail,74percentinthehigh-techand52percentindefense.(Capgemini,2023)AIasGameChanger|03.PilotProjectsandImplementationInterimconclusion:TheautomotivementinAIhasbeenbelowaveragetodate;thisbudgetsandspecializedteams.GiventhehugeofAIontheindustry,itisadvisabletothistionAIasGameChanger|03.PilotProjectsandImplementationProportionofcompanieswithadedicatedteamandbudgetforAI30%74%62%30%74%62%52%36%40%

High

Retail defense

Tele-communi-cationsFigure5:CompanieswithteamandbudgetforAI(Capgemini,2023)2122AIasGameChanger|AIasGameChanger|04.AIModels,Levels,andUseCases04.AIModels,Levels,andUseCases23InterconnectedAIconceptsEachconceptisaspecializedpartoftheoneprecedingit.Figure6:InterconnectedAIconceptsAIcoversawidefieldthatcanbedividedintoseveralareasandtermsusingahierarchicaldiagram:ArtificialIntelligence(AI):Researchareafocusingonthecreationofintelligentmachines.Machinelearning(ML):BranchofAIfocusingonthedevelopmentofmachinesthatcanlearnfromdata.Deeplearning:Asub-categoryofmachineingbasedonartificialneuralnetworks.Examplesconvolutionalneuralnetworks(CNNs)andneuralnetworks(RNNs).GenerativeAI:Aspecialtypeofartificialneuralnet-worksthatgeneratedatasimilartothetrainingdata.Examplesaregenerativeadversarialnetworks(GANs)andlargelanguagemodels(LLMs).WithAIapplications,variouscategoriesofusecasescanbeimplemented:Datamanagement:Thisinvolvesharmonizingdataandobtainingfindings.Itisessentialforefficientuseofinformation.

Patternrecognition:Thiscategoryincludesanomalydetection,categorization,andprediction,andalsotimeseriesanalysis.Thesetechniquesen-abletheidentificationoftrendsandpatternsinlargedatasets.Decision-making:Includesrecommendationsys-temsandrecommendedactionsbasedondataanalyses.Theseareusedtosupportdecision-mak-ingprocesses.Textandimageprocessing:Includessemanticsearch,textsummaries,andtranslations,whichenabletheprocessingandanalysisoftextandim-agedata.Communication:Thiscategoryincludesvoicerec-ognition,chatbots,voicecontrol,intelligentpro-cessautomation,andvoiceoutput,whichfacili-tateshuman-machineinteraction.Creativityandcontentgeneration:Thegener-ationoftextsandimagescanautomateandaccel-eratecreativeprocesses.24AIsystemscanbevisualizedasapyramidwithaberoflayers.Thelowestlayerdescribesthecomputingpowerrequiredfortheotherlayers.CompaniesasNVIDIA,ARM,andAMDareestablishedcompaniesthatprovidechipsetsandhigh-performancecomput-ersforthisPerformancehasdevelopedrapidlyrecentyears.Forexample,marketleaderNVIDIAlivereda1000-foldincreasebetween2016and2024.Thesecondlayerfrombottomconsistsofcloudfrastructureandenablers.Thebiggest,mostknowncompaniesthatprovidetheinfrastructurearecompaniessuchasAmazonwithAWS,AlphabetwithGoogleCloud,andMicrosoftwithAzure.Asthecloudcomputingsegment,theprovidersaregion-specific.InChina,theinfrastructurelayeris

eratedbycompaniessuchasTencent,Alibaba,andBaidu;theyareamongthechipproviders’largestcus-tomers.AIasGameChanger|AIasGameChanger|04.AIModels,Levels,andUseCasesUsersmostlikelycomeintocontactwiththeapplica-tionlayeroftheAIpyramid,forexamplewhentheyaskaquestionintoolslikeChatGPTfromOpenAI,ClaudefromAnthropic,orGeminifromGoogle.VisualizationofAIasapyramidTrainingConvenienceTrainingConvenienceADAS/ADApplicationsFoundationalmodelsInfrastructure&enablerSemiconductorecosystemOutsidevehicle(e.g.cloud)Invehicle(e.g.computeunits,software)Figure7:VisualizationofAIasapyramidOutsidevehicle(e.g.cloud)Invehicle(e.g.computeunits,software)25ItisimportanttodistinguishbetweentheingAIterms:AImodelsaremathematicalandstatisticalcon-structsthataredevelopedandtrainedtocarryoutspecifictasksbasedondata.AIcomponentsconsistofAImodelsandincludeapipelineforprocessinginputandoutputdata.coverpreprocessing(processingofrawdata)andprocessing(preparationofresults).AIcapabilitiesarespecificfunctionsorthattheAIsystemcanexecute,suchasvoiceorrecognition.AIusecasesdefinethepurposeandcontextinwhichAIcapabilitiesareapplied.Ausecasecanin-cludeseveralAIcapabilities.Figure8showshowtheseelementsbuildononean-other.TheGameChanger:WhatCanBeAchievedWithAIThedevelopmentofAImodelscanbedividedintoferenttimephases.Wearecomingoutofthephaseartificialnarrowintelligence(ANI),whichwasbymachinelearningandapplicationslikeSiriandexa.GenerativeAImodelscanberegardedasadevel-opmentalsteponthepathtoartificialgeneralgence(AGI).Inthisphase,machineswouldbeablematchhumansinawiderangeofcognitiveabilities.Inthephaseofartificialsuperintelligence(ASI),AIplicationsandmachineswithembeddedAIwillsurpassedhumansasregardscognitiveabilities.

TheuseofAIwillbecomeagamechangerbecausethetechnologyenablesthefollowingfunctionsforfirsttime:HumaninteractionFeatures:Abilitytohavecolloquialandmulti-levelconversations,aswellasusecontextandmemoryImportance:ThesecharacteristicsenableAIcommunicatemorenaturallyandintuitivelypeople–particularlycrucialforapplicationsasvirtualassistants.HumanthinkingFeatures:Logicalthinking,thecreationofworldmodelsandpredictiveabilitiesImportance:Theseabilitiesareimportantforsolvingcomplexproblemsandmakingwell-foundedsions,similartoahuman.VisualunderstandingFeatures:Profoundinterpretation,opensetofcon-clusions,understandingofconsequencesImportance:TheseaspectsenableAItoanalyzeandinterpretvisualdata,whichisimportantinareassuchasimageprocessingandautonomousnaviga-tion.Internet-scalableknowledgeFeatures:Trainingonextensivetextualandvisualdata,adaptationtonewsituations,awarenessoflocalcustomsandsignsImportance:ThesecharacteristicsenableAItoin-tegrateknowledgefromavarietyofsourcesandapplyitindifferentcontexts.26AIasGameChanger|AIasGameChanger|04.AIModels,Levels,andUseCasesAIusecasesEmbeddingAIcapabilitiesAIusecasesEmbeddingAIcapabilitiesAIcomponentsAImodelsPreprocessingPostprocessingOutputFigure8:ClassificationofAItermsAImodeldividedintophasesArtificialNarrowIntelligence(ANI)GenerativeArtificalIntelligenceArtificialGeneralIntelligence(AGI)ArtificialSuperIntelligence(ASI)TodayInthenextfewyearsExample:Siri,AlexaExample:OpenAIMachineLearningMachineIntelligenceMachineConsciousnessReturnsanswersbasedonexistingdata,basedonalgorithmsandstatisticaltechniques.AIcancreatedatainstancesindependently.Notyetclosetohumanintelligence.Comparablewithacomputerthatcomesclosetohumanintelligenceinallareas.Intellectthatissuperiortohumansinmanyorareas.AIcannotcreatedatainstancesindependently.Table1:ThedevelopmentofAImodelsdividedintodifferenttimephases27AItechsimpletoolsthatcanbeusedtoshowthepossibleusesofAI.Themethodallowsthedocumentationofapplicationexamples,addedandlimitsofAIapplications.Inyears,thecapabilitiesofAImodelshavecontinuouslyForsometasks,thecomingclosetohumancapabilities.InabenchmarktestcalledMassiveMultitaskLanguagederstanding(MMLU),whichwasdevelopedtoatelanguagemodels,modelsachievingcuracyvaluesof86to94Thehumanopinionis89.8percent.ThismeansthefirstAIperformingbetterthanhumanexperts(iAsk,TheMMLUtestconsistsof57tasksthefieldsofmathematics,UShistory,andMeasuringMassiveMultitaskLanguageUnderstanding(MMLU)Ranking100%93.8995% ExpertAGI:93.0%93.8990%85%80%75%GPT-4oGeminiGPT-4oGeminiGPT-4Qwen2MistralLlama3Yi-largeProInstructLarge2

Humanexpert:89.8% 86.584.584.282.782.482.480.079.379.3iAskProClaude3.586.584.584.282.782.482.480.079.379.3Sonnet

Llama3.1TurboFigure9:TheperformanceofAImodelscomparedtohumancapabilitiesintheMMLUtest(iAsk,2024)28AutomobileManufacturersWithAIInvestmentBasedondifferentresearch(Shilov,2023),(Mu,itcanbeassumedthatNVIDIAsoldapprox.500,000to620,000H100andA100chipsin2023.450,000ofthechipswerepurchasedbybigtechhyperscalercompaniesincludingAmazon,Meta,andAlphabet.Advanceddriverassistancesystems(ADAS)supportdriversincertaindrivingsituations,in-creasesafety,andenhancedrivingcomfort.Automated/autonomousdriving(AD):self-drivingvehiclesInferenceistheprocessthatatrainedmachinelearningmodelusestodrawconclusionsfromnewdata.Example:Aself-drivingcarrecognizesastopsignonaroadithasneverdrivenonbefore.TheabilitytorecognizethisstopsigninacompletelynewcontextindicatesanAImodelcapableofmak-inginferences.OneofthefewautomotivemanufacturerstoupcomputingpoweronalargescaleisByendof2024,thenumberofH100GPUsisedtoincreasefrom35,000to85,000(Walz,2024).ThesupercomputeratwillbeusedtotrainmodelsfortheADAS(advanceddriverassistancesystems),amongotherthings.TheinferenceofAdvanceddriverassistancesystems(ADAS)supportdriversincertaindrivingsituations,in-creasesafety,andenhancedrivingcomfort.Automated/autonomousdriving(AD):self-drivingvehiclesInferenceistheprocessthatatrainedmachinelearningmodelusestodrawconclusionsfromnewdata.Example:Aself-drivingcarrecognizesastopsignonaroadithasneverdrivenonbefore.TheabilitytorecognizethisstopsigninacompletelynewcontextindicatesanAImodelcapableofmak-inginferences.OtherautomotivemanufacturershavesofarbeenluctanttoinvestintheirowndatacenterswhichbeusedforAIapplications.ItistobeassumedtheywillusethedatacentersoflargetechnologypaniestotrainAIapplicationsandrunthem

ForecastspredictthatthedemandforcomputingpacityfortheinferenceofmodelswillincreaseinlongtermandthatthedemandforAItrainingwillclineinrelativeterms.Inferencecantakeplaceinthecloudorinprocessingunitsinthevehicles.partofthedevelopmentofsoftware-definedvehicles(SDV),thecomputingpowerinstalledinthevehicleswillcontinuallyincrease,whichispositivefortheofAIapplicationsinvehicles.Distributionoftheofcomputingpowerinthecloudorinthevehiclede-pendsverymuchontheusecases.AIModels:MakeorBuy?Anumberoflargelanguagemodelsareinexistencealready,andagreatdealoftimeandmoneyhasspentondeveloping