Tesla (TSLA) Q1 2024 Earnings Call Transcript财报电话会议记录.docx

Tesla,Inc.(TS1.A)Q12024EarningsCallTranscriptApr.23,202410:30PMETTesla,Inc.(TS1.A)Stock,TS1.A:CAStockCompanyParticipantsMartinViecha-VicePresident-InvestorRelationsElonMusk-ChiefExecutiveOfficerVaibhavTaneja-ChiefFinancialOfficer1.arsMoravy-VicePresident-VehicleEngineeringAshokElluswamy-Director-AutopilotSoftwareConferenceCallParticipantsTonySacconaghi-BernsteinAdamJonas-MorganStanleyAlexPotter-PiperSandlerMarkDelaney-GoldmanSachsColinRusch-OppenheimerBenKallo-BairdShreyasPatil-WolfeResearchMartinViechaTesla'sFirstQuarter2024Q&AWebcast.MynameisMartinViecha,VPofInvestorRelations,andmjoinedtodaybyElonMusk,VaibhavTaneja,andanumberofotherexecutives.OurQ1resultswereannouncedatabout3.00p.m.CentralTimeintheUpdateDeckwepublishedatthesamelinkasthiswebcast.Duringthiscall,wewilldiscussourbusinessoutlookandmakeforward-lookingstatements.Thesecommentsarebasedonourpredictionsandexpectationsasoftoday.Actualeventsandresultscoulddiffermateriallyduetoanumberofrisksanduncertainties,includingthosementionedinourmostrecentfilingswiththeSEC.Duringthequestion-and-answerportionoftoday'scall,pleaselimityourselftoonequestionandonefollow-up.Pleaseusetheraisehandbuttontojointhequestionqueue.ButbeforewejumpintoQ&A,Elonhassomeopeningremarks.Elon?ElonMuskThanks,Martin.SotorecapinQ1wenavigatedseveralunforeseenchallengesaswellastherampoftheupdatedModel3inFremont.Therewas,asweallhaveseen,theEVadoptionrategloballyisunderpressureandalotofotherautomanufacturersarepullingbackonEVsandpursuingplug-inhybridsinstead.Webelievethisisnottherightstrategyandelectricvehicleswillultimatelydominatethemarket.Despitethesechallenges,theTeslateamdidagreatjobexecutinginatoughenvironmentandenergystoragedeployments,theMegapackinparticular,reachedanalltimehighinQ1,leadingtorecordprofitabilityfortheenergybusiness,andthatlookslikelytocontinuetoincreaseinthequartersandyearsahead.Itwillincrease.Weactuallyknowthatitwill,sosignificantlyfasterthanthecarbusinessasweexpected.WealsocontinuetoexpandourAltrainingcapacityinQ1,morethandoublingourtrainingcomputesequentially.Intermsofthenewproductroadmap,therehasbeenalotoftalkaboutourupcomingvehiclelineinthenext-inthepastseveralweeks.We'veupdatedourfuturevehiclelineuptoacceleratethelaunchofnewmodelsahead,previouslymentionedstartupproductioninthesecondhalfof2025,soweexpectittobemoreliketheearly2025,ifnotlatethisyear.Thesenewvehicles,includingmoreaffordablemodels,willuseaspectsofthenextgenerationplatformaswellasaspectsofourcurrentplatforms,andwillbeabletoproduceonthesamemanufacturinglinesasourcurrentvehiclelineup.Soit'snotcontingentonanynewfactoryormassivenewproductionline.It'llbemadeonourcurrentproductionlinesmuchmoreefficiently.Andwethinkthisshouldallowustogettoover3millionvehiclesofcapacitywhenrealizedtothefullextent.RegardingFSDVersion12,whichisthepureAl-basedself-driving,ifyouhaven'texperiencedthis,Istronglyurgeyoutotryitout.It'sprofoundandtherateofimprovementisrapidso-andwe,venowturnedthatonforallcarswiththecamerasandinferencecomputerandeverythingfromHardware3oninNorthAmerica.Andsoit'sbeenpushedoutto,Ithink,around1.8millionvehiclesandwe'reseeingabouthalfofpeopleuseitsofarandthatpercentageisincreasingwitheachpassingweek.Sowenowhaveover300billionmilesthathavebeendrivenwithFSDV12.Sincethelaunchoffullself-driving,supervisedfullself-driving,it'sbecomeveryclearthatthevision-basedapproachwithendtoendneuralnetworksistherightsolutionforscalableautonomy.It'sreallyhowhumansdrive.Ourentireroadnetworkisdesignedforbiologicalneuralnetsandeyes.Sonaturallycamerasanddigitalneuralnetsarethesolutiontoourcurrentroadsystem.Tomakeitmoreaccessible,we'vereducedthesubscriptionpriceto$99amonth,soit'seasytotryout.Andaswe'veannounced,we'llbeshowcasingourpurpose-builtrobotaxi,orCybercab,inAugust.Yes.RegardingAlcompute,overthepastfewmonths,we'vebeenactivelyworkingonexpandingTesla'scoreAlinfrastructure.Forawhilethere,weweretrainingconstrainedinourprogress.Weare,atthispoint,nolongertrainingconstrainedandsowe'remakingrapidprogress.We'veinstalledandcommissioned,meaningthey'reactuallyworking35,000H100computersorGPUs,GPUiswrongword,theyneedanewword.IalwaysfeellikeawincewhenIsayGPUbecauseit'snot-GPUstands-Gstandsforgraphics,anditdoesn'tdographics.Butanywayroughly35,000H100sareactive,andweexpectthattobeprobably85,000orthereaboutsbytheendofthisyearandtraining,justfortraining.Wearemakingsurethatwe'rebeingasefficientaspossibleinourtraining.It'snotjustaboutthenumberofH100s,buthowefficientlythey'reused.Soinconclusion,we'resuperexcitedaboutourautonomyroadmap.Ithinkitshouldbeobvioustoanyonewho'sdrivingVersion12andittellsthatthatitisonlyamatteroftimebeforeweexceedthereliabilityofhumansandnotmuchtimewiththat.Andwe'rereallyheadedforanelectricvehicle,anautonomousfuture.AndllgobacktosomethingIsaidseveralyearsagothatinthefuture,gasolinecarsthatarenotautonomouswillbelikeridingahorseandusingaflipphone.Andthatwillbecomeveryobviousinhindsight.WecontinuetomakethenecessaryinvestmentsthatwilldrivegrowthandprofitsforTeslainthefuture,andIwantedtothanktheTeslateamforincredibleexecutionduringthisperiodandlookforwardtoeverythingthatwehaveplannedahead.Thanks.MartinViechaThankyouverymuch,andVaibhavhassomecommentsaswell.VaibhavTanejaThanks.It'simportanttoacknowledgewhatElonsaid,fromourautobusinessperspective.Wedidseeaseasonaldeclineinrevenuesquarter-over-quarterandthosewereprimarilybecauseofseasonality,uncertainmacroeconomicenvironmentandtheotherreasons,whichElonhadmentionedearlier.Automarginsdeclinedfrom18.9%to18.5%.ExcludingtheimpactofCybertruck,theimpactofpricingactionswaslargelyoffsetbyreductionsinperunitcostsandtherecognitionofrevenuefromAutoparkfeatureforcertainvehiclesintheU.S.thatpreviouslydidnothavethatfunctionality.Additionally,whilewedidexperiencehighercostduetotherampofModel3inFremontanddisruptionsinBerlin,thesecostswerelargelyoffsetbycostreductioninitiatives.Infact,ifweexcludeCybertruckandFremontModel3rampcosts,therevenuefromAutopark,automarginsimprovedslightly.CurrentlynormalizedModelYcostpervehicleinAustinandBerlinarealreadyveryclosetothatofFremont.Ourabilitytoreducecostswithoutsacrificingonqualitywasduetotheamazingeffortsoftheteam,inexecutingTesla'srelentlesspursuitofefficiencyacrossthebusiness.We'vealsowitnessedthatasotherOEMsarepullingbackontheirinvestmentsinEV,thereisincreasingappetiteforcredits,andthatmeansasteadystreamofrevenueforus.Obviously,seeingotherspullbackfromEVisnotthefuturewewant.Wewouldpreferitthewholeindustrywentallin.Onthedemandfront,we'veundertakenavarietyofinitiatives,includingloweringthepriceofboththepurchaseandsubscriptionoptionsforFSDlaunchingextremelyattractiveleasingspecialsfortheModel3intheU.S.for$299amonthandofferingattractivefinancingoptionsincertainmarkets.Webelievethatourawarenessactivities,pairedwithattractivefinancing,willgoalongwayinexpandingourreachanddrivingdemandforourproducts.OurEnergybusinesscontinuestomakemeaningfulprogresswithmarginsreachingarecordof24.6%.Weexpecttheenergystoragedeploymentsfor2024togrowatleast75%higherfrom2023.Andaccordingly,thisbusinesswillbegincontributingsignificantlytoouroverallprofitability.NotethatthereisabitofIumpinessinourstoragedeploymentsduetoavarietyoffactorsthatareoutsideofourcontrol,sodeploymentsmayfluctuatequarter-over-quarter.Ontheoperatingexpensefront,wesawasequentialincreasefromourAlinitiatives,continuedinvestmentinfutureprojects,marketingandotheractivities.Wehadnegativefreecashflowof$2.5billioninthefirstquarter.Theprimarydriverofthiswasanincreaseininventoryfromamismatchbetweenbuildsanddeliveriesasdiscussedbefore,andourelevatedspendonCapExacrossvariousinitiatives,includingAlcompute.Weexpecttheinventorybuildtoreverseinthesecondquarterandfreecashflowtoreturntopositiveagain.Aswepreparethecompanyforthenextphaseofgrowth,wehadtomakethehardbutnecessarydecisiontoreduceourheadcountbyover10%.Thesavingsgeneratedareexpectedtobewellinexcessof$1billiononanannualrunratebasis.WearealsogettinghyperfocusedonCapExefficiencyandutilizingourinstalledcapacityinamoreefficientmanner.Thesavingsfromtheseinitiatives,includingourcostreductionswillhelpimproveouroverallprofitabilityandultimatelyenableustoincreasethescaleofourinvestmentsinAl.Inconclusion,thefutureisextremelybrightandthejourneytogettherewhilechallengingwillbeextremelyrewarding.Onceagain,IwouldliketothankthewholeTeslateamfordeliveringgreatresults.AndwecanopenituptoQ&A.Question-and-AnswerSessionA-MartinViechaOkay.1.et'sstartwithinvestorQ&A.Thefirstquestionis,whatisthestatusof4680.Whatisthecurrentoutput?1.ars?1.arsMoravySure.4680productionincreasedabout18%to20%fromQ4reachinggreaterthan1KaweekforCybertruck,whichisabout7gigawatthoursperyearaswepostedonX.WeexpecttostayaheadoftheCybertruckrampwiththecellproductionthroughoutQ2aswerampthethirdoffourlinesinPhase1,whilemaintainingmultipleweeksofcellinventorytomakesurewe'reaheadoftheramp.Becausewe'reramping,COGScontinuestodroprapidlyweek-overweekdrivenbyyieldimprovementsthroughoutthelinesandproductionvolumeincreases.Soourgoal,andweexpecttodothisistobeatsuppliercostofnickel-basedcellsbytheendoftheyear.MartinViechaThankyou.ThesecondquestionisonOptimus.SowhatisthecurrentstatusofOptimus?Aretheycurrentlyperforminganyfactorytasks?Whendoyouexpecttostartmassproduction?ElonMuskWeareabletodosimplefactorytasksoratleast,Ishouldsay,factorytasksinthelab.Intermsof-wedothinkwewillhaveOptimusinlimitedproductioninthenaturalfactoryitself,doingusefultasksbeforetheendofthisyear.AndthenIthinkwemaybeabletosellitexternallybytheendofnextyear.Thesearejustguesses.Asvesaidbefore,IthinkOptimuswillbemorevaluablethaneverythingelsecombined.Becauseifyou'vegotasentienthumanoidrobotsthatisabletonavigaterealityanddotasksatrequest,thereisnomeaningfullimittothesizeoftheeconomy.Sothat'swhatisgoingtohappen.AndIthinkTeslaisbestpositionedofanyhumanoidrobotmakertobeabletoreachvolumeproductionwithefficientinferenceontherobotitself.ImeanthisperhapsisapointthatisworthemphasizingTesla'sAlinferenceefficiencyisvastlybetterthananyothercompany.ThereisnocompanyevenclosetotheinferenceefficiencyofTesla.We'vehadtodothatbecausewewereconstrainedbytheinferencehardwareinthecar,wedidn'thaveachoice.Butthatwillpaydividendsinmanyways.MartinViechaThankyou.Thethirdquestionis,whatisthecurrentassessmentofthepathwaytowardsregulatoryapprovalforunsupervisedFSDintheU.S.Andhowshouldwethinkabouttheappropriatesafetythresholdcomparedtohumandrivers?ElonMuskSure.1.arsMoravyIcanstart.Thereareahandfulofstatesthatalreadyhaveadoptedautonomousvehiclelaws.Thesestatesarepavingthewayforoperations,whilethedataforsuchoperationsguidesabroaderadoptionofdriver-lessvehicles.IthinkAshokcantalkalittlebitaboutoursafetymethodology,butweexpectthatthesestatesandtheworkongoingaswellasthedatathatwe,reprovidingwillpaveawayforabroad-basedregulatoryapprovalintheU.S.atleastandtheninothercountriesaswell?AshokElluswamyYes.It'sactuallybeenprettyhelpfulthatotherautonomouscarcompanieshavebeencuttingapaththroughtheregulatoryjungle,whichisabsurd.That'sactuallyquitehelpful.AndtheyhaveobviouslybeenoperatinginSanFranciscoforawhile.IthinktheygotapprovalforCityof1.A.Sotheseapprovalsarehappeningrapidly.Ithinkifyou'vegotatscale,astatisticallysignificantamountofdatathatshowsconclusivelythattheautonomouscarhas,let'ssay,halftheaccidentrateofahuman-drivencar,Ithink,that'sdifficulttoignorebecauseatthatpoint,stoppingautonomymeanskillingpeople.SoIactuallydonotthinkthattherewillbesignificantregulatorybarriersprovidedtherewasconclusivedatathattheautonomouscarissaferthanahuman-drivencar.Andinmyview,thiswillbemuchlikeelevators.Elevatorsusedtobeoperatedbyaguywithrelayswitch.Butsometimesthatguywouldgettiredordrunkorjustmakeamistake,andshattersomebodyinhalfbetweenfloors.Sowejustgetanelevatorandpressbutton,wedon'tthinkaboutit.Infact,it'skindofweirdifsomebodyisstandingtherewitharelayswitch.Andthatwillbehowcarswork.Youjustsummonthecarusingyourphone,yougetin,ittakesyoutoadestination,yougetout.VaibhavTanejaYoudon'teventhinkaboutit?ElonMuskYoudon'teventhinkaboutit.Justlikeanelevator,ittakesyoutoyourfloor.That'sit.Don'tthinkabouthowtheelevatorisworkingoranythinglikethat.AndsomethingIshouldclarifyisthatTeslawillbeoperatingthefleet.SoyoucanthinkoflikehowTesla,thinkofit'slikesomecombinationofAirbnbandUber,meaningthattherewillbesomenumberofcarsthatTeslaownsitselfandoperatesinthefleet.Therewillbesomenumberofcarsandthenthere'llbeabunchofcarswherethey'reownedbytheenduser.Thatendusercanaddorsubtracttheircartothefleetwhenevertheywant,andtheycandecideiftheywanttoonlyletthecarbeusedbyfriendsandfamilyoronlyby5-starusersorbyanyoneatanytimetheycouldhavethecarcomebacktothemandbeexclusivelytheirs,likeanAirbnb.Youcouldrentoutyourguestroomornot,anytimeyouwant.Soasourfleetgrows,wehave7millioncarsgoingto-9millioncarsgoingto,eventuallytensofmillionsofcarsworldwide.Withaconstantfeedbackloop,everytimesomethinggoeswrong,thatgetsaddedtothetrainingdataandyougetthistrainingflywheelhappeninginthesamewaythatGoogleSearchhasthesortOfflywheel,it'sverydifficulttocompetewithGooglebecausepeopleareconstantlydoingsearchesandclickingandGoogleisgettingthatfeedbackloop.IfsthesamewithTesla.Butatascalethatismaybedifficulttocomprehend,butultimately,itwillbetensofmillions.Ithinkthere'salsosomepotentialhereforanAWSelementdowntheroadwhereifwe,vegotverypowerfulinferencebecausewe'vegotaHardware3inthecars,butnowallcarsarebeingmadewithHardware4.Hardware5isprettymuchdesignedandshouldbeincars,hopefullytowardstheendofnextyear.Andthere'sapotentialtorun-whenthecarisnotmovingtoactuallyrundistributedinference.SokindoflikeAWS,butdistributedinference.1.ikeittakesalotofcomputerstotrainanAlmodel,butmanyordersofmagnitudelesscomputetorunit.Soifyoucanimaginefuture,perhapswherethere'safleetof100millionTeslas,andonaverage,they'vegotlikemaybeakilowattofinferencecompute.That's100gigawattsofinferencecomputedistributedallaroundtheworld.It'sprettyhardtoputtogether100gigawattsofAlcompute.Andeveninanautonomousfuturewherethecaris,perhaps,usedinsteadofbeingused10hoursaweek,itisused50hoursaweek.Thatstillleavesover100hoursaweekwherethecarinferencecomputercouldbedoingsomethingelse.Anditseemslikeitwillbeawastenottouseit.MartinViechaAshok,doyouwanttochimeinontheairprocessandsafety?AshokElluswamyYes,wehavemultipletiersofvalidatingthesafetyinanygivenweek,wetrainhundredsofneuralnetworksthatcanproducedifferenttrajectoriesforhowtodrivethecar,wereplaythemthroughthemillionsofclipsthatwehavealreadycollectedfromourusersandourownQA.Thosearelikecriticalevents,likesomeonejumpingoutinfrontorlikeothercriticaleventsthatwehavegathereddatabaseovermany,manyyears,andwereplaythroughallofthemtomakesurethatwearenetimprovingsafety.Andontopofit,wehavesimulationsystemsthatalsotrytorecreatethisandtestthisinclosedloopfashion.Andsomeofthisisvalidated,wegiveittoourownQAdrivers.Wehavehundredsofthemindifferentcities,inSanFrancisco,1.osAngeles,Austin,NewYork,alotofdifferentlocations.Theyarealsodrivingthisandcollectingreal-worldmiles,andwehaveanestimateofwhatarethecriticalevents,aretheyanetimprovementcomparedtothepreviousweek'sbuilds.Ando