DeutscheBankResearchTheupcomingAIBoominChinaDBChinaEconomics&AsiaCompanyResearchMarch2025DistributedDistributedon:26/03/2025DeutscheBankResearchIMPORTANTRESEARCHDISCLOSURESANDANALYSTCERTIFICATIONSLOCATEDINAPPENDIX1.UNTIL19thMARCH2021INCOMPLETEDISCLOSUREINFORMATIONMAYHAVEBEENDISPLAYED,PLEASESEEAPPENDIX1FORFURTHERDETAILS.Outlines口WhyisAIWhyisAIadoptionacceleratinginChina?Which?LowercostenablesrapidexpansionofAIinvestment?China’scompetitiveedgeinAIadoption?ChartingthepathforemergingAIindustriessectorssectorsshouldbenefitthemostfromAI??AIInfrastructure:semiconductor,cloud,datacenter,energy?AIapplications:mega-apps,AIagent,applicationtohealthcare/financial/consumersectors?EmbodiedAI:Intelligentcars,humanoidrobotsEconomicsYiXiongDeyunOuCompanyResearchPeterMillikenLeoChiangCyrusNgBinWangJohnnyXieJessieXuHanZhangIrisZhengGaryZhou1DeutscheBank1ResearchSummaryChina’sadoptionofAIhassuddenlyaccelerated.DeepSeek'sreleaseservedasacatalyst,shiftingthefocusfromdevelopingbiggermodelstoreal-worldapplications.ThisalsotriggeredasurgeinAIinvestmentacrossChineseindustries.Whyisthishappeningnow?Inthisreport,wediscusswhyChinesecompanieslikelypossessacompetitiveadvantageincertainapplicationsofAI.Firstly,AIapplicationrequirescross-industryinnovation,combiningAIwiththeexpertiseanddatafromsectorssuchassmartphones,mega-apps,andelectricvehicles.ThisisfurtheramplifiedbyChina’scomprehensiveandagileindustrialsupplychain,enablingrapidprototypinganditerationofAI-poweredproducts.Furthermore,adynamicdomesticmarketprovidesafertiletestinggroundfornewapplications,whilealargepoolofAItalentsandengineersfuelsdevelopment.TheneartermprioritizesAIinfrastructure,includingsemiconductors,cloudcomputing,datacenters,andenergysolutions.Thenear-tomedium-termfocusesonAIapplicationslikemega-apps,AIagents,andintegrationwithinhealthcare,consumer,andfinancialindustries.Lookingfurtherahead,themedium-tolong-termvisioncentersonembodiedAI,connectingAItothephysicalworldthroughvesselssuchasintelligentcarsandhumanoidrobots.2DeutscheBank2ResearchWhyisAIadoptionacceleratinginChina?DeutscheBankResearchThe“Androidmoment”andChina’scompetitiveedgeindisruptiveinnovationinnovationTounderstandthepotentialofAIinnovationinChina,considertheparalleloftheChinesesmartphoneindustry'sriseinthe2010s,whenGoogle’sinnovationinAndroidOSfueledadecadeofexplosivegrowthinsmartphones.AhandfulofChinesecompaniesemergedandhavenowbecometopsmartphonemakersinChinaandglobally.Thissuccessstoryoffersseveralkeylessons:1)Dramaticcostreduction:Android'sopen-sourcenaturedrasticallyreduceddevelopmentcosts,drivingexponentialgrowthinsmartphonedemand.2)Intensecompetition:numerouscompaniesenteredthemarket,fosteringahighlydynamicandcompetitiveenvironment.3)Rapiditeration:LeveragingChina'sextensiveindustrialbaseandreadilyavailablesuppliers,manufacturerssignificantlyshortenedproductdevelopmentcycles,rapidlyintroducingnewfeaturesandincorporatingcustomerfeedbackwhilesimultaneouslycuttingcosts.WebelievetheforthcomingAIrevolutioninChinawilllikelyexhibitsimilarcharacteristics.AndroidAndroidOSacceleratedsmartphonesalesglobally…Globalsmartphoneshipment,100mnunits50Android‘07‘08‘09‘10‘11‘12‘13‘14‘15‘16‘17‘18‘1…with…withChinesephonemakerswinninglargemarketshares%ofsmartphoneshipmentinChina0Datasource:DeutscheBankResearch,IDC,statcounterDatasource:DeutscheBankResearch,ChinaMITTDeutscheBankResearchYiXiong/DeutscheBankResearch4(+852)22036139|yi.xiong@/deyun.ou@4DeutscheBankResearchYiXiong/DeyunDeutscheBankResearch5(+852)22036139|yi.xiong@/deyun.ou@5DeepSeekisenablingrapidexpansionofAIusersandapplications?DeepSeekissmartyetaffordable.TheDeepSeekV3andR1open-sourcemodelsoffercomparableperformancetoleadingLLMsatafractionofthecost,significantlyloweringthebarriertoentryforanyonetointegrateAIintoitsownproducts.?Rapidexpansionofusers.DeepSeekwaswidelyreportedtohavereached100milliondownloadswithinweeksoflaunchandhasaround50millionofactivedailyusersasoftoday.OtherAIappsinChina,suchasBytedance'sDoubaoandTencent'sYuanbao,havealsoseenacceleratedusergrowth.?Open-sourceAImodelshavebecomemainstream.DeepSeek‘sopen-sourcingofmodelweights,methodologypapers,andcodeacceleratestheoveralldevelopmentoftheopen-sourceAIecosystem.ManyChineseAIdevelopershavereleasedtheirownreasoningmodelsfollowingDeepSeekR1.LLMsarenolongeraluxurybuthavebecomeacommodity.AccelerationofReasoningModelsinChinaTencentQwenQwQ32BMoonshot?Notjustachatbot.DeepSeek’simprovementstoTransformerAIarchitecturearen’tjustforchatbots,butcanalsobeAccelerationofReasoningModelsinChinaTencentQwenQwQ32BMoonshotDeepSeeksignificantDeepSeeksignificantlyliftedperformanceofopen-sourceLLMsQwQQwen2.5MaxQwen2.5Max22Datasource:DeutscheBankResearch,artificialanalysisDatasource:DeutscheBankResearch,ChinalocalnewsDeutscheBankResearchYiXiong/DeyunDeutscheBankResearch6(+852)22036139|yi.xiong@/deyun.ou@6ByebyeScalingLaw,HelloJevonsParadox:ChinesetechfirmsarescalingupAIinvestmentWhenUSWhenUStechcompaniesscaledupinvestmentinGenerativeAIinrecentyears,Chinesetechcompaniesinitiallylagged.Butachangeisunderway:investmentbyChinesetechcompaniesacceleratedsignificantlyinQ42024.Why??Diminishingreturnsofscale:theScalingLawinAI,wherelargermodelsconsistentlyyieldedbetterperformance,ismaybereachingapointofdiminishingreturnsasrecentreleasesoflargermodelshavenotgeneratedsufficientperformanceimprovements.Thisshiftinthecost-benefitdynamiccreatesanopportunityforalternativeapproaches.?Jevons'Paradoxdrivesdemand:AsAImodeldevelopmentcostsdecrease,theaccessibilityofAIincreases,leadingtoasurgeinconsumerdemand.Thisincreaseddemandisnotnecessarilyforthemostsophisticatedmodels,butratherforefficient,practicalapplicationsofAIthatmeetspecificconsumerneeds.Insteadofsolelypursuinglargerandmorecomplexmodels,ChinesetechcompaniesareprioritizingthedevelopmentofmoreefficientmodelstailoredtomeettheburgeoningconsumerdemandforpracticalAIapplications.Thisfocusonefficiencyandapplicationallowsthemtocapitalizeonthegrowingmarketopportunitieswhilemanagingcostseffectively.CapexbyChinesetechshaslaggedUSpeerssince2021.Theyarenowcatchingup86420Capex,topChinatechcompaniesCapex,topUStechcompanies(rhs)0Alibaba’splannedinvestmentinAIandcloudin2025-27willexceeditscapexinthepastdecadeRMBbnAlibabaCAPEXRmb380bn14012010080604020FY12FY13FY14FY15FY16FY17FY18FY19FY20FYFY12FY13FY14FY15FY16FY17FY18FY19FY20FY21FY22FY23FY24FY25EFY26EFY27EDatasource:DeutscheBankResearch,BloombergFinanceLPDatasource:DeutscheBankResearchDeutscheBankResearchYiXiong/DeyunDeutscheBankResearch7(+852)22036139|yi.xiong@/deyun.ou@7China’scompetitiveedgeinAI:cross-industryinnovation??AIapplicationrequirescross-fieldinnovation.AI'struepotentialunfoldswhenappliedacrossdiversesectors.Thisprocesstakestimeandresources,particularlybecauseeffectiveAIdeploymenthingesonvastquantitiesofdata,oftengeneratedonlyduringreal-worldusage.Thiscreatesacyclicaldependencywhereapplicationfuelsdatacreation,whichinturnrefinestheAI,leadingtofurtheradvancements.?Chinaiswell-positionedtocapitalizeonthiscross-industrydynamic.ItholdssignificantmarketshareinkeysectorspoisedforAItransformation,includingsmartphones,mega-apps,cloudcomputing,electricvehicles,androbotics.Theseindustriesgeneratemassivedatasets,providingChinesecompanieswithacrucialdataadvantage.?Agilesupplychainsenablerapiditeration.Theabilitytorapidlyprototype,test,anditerateonhardwareandsoftwarecomponentsallowsChinesecompaniestoquicklytranslateAIadvancementsintotangibleproducts.Thisspeedandflexibilityarecrucialinthefast-evolvingAIlandscape,enablingcompaniestoadapttomarketdemandsandmaintainacompetitiveedge.ThiscombinationofdataadvantageandmanufacturingprowesspositionsChinaasasignificantplayerintheglobalAIrace,inourview.MarketshareofChineseMarketshareofChineseproducersinindustriesthatareexpectedtobenefitfromAIrevolutionAutosCloudservicesMega-appsSmartphonesElectricvehiclesRoboticsAImodelsDatasource:DeutscheBankResearchDeutscheBankResearchYiXiong/DeyunDeutscheBankResearch8(+852)22036139|yi.xiong@/deyun.ou@8China’scompetitiveedgeinAI:adynamicconsumermarket??Massivemarketfuelsdataadvantage:China'senormousconsumermarketprovidesavastpoolofdatacrucialfortrainingandrefiningAImodels.ThisdataadvantagecouldallowChinesecompaniestodevelophighlytailoredandeffectiveAIsolutions,givingthemasignificantedgeinalgorithmdevelopmentandperformance.?Consumerreceptivenessdrivesrapidadoption:Chineseconsumersareknownfortheirearlyadoptionofnewtechnologies,asdemonstratedbytherapiduptakeofsmartphonesandelectricvehicles.ThiswillingnesstoembracenewtechnologiesallowsChinesecompaniestoquicklyiterateandrefinetheirAIofferings,gainingafirst-moveradvantageinemergingAImarkets.?Intensecompetitionfostersinnovationandknowledgediffusion:ThehighlycompetitivenatureoftheChineseconsumermarketpushescompaniestoconstantlyinnovateandimprovetheirproductsandservices.Thisdynamicenvironmentfostersrapidknowledgediffusionandacceleratesthedevelopmentofcutting-edgeAItechnologies.FromFromsmartphonestoEVs,Chineseconsumersareknownfortheirearlyadoptionofnewtechnologies0Smartphonesalesas%oft50Datasource:DeutscheBankResearchDatasource:DeutscheBankResearchDeutscheBankResearchYiXiong/DeyunDeutscheBankResearch9(+852)22036139|yi.xiong@/deyun.ou@9China’scompetitiveedgeinAI:AItalentsandengineers??AlthoughtheUSremainedtheworld’stopdestinationfortopAItalentstowork,Chinaisthetopcountryoforiginfortop-tierAIresearchers.ThismeansifChineseuniversitiesandcompaniescanofferresearchconditionsandjobopportunitiesforAItalentscomparabletotheUScounterparts(whichisincreasinglythecase),itshouldn’tbedifficultforthemtofindenoughtalent.DeepSeek’soverwhelminglydomestic-educatedteamwasacaseinpoint.?TheadoptionofAItoreal-worldproductsrequiresnotonlyonAI-specificknowledgebutalsohowtointegrateAIintoexistingproducts.Chinaisthelargestsupplierofengineers;thesizeofSTEMgraduatesfromChineseuniversitieseveryyearislargerthanthosefromUS,France,GermanyandJapancombined.ChineseEVmanufacturerBYDcurrentlyemploys~110,0000engineers,whichisalmostequivalenttoTesla’stotalworkforce.Chinsecompaniescouldleveragetheirvastengineeringworkforcetoprototype,test,andscaleAIsolutionsfasterthancompetitors.AlmostAlmosthalfofAItoptalentsgraduatedfromChineseuniversitiesTop20%AIresearchersbycountryofundergraduatedegrees0852425ChinaChinaalsoleadstheworldinthesupplyofengineersMexico0.24Datasource:DeutscheBankResearch,PaulsonInstituteDatasource:DeutscheBankResearch,Olissandothers(2023)DeutscheBankResearchYiXiong/DeyunDeutscheBankResearch10(+852)22036139|yi.xiong@/deyun.ou@10ChartingthepathforemergingAIindustriesinChinaEmbodiedEmbodiedAIAIAppsIntelligentVehiclesHumanoidRobotsFinancialservicesHealthcareservicesCateringservicesSemiconductorsCloudservicesDatacentersEnergyAI-enhancedmega-appsAIagentsLookingforward,wethinktheAIrevolutioninChinawilllikelyunfoldinthreestages:1.AIInfrastructure:TheinitialphasewillbemarkedbyasurgeindemandforAIinfrastructure,includingcloudcomputingservices,datacenters,andspecializedhardwarelikeGPUs.ThisfoundationallayerwillpavethewayforthedevelopmentanddeploymentofincreasinglysophisticatedAImodels.2.AI-PoweredApplications:ThesecondstagewillwitnessaproliferationofAI-poweredapplicationstransformingbothconsumerhabitsandworkingprocedures.Whilethisencompassesawiderangeofsectors,inthecontextofChina,wewillfocusontheapplicationofAItoexistingmega-apps,theemergenceofAIagents,andapplicationswithinfinancialandhealthcareservices.3.TheRiseofEmbodiedIntelligence:Inthelongerterm,AIwillbeintegratedintothephysicalworld,unlockingavastarrayofnewpossibilities.WebelievethisistheareainwhichChinaismostlikelytoexcel.Ourfocuswillbeontwoparticularlypromisingfields:intelligentvehiclesandhumanoidrobots.Datasource:DeutscheBankResearchDeutscheBankResearchYiXiong/DeyunDeutscheBankResearch11(+852)22036139|yi.xiong@/deyun.ou@11Risks:whatcouldgowrongforChina’sAIdevelopment?ChipsBottleneck:tightenedU.S.exportrestrictionscreateacriticalshortageofadvancedAIchipsinChina,limitingcomputationalpowerwhilealsohinderingChina’schipeffortsindevelopingandmanufacturingchipsdomestically.TechnologyBottleneck:CurrentgenerativeAItechnologyfaceslimitations,hinderingpracticalapplicationstotherealworld. AGIEmergenceElsewhere:amajorAIbreakthrough(e.gemergenceofa powerfulArtificialGeneralIntelligence)erodesChina’scompetitiveedgeRegulatoryEnvironment:TighteningregulationsonAIdevelopment,drivenbyethicalconcernsandsecurityconsiderations,stifleinnovationandslowdownprogress.Protectionism:Nationalsecurity,dataprivacy,andprotectionismconcerns,resultinginwidespreadprotectionismoncross-borderflowsofAIproductsinternationally.SocialStability:RapiddisplacementofhumanlaborbyAIleadstowidespreadunemployment,triggeringeconomicandsocialinstability.AIInfrastructure:Semiconductor,Cloud,Datacenter,EnergyDeutscheBankResearchUSrestrictionsonchipsandAIareheadwindstoChina’sAIinnovation…??TheUSgovernmenthasimposedaseriesofrestrictionsrelatedtohigh-performancesemiconductorsonChinasince2019.ThesemeasuresincludebansonsellingadvancedchipprocessingequipmenttoChina,restrictionsonchipexports,limitationsoninvestmentinChineseproductionplants,andevenconstraintsontechnologyexchange.?Since2023,theserestrictionshavebecomemorefocusedontheAIsector,withtheUSimplementingbroaderexportcontrolcriteriatofurthertightenitsgrip.MostofNvidia’shigh-performanceAIchips,includingthosebasedonthelatestBlackwellplatform,cannotbeexportedtoChina.?WhilethedowngradedH20AIGPUisstillavailableinChina,there’sariskthateventheH20couldberestrictedinthefuture.ShortageofchipsmightbecomeaconstrainttothefuturedevelopmentinChina.TimelineofTimelineofUS’srestrictionon(AI)chipsshippedtoChinaAug:Government-subsidized20192022Source:DeutscheBankResearch,BloombergNews,Reuters,XinhuaNewsDeutscheBankResearchYiXiong/DeutscheBankResearch13(+852)22036139|yi.xiong@/deyun.ou@13DeutscheBankResearchYiXiong/DeyunDeutscheBankResearch14(+852)22036139|yi.xiong@/deyun.ou@14…butrecentdevelopmentscouldhelpmitigatethenegativeimpact??AImodelshavebecomeeasiertotrain:DeepSeekhassuccessfullyutilizedthecomparativelylower-performanceH800GPUformodeldevelopment.Additionally,on11thMar,AntGroupintroducedanovelmethodologyinaworkingpaper,thatenablesAIdevelopmentusingChinese-manufacturedGPUs,achievingresultscomparabletothoseusingmoreadvancedGPUs.Thismethodreportedlyreducestrainingcostsby20%.?Smallerandmoreefficientmodels:Alibaba’sQwenintroduceditslatestpretrainedmodel,QwQ-32B,on6thMar.Withonly32billionparameters,itachievesperformancecomparabletomuchlargermodelslikeDeepSeek-R1(671billionparameters)andGPT-o1(100billionparameters).Withthesedevelopments,deploymentofsmallerAImodelsonlocaledgedevicessuchasphonesandcarstoenablereal-timeanalysiswithoutrelianceoncloudisgrowinginpopularity,whichdoesnotrequirethemostadvancedchips.Inthelongerterm,Chinahopestoaddressthechipsbottleneckbyadvancingitsownadvancedchipsmanufacturingtechnology,rangingfromlithographymachines(EUV)tochipdesigningtools(RISC-V).QwQQwQ32BintelligencecomparedtolargermodelsQwQ32BQwQ32B0Datasource:DeutscheBankResearch,QwenLargeLanguageModelCloudoperatorstobenefitfromsurgingAIdemand??TheriseofDeepSeekhasfueledChina’sAIdevelopments,resultinginasurgeincomputingpowerdemand,especiallyinGPU,benefitingcloudoperators.TheindustryisrapidlyevolvingwithfasteradoptionofAIapplications,boostinginferencingdemandontopofmodeltraining.?Amongourinternetcoverages,hyperscalerslikeAliCloud,TencentCloud,andBaiduCloudshouldbenefitfromthetrendinviewoftheirleadingcloudbusinessesandinfrastructures.?Weanticipatestrongcloudrevenuegrowthandrisingcapexforthesecloudplayersinthecomingyears.Alibaba’sAlibaba’scloudrevenuegrowthisacceleratingRmbbnAlibabacloudrevenue——yoy%(RHS)18%16%14%12%10%8%6%4%2%0%-2%-4%34323028262422202Q233Q234Q231Q242Q243Q244Q241Q252Q253Q254Q25E2Q233Q234Q231Q242Q243Q244Q241Q252Q253Q254Q25EDatasource:DeutscheBankResearchTencent’sTencent’scapexsurgedin2024andwillrisefurtherin2025Datasource:DeutscheBankResearchDeutscheBankDeutscheBankResearch15(+852)22035880|jessie.xu@15CloudoperatorstobenefitAliCloudAlibabaannouncedplanstoinvestatleastUS$53bnoverthenext3yearstoadvanceitscloudcomputingandAIinfrastructure.Thisexceedsthetotalcapexofthecompanyinthepastdecade.OnMarch6th,AlibabaCloudintroducedQwQ-32B,acompactreasoningmodelwithonly32bnparameters.QwQ-32BmatchestheperformanceoftheDeepSeek-R1-671Bmodel,whilebreakingnewgroundbyenablinghigh-performanceinferencemodelstorunlocallyonconsumption-gradegraphicscards,sharplyreducingdeploymentcosts.ThefasterAIadoptionhasbeenboostingAlibaba’scloudrevenuegrowthtoacceleratetodouble-digitsinQ42025,whichshouldfurtheraccelerateinto2025.BaiduCloudCloudBaiduCloudCloudrevenuerepresented16%oftotalgrouprevenuein2024Gen-AIrelatedcloudrevenuenearlytripledyoyin2024,leadingtototalcloudrevenueup+17%yoy/+26%yoyin2024/4Q24,drivenbysurgingGPUdemand.Baiduhasestablishedpartnershipsacrossdiversesectors,includinginternet,automotive,smartdevices,manufacturing,energy,finance,utilities,andalotofAIGCstartups.WeanticipateBaidu’scloudrevenuetomaintainstrongmomentumin2025,supportedbyacceleratingAIadoptionandrisingcomputingdemand.TencentCloudCloudrevenuerepresented6%-7%(DBe)oftotalgrouprevenuein2024TencenthasbeenacceleratingAIintegrationintoitscloudservicesacrossIaaS,PaaSandSaaStocapturethesurgingdemand.AIcloudrevenueapproximatelydoubledyoyin2024despiteconstrainedGPUcapabilities.Tencentcontinuestooptimizetrainingandinferenceefficiencyofself-developedHunyuanLLM;andintegratesDeepSeektoenhanceservices.2025Capexsetatlowteenspercentageofrevenue.WenowestimateRMB87bn,up13%yoy.Weexpectcloudrevenuetoacceleratein2025asithassteppeduppurchasesofGPUssince4Q24.Marketshare-Iaas(1H24)Alibaba,25.8%Others,29.9%Tencent,8.5%HuaweiTencent,8.5%ChinaMobile,ChinaChinaMobile,9.1%13.2%DeutscheBankResearchDatasource:DeutscheBankResearch,CompanyDeutscheBankResearchMarketshare-Paas(1H24)Alibaba,24.9%Others,35.3%Tencent,11.5%ChinaTelecom,Huawei,11.0%7.3%AWSHuawei,11.0%16JessieXu16(+852)22035880|jessie.xu@ChinatelcostobeoneoftheleadingpowersandcorebeneficiariesduringtheAIboom?Soliddigitalfoundation:~30%ofrevenuecontributionand~40%ofcapextilt?Dominantcomputinginfra:>50%ofChinadatacentermarketshare,withmultiplehyperscalenewbuildsformodeltrainingandmassiveedgenodesforinference(potentialassetre-rating)o60EFLOPSintelligentcomputilityhasbeenbuiltbyChina’stelcosasof4Q24,withseveral10k-GPUpools.ZacharyLehman,ina2023paper,GlobalComputingCapacity,estimatedtheworldcomputingpowerat3980EFLOPS,sotheChinesetelcoswouldhavemorethan1%ofglobalcomputingpowero400Gbpsfull-fibrebackbonenetworkacrossnationalhubs,facilitatingdatacenterinterconnection?Comprehensivecloudofferings:full-stackcapabilitiesinIaaS/MaaS/DaaS/PaaS(monetizationdeveloping)oSuccessfullyhelpedSOEslikeSinopecdeployingDeepSeekrecently?AIApplicationslikelydrivedataconsumptionandrevenueoHuawei(MWCBarcelona)projectsChina’stoken-drivendailydatatraffictoreach5mTBby2030,a3.5xgrowth200180160140120200180160140120100806040200202220232024E——CM:digitaltransformation——CT:industrialdigitalization35%CU:industryinternet30%25%20%15%10%5%0%2021202220231H24Datasource:DeutscheBankResearch,CompanydataDeutscheBankResearchDeutscheBankResearch17(+852)22036190|liken@17Inferencedemandpoisedtounlockanewcycleofindustrygrowth?ChinaIDCshadaperiodofdarktimeduring2021-’23,withovercapacityandpricingpressure.Theturnaroundoccurredin2024,whenbothrevenueandcapacitygrowthstartedtoreaccelerate.Inventorybecameanadvantage.Notonlythesecuredlandintoptiercities,alsotheenergypermits,operationalandmaintenanceexpertise.CSPscapexplansprovidehigh-levelordervisibilityinthenearterm;widerapplicationsadoptionoffersignificantlong-termupside.Valuationstillattractive:~20xEV/EBITDAvs.globalpeersat~30x?ChinaTowerhas1mshelters,whichcanbeusedtohostservers.Usecasesincludelocalserversforhostingmedium-sized-languagemodels,forgamingandotherlowlatencyuses.Theprojectsarealreadytrialinginsixprovinces.Itwouldhavelowincrementalcosts,sowouldbecomeasignificantprofitdriverifusagecomesthrough.capacityTotalYoY100%ServicerevYoYcapacityTotalYoY100%90%80%70%60%50%40%30%20%10%0%DeutscheBankResearchDatasource:DeutscheBankResearch,CompanyDeutscheBankResearch18PeterMilliken18(+852)22036190|liken@AI-accelerateddeploymentofSmartGrids??AI-backedtoolsareacceleratingthedeploymentofSmartGridsinpowerindustry,whichincludes1)abetterintegrationandplanningforrenewablepowergeneration;2)improvingpowertransmissionefficiency;3)optimizingpowerdistributionnetworksbasedonend-userdemand.?AIsupportspowergridmanagement.InDecember2024,theStateGridlaunchedtheBrightPowerModel,abig-datamodelforpowerindustry,whichcanintegrate100billionsofdataparameters,andactasanexpertinover600applicationscenariosofpowergridoperation.InMarch2025,italsolaunchedanewapplicationintegratingtheBrightPowerModelandDeepSeek,accordingtolocalnews(Sohu).?AImodelingimprovesrenewablepowercapacityplanningbyleveragingadvancedweatherforecastingtopredictsolar/windpowergenerationoutputs.AccordingtoBJXNews,HuadianGroup,incollaborationwithHuaweiandothercompanies,testedanAI-basedpredictionmodelatarenewablepowerstation,whichcouldachieve97.24%accuracyfor15-minuteoutputforecasts,and91.72%for4-hourforecasts.SmartSmartgridframeworkDatasource:DeutscheBankResearch,StateGrid,YicaiGlobal,BJXnewsDeutscheBankResearchDeutscheBankResearch19(+852)22035889|gary.zhou@19AdoptingAIinIntegratedEnergyprojects??Chinacity-gascompanies(suchasENNEnergy[2688.HK])havebeeninvestinginintegratedenergy(IE)projects,whichextendstheirtraditionalgasdistributionbusinessfurthertoheat/steamgeneration,powersupplyandothertypesofenergydependingoncustomerneeds.?AdoptingAIinintegratedenergyprojectshelpscreatea"Load-source-grid-storage"micro-gridmodelthatoptimizesforbotheconomicbenefitsandreducedc