Preface

TherapiddevelopmentofartificialintelligenceindustrydrivestheexplosivegrowthofvariousAIapplications.Asthecriticalinfrastructurebridgingendusersandcomputingresources,metropolitanareanetworks(MANs)arenowfacingtransformativerequirementsinnetworkarchitecture,functionalcapabilities,andserviceparadigms.

In2024,ChinaTelecompioneeredtheindustry-first‘computingservice-orientedmetropolitanareanetwork’conceptandreleasedthe‘computingservice-orientedmetropolitanareanetworkWhitepaper’,generatingsignificantindustry-wideattentionanddiscourse.Asacontinuation,thiswhitepaperprovidesin-depthanalysisofmetropolitanareanetworkevolutionintheAIera.Thiswhitepaperfirstanalyzesthedevelopmentlandscapeofartificialintelligencefromtheperspectivesofindustryadvancementandmacropolicies.Subsequently,itconductsanin-depthanalysisofAIapplicationrequirementstodefinetheessentialnetworkcapabilitiesthatmetropolitanareanetworksmustpossess.Thiswhitepaperthenexaminesthedesignobjectives,elaboratingontheoverallarchitectureandkeytechnologiesofmetropolitanareanetworksfortheAIera.Finally,itprovidestechnicalsolutionstailoredfortypicalscenarios.

Thefollowingorganizationsandprincipalmemberscontributedtothepreparationofthiswhitepaper:

lChinaTelecomResearchInstitute:YongqingZhu,ZehuaHu,XiaGong,ShizhangYuan

lZhongguancunUltraCrossConnectionNewInfrastructureIndustryInnovationAlliance:BoYuan

lHuaweiTechnologiesCo.Ltd.:HaobinZhao,JieDong,LiZhang

lZTECorporation:WenqiangTao,HaidongZhu,XiaoweiJi

Directory

CHAPTERIDevelopmentTrendsofArtificialIntelligence 1

1.1AIIndustryentersaphaseofacceleratedgrowth 2

1.2AIisfocalpointofglobalindustrialpolicies 4

1.3AItechnologyisdevelopingexplosively 5

1.4ChallengestoMANfromlarge-scaleAIcommercialization 9

CHAPTERIIAI-DrivenRequirementsforMAN 12

2.1AIapplicationsannovationaontinuestoaccelerate 13

2.2AIapplicationsexhibitdiversedeploymentmodels 15

2.3AIapplicationsimposenewrequirementsonMANs 17

2.4AIapplicationsdrivenMANstowardnext-generationevolution 24

CHAPTERIIIMANArchitecturefortheAIEra 25

3.1MANsdesignobjectives 26

3.2OverallMANarchitecture 28

3.3KeymodulesofMAN 30

CHAPTERIVMANKeyTechnologiesfortheAIEra 35

4.1Integratedcomputingandnetwork,convergedbearernetwork 36

4.2Elasticity,agility,flexibilityandefficiency 37

4.3Precisecontrolanddynamicconvergence 41

4.4IntelligentO&M,securityandreliability 45

CHAPTERVTypicalDeploymentScenarios 50

5.1Scenario1:TransmittingmassivesampledatatoAIDC 51

5.2Scenario2:Modeltrainingwithstorageandcomputedisaggregated 52

5.3Scenario3:CollaborativemodeltrainingacrossmultipleAIDCs 53

5.4Scenario4:Cloud-Edgecollaborativemodeltraining/inference 54

5.5Scenario5:Inferencedelivery 54

5.6Scenario6:Federatedlearning 55

5.7Scenario7:Multi-agentsystem/A2A 56

CHAPTERVIConclusionsandFuturePerspectives 58

1

ChapterI

DevelopmentTrendsofArtificialIntelligence

2

1.1AIIndustryentersaphaseofacceleratedgrowth

AsthecoredrivingforceleadingtheFourthIndustrialRevolution,theartificialintelligence(AI)industryisexperiencingunprecedentedrapiddevelopment,demonstratingenormousmarketpotential.AccordingtoGrandViewResearch,theglobalAImarketsizereached196.63billionin2023andisprojectedtoincreaseto1,811.75billionby2030,withacompoundannualgrowthrate(CAGR)of37.3%from2024to2030.InChina,researchreportsindicatethatthescaleoftheAIindustryisexpectedtoexpandfrom398.5billionyuanin2025to1,729.5billionyuanin2035,withanestimatedCAGRof15.6%.Artificialintelligencehasundoubtedlybecomeapowerfulengineforglobaleconomicgrowth.

Figure1-1:Globalartificialintelligentmarket

TheglobalAIindustrydemonstratesatrendfordevelopmentof‘dual-trackadvancementanddiversifiedflourishing’.Ontheonehand,globaltechnologygiantscontinuetointensifytheirAIinvestments:companieslikeGoogleandMicrosoftaredeepeningresearchanddevelopment(R&D)incoreAItechnologies;AmazonandApplepersistininnovatingintelligentcloudservicesandend-devicesmartapplications,whileChina'smajortechfirmssuchasBaidu,Alibaba,Tencent,andHuawei(BATH)arealsomakingrapidprogressinkeyareassuchasAIchipdevelopment,largeAImodelconstruction,computervision,andembodied

3

intelligence.Ontheotherhand,theexplosivebreakthroughsingenerativeAItechnologyhavespurredawaveofinnovativeenterprisesworldwide:OpenAIpioneeredthecommercializationofgenerativeAIwithChatGPT;AnthropicandCoherespecializeinvertical-orienteddevelopment;andin2025,China'sDeepSeeksignificantlyacceleratedthecommercialapplicationoflargeAImodelsininferencescenarios.NumerousemergingAIsupplychaincompanieshavebecomeinvestmenthotspots,collaboratingwithindustryleaderstoformasynergisticinnovationecosystem.Thisdynamicdevelopmentpatternthatfeaturescompetitionandsymbiosisamongdiverseplayersnotonlyacceleratesthecommercialdeploymentoflargelanguagemodelsinfinance,healthcare,andmanufacturing,butalsoprovidesrobustmomentumforthehigh-qualitydevelopmentofthedigitaleconomy.

BenefitingfromtherapiddevelopmentofAIindustry,AItechnologiesarebecomingpowerfulenginesforurbandevelopment,injectingunprecedentedvitalityintovarioussectorsofcities:Intransportationfield,leveragingtheprecisepredictivecapabilitiesoflargeAImodelsoptimizestrafficflowandenhancestravelefficiency.Inhealthcareindustry,AI-assisteddiagnostictechnologiesenabletherapidandaccurateanalysisofmedicalimages,helpingdoctorstoformulatetreatmentplans.Ineducation,customizedteachingcontentisprovidedbasedonstudents'learningprogressandcharacteristics,stimulatingtheirinterestandpotential.ThefinancialsectorutilizeslargeAImodelsforriskassessmentandinvestmentdecision-making,improvingtheprecisionandsecurityoffinancialservices.Furthermore,numerousfieldssuchasintelligentmanufacturing,intelligentgovernmentservices,andenvironmentalmonitoringhavebecomemoreefficient,intelligent,andsustainablethroughtheempowermentofAI.TheapplicationofAItechnologiesprovidesresidentswithmoreconvenient,comfortableandsecurelivingexperiences,leadingcitiestoanintelligentanddigitalfuture.

4

1.2AIisfocalpointofglobalindustrialpolicies

AIhasbecomeoneofthecoredrivingforcesforurbanandsocialdevelopment,formingaglobalconsensus：

lTheUnitedStateslaunchedthe‘WhiteHouseSmartCitiesInitiative’in2015,leveragingAI,bigdata,andtheInternetofThings(IoT)technologiestohelpcitiesaddresschallengessuchastrafficcongestion,energymanagement,andpublicsafety.By2025,itwouldfurtherstrengthenAIinfrastructurethroughthe‘StargateProgram’.

lTheEuropeanUnionproposedthe‘EuropeanDataUnionStrategy’in2025topromoteAIandbigdataapplicationsinhealthcare,education,andurbangovernance,supportedbythe‘DigitalEuropeProgramme’toimplementAIincriticalsocialandlivelihoodsectors.

lJapanintroducedthe‘SuperCity’vision,integratingAIandIoTtocreatedata-driven‘smartcities’.

lSingaporeimplementedits'NationalAIStrategy2.0',whichcombinestalentattraction,industrialapplications,R&Dinnovation,andinfrastructuretobuildanAIecosystemthatimprovespublicservicesandindustrialcompetitiveness.

lTheChinesegovernmentalsoprioritizesAI-drivenurbandevelopment.In2024,China’sNationalDataAdministrationissuedguidelinestodeepensmartcityinitiatives,encouragingAI-poweredsolutions,suchasintelligentanalysis,scheduling,regulationanddecisionmaking,tocomprehensivelyempowerurbandigitaltransformation.

NetworkshavebecomecriticalinfrastructuresupportingglobalAIindustrydevelopmentandarereceivinghighpriorityfromnationsworldwide：

lInChina,‘empoweringcomputingthroughnetworks’hasbeenestablishedasafundamentalprincipleforbuildingsmartcities.InOctober2023,China'sMinistryofIndustryandInformationTechnology(MIIT)introducedtheHigh-QualityDevelopmentActionPlanforComputingInfrastructure,whichaimstocreateagroupofcomputingpowernetworkcitybenchmarksinkeyregions.

5

lInJune2023,theSingaporegovernmentlauncheditsDigitalConnectivityBlueprint,proposingtheconstructionofseamlessend-to-end10GbpsdomesticconnectivitywithinfiveyearstoensureSingapore'sdigitalinfrastructureremainsworld-classandsetsthedirectionforitsdigitalfuture.

lInApril2024,SaudiArabia'sMinistryofCommunicationsreleasedtheSaudiArabia's10GbpsSocietyWhitePaper,becomingthefirstgloballytoproposeanend-to-endhigh-speed,high-qualityNet5.5Gnetworkarchitecturetosupportthecountry'sintelligenttransformation.

lIn2025,theEuropeanCommissionDigitalEuropeProgramme(DIGITAL)2025-2027alsoemphasizedtheneedtoenhancenetworkresilienceinvariousAIscenarios.

WiththewidespreadadoptionoflargeAImodelsandgrowingdemandforapplicationssuchasdistributedinference,theroleofnetworksinAIdevelopmentisbecomingincreasinglyprominent.Buildingasecond"informationsuperhighway"dedicatedtoAIhasemergedasaglobalpriority.

1.3AItechnologyisdevelopingexplosively

1.3.1AItechnologyisadvancingcomprehensively

ThedevelopmentofAItechnologydemonstratesnotabletrendsofdiversifiedcollaboration,high-efficiencyevolution,andmulti-ecosystemintegration：

lAtthehardwarelevel,thesignificantincreaseininferencescenarioshasdrivenrapidadvancementsindedicatedAIchipssuchasTPUsandLPUs,whilegeneral-purposeGPUs,combinedwithcutting-edgetechnologieslikechiplet,3Dstacking,andquantumcomputing,provideenhancedcapabilitiesforultra-large-scaleAImodeltraining.

lInstoragetechnology,protocolssuchasHBM3andCXLhaveachievedleapsinmemorybandwidthandcapacity,whilearchitecturessuchasstorage-computedisaggregationmeetthedemandforbuildingprivateknowledgebasesbasedonlargeAImodels.

6

lHigh-speedinterconnecttechnologiessuchasUEC,NVLink,UCIeandFalconbreakdowndatatransmissionbarriers,enablingefficientcollaborationbetweendistributedcomputingandheterogeneousarchitectures.

lOnthesoftwareecosystem,open-sourceframeworkssuchasPyTorchandTensorFlowaredeeplyintegratingwithautomatedtoolchains,combinedwithcloud-edge-deviceunifieddeployment,toachieveend-to-endoptimizationfrom

trainingtoinference.

lInaddition,greencomputingtechnologies,includingliquidcoolinganddynamicpowermanagement,contributetothesustainabledevelopmentofAI.

1.3.2LargeAImodeltechnologyentersrapiddevelopment

phase

LargeAImodelshavebecomeoneofthemostwidelyappliedkeyAItechnologiestoday.FromthelaunchofChatGPTin2022totheriseofDeepSeekin2025,thefieldoflargeAImodelshasexperiencedexplosivegrowth.Thedevelopmentoflargemodelsexhibitsmulti-dimensionaltrends:ononehand,modelscalecontinuestoexpandwithincreasingparametercounts,enablingthecaptureofmorecomplexpatternsandrelationshipstoenhanceperformanceacrossvarioustasks;ontheotherhand,multi-modalfusionhasbecomeanimportantdevelopmentdirection,aslargemodelscombinetext,image,speechandothermulti-modaldatatoachievemorecomprehensiveunderstandingandgenerationofinformation,expandingtheirapplicationscenarios.Additionally,greaterattentionisbeingpaidtomodelsafety,reliabilityandinterpretability,withresearcherscommittedtodevelopingmorerobustmodelarchitecturesandtrainingmethodstoensurestableoperationandtrustworthyapplicationoflargeAImodelsincomplexenvironments.ThesetrendscollectivelydrivethecontinuousadvancementoflargeAImodeltechnology,layingasolidfoundationforthewidespreadapplicationofAI.Currently,largeAImodelsareevolvinginthefollowingtechnicaldirections:

Direction1:AstheparametersandtrainingdatascaleoflargeAImodelscontinue

7

toincrease,thedemandforcomputingpowerisalsogrowingrapidly.Single1K+GPUsor10K+GPUsAIDataCenters(AIDC)canhardlymeettherequirementsofultra-large-scaletraining.TakingLlama3.1releasedin2024asanexample,itslargestmodelhas405Bparametersandrequiresapproximately15trillionTokensforpre-training,withtheentiretrainingprocessdemanding39.3millionGPU/hours(H100)ofcomputingpower.Therefore,adoptingdistributedtrainingmethodsandutilizinghigh-performancenetworkstoenhancethecollaborativetrainingefficiencyacrossmultipleAIDCshasbecomeanecessityforAIdevelopment.Currently,multipleoperatorshavecompletedthecommercialdeploymentofdistributedtraining,achievingthedistributedtrainingfor10K+GPUs,100BparameterslargeAImodelsacrossAIDCsoverdistances100+kilometers.Amongthem,ChinaTelecomandHuaweijointlydeployedthedistributedtrainingservicesupporting120KMwide-areaRDMAlosslesstransmission,withtrainingefficiencyreachingover95%.

Direction2:SoftwareengineeringoptimizationhasbecomethekeypathwaytobreakthroughAIhardwarebottlenecks,drivinglargeAImodelstowardcost-effectivedevelopment,andacceleratingtheadoptionofAIacrossindustries.Theopen-sourceDeepSeek-V3in2025completedpre-traininginjusttwomonthsusingonly2,048GPUsthroughalgorithmicoptimization,whiletheDeepSeek-R1modelfurthercompressedthetrainingcycleto2-3weeks.This‘low-cost&open-source’solutionsignificantlyloweredthetechnicalthresholdforlargeAImodels,directlyleadingtotwonotablechanges:First,therelativelylowusagecoststriggeredexplosivegrowthinlargeAImodel-basedapplications,resultinginsurgingAItrafficwithincitiesthatrequiresnetworktoensureefficientAItrafficsteer;Second,throughfull-stacksoftwareengineeringoptimizationspanning‘a(chǎn)lgorithm&hardware&system’,AIinferencelatencywasreducedbyover60%,drivingexponentialgrowthinAIinferencedemand.

Direction3:Theintelligentinteractionofmulti-edgeagentsreflectsAItechnology'stransformationfromcentralizedtodistributedsystemsandfrom

8

singleintelligencetocollectiveintelligence,drivingbreakthroughsinreal-timeperformance,autonomy,andcollaborationofAI.LargeAImodelscanachievelightweightdeploymentthroughtechniqueslikemodeldistillation,makingthemcompatiblewithresource-constrainedscenariossuchasconsumer-gradeGPUs,mobiledevices,andIoTequipment,therebypromotingthedevelopmentofedge-basedsmallintelligentdevices.Atthesoftwarelevel,thewidespreadadoptionofMulti-Agenttechnologyenablesmultipleterminalstocollaborativelycompletecomplextasks,furtheradvancinglarge-scaleinteractiveapplicationsofedgeagents.Google'sintroductionoftheA2AandMCPprotocolsforagentinteractionin2025signalsAI'simpendingtransitionfromthe‘cloudcomputing’architectureofB2B,B2C,andC2Ctothe‘granularcomputing’architectureofA2A,M2M,andX2X,withthefrequentinteractionsbetweenintelligentcomputingparticlesplacehigherdemandsonthereliability,andbandwidthcapacityofnetwork.

Direction4:InSeptember2024,OpenAIlaunchedtheo1modelwithChain-of-Thought(CoT)mechanism,whichachieveshigheraccuracybyextendingthinkingtimeduringinference,markingaparadigmshiftfrompursuingresponsespeedtoemphasizingdeepreasoning.Thistransformationhasdriventheshiftofcomputingpowerdemandfrompre-trainingtoinference,breakingthroughthelimitationsofScalingLaw:whilepre-trainingrelieson10K+GPUsScaling-upclusters,inferencecanbeimplementedthroughScaling-outarchitecturescomposedofasmallnumberofGPUs,promotingtheevolutionofAIinfrastructuretowarddistributedandflexiblyscheduledsystems.Additionally,thesignificantlyincreaseddeploymentdemandsonAIinferencehaveraisedrequirementsforlarge-scaleinferenceperformanceimprovements.Network-baseddistributedinferencehasbecomeakeydirectionforfutureurbanAIapplications,necessitatingnetworkstosupportdistributedAIinferencedeployment.Inresponse,NVIDIAintroducedtheDynamoframework,adoptingaPD-separatedarchitecturetooptimizeresourceschedulingandcomputingefficiencyinlarge-scaleAIinference.

9

1.4ChallengestoMANfromlarge-scaleAIcommercialization

BuildingacomprehensiveAIurbanecosystemhasbecomethecorepathwayforupgradingurbansystemstoadvancedintelligence.Inthisprocess,theconceptof‘CityasaComputer’hasgraduallygainedglobalconsensus:bydeeplyintegratingcomputingpower,storage,andterminalsthroughmetropolitanareanetworks(MANs),citiesaretransformedintodistributedultra-large-scalecomputingsystems,enablingcitywideintelligentmanagementthroughmillisecond-leveldataflowandreal-timedecision-making.Existingbroadbandnetworks,mobilenetworks,dedicatedgovernmentandenterprisenetworks,andcloudnetworkswithincitiesconnectedvarioususers.However,traditionalMANsstruggletomeettherequirementsforcarryingurbanAIservices,whetherintermsofnetworkarchitectureorcoretechnologies.

1.4.1Challengesindatacirculation

ThetrainingoflargeAImodelsandtheconstructionofknowledgebasestypicallyrequiredatavolumesattheTB/PBscale,whichimposeshigherthroughputrequirementsondatatransmissionnetworks.Simultaneously,thecomputingtrafficoflargemodelsexhibitssignificantelasticcharacteristics,demandingextremelyhighnetworkreliability.Substandardandnon-deterministicnetworksmayresultininsufficientdatatransmissionbandwidth,excessivelatency,orfrequentpacketloss,therebycompromisingtheavailabilityofcomputingresources.Furthermore,versioniterationsoflargemodelsandknowledgebaseupgradesinAIsystemsalsodependonstablenetworksupport.Poornetworkqualitycanconstraintheimplementationofthesefunctions,ultimatelyreducingtheoveralloperationalefficiencyofAIinfrastructure.

Therapiddevelopmentoflarge-scaleinferenceapplicationsandA2AcomputingparadigmshasintroducednewchallengestourbanAIdatacirculation:ononehand,

10

MANsneedtomeettheefficientdatacommunicationandinteractionrequirementsbetweendistributedinferencenodes;ontheotherhand,theA2Amodehasledtoexponentialgrowthinhigh-frequencyinteractiontrafficacrossintelligentagents,whichnotonlysignificantlyincreasesthebandwidthrequirementsofedgenetworksbutalsorequiresMANstoensurethereliabilityofinformationinteractionbetweenintelligentagents.Therefore,torealizethevisionof‘CityasaComputer’,itisurgenttobuildanewultra-interconnectednetworkdifferentfromtraditionalMANstomeetthetransmissionrequirementsofAIcomputingdataflowsandenableMANstoeffectivelysupportefficientcomputationaldatacirculation.

1.4.2ChallengesinO&M

WhenMANscarryAIservices,networkmanagementandmaintenance(O&M)facegreaterchallenges.Fromservicemodelperspective,AIhastransformednetworktrafficpatterns:largeAImodeltrainingcancausesuddentrafficsurges,whilefrequentinteractionsbetweenintelligentagentsalsogenerateburstycommunication,requiringnetworkstopossesspredictiveplanningandmaintenancecapabilities.AIservicesalsodemandhighernetworkreliability,evenminorfaultsduringmodeltrainingmayleadtocompletetaskresets.Whenlarge-scaleinferenceservicesreplacemanualservicesincities,networksmustensureserviceexperience.

Consequently,traditionalmanagementmodelsthatrelyonmanualinterventionandrouteconvergencetoensurebasicnetworkavailabilitycannolongermeettheperformancedemandsofAIservices.AIservicesrequirehigherfaultself-healingratesandlowerlatencyinnetworkoptimizationdecisions,pushingnetworkoperationstowardhighautonomytofulfillneedslikepredictivemaintenance,serviceawareness,andelasticoptimization.Thequestionofhowtoequipnetworkswithhighlyintelligentmanagementandoperationalcapabilities,namelyautomatingtheadjustmentofnetworkresourcesandconfigurationsbasedontheintentionsandstatesofcomputingservices,hasbecomeakeyfocusforAI-orientedMANs.

11

1.4.3Challengesinsecurityandtrustworthiness

Withtherapidadoptionoflargemodels,vastamountsofurbandataarebeingutilizedforanalysis,computation,andprocessing.Datafromenterprises,households,andindividualsconstituteprivatedomaintraffic,posingsignificantsecurityrisks:Forhouseholdsandindividuals,privatedomaintrafficinvolvessensitivedatasuchaspersonalinformationandconsumptionbehaviors,whereleakscouldleadtoprivacyviolations;forenterprises,privatedomaintrafficencompassesR&Ddata,productiondata,andoperationaldata,wherebreachescouldunderminecompetitivenessoreventriggerlegaldisputes.Sincedatatransmissionfacespotentialthreatssuchastheft,tampering,andloss,MANsmusthaverobustdataprotectioncapabilitiestoensuredataconfidentiality,integrity,andavailability.

TraditionalAAA(Authentication,AuthorizationandAccounting)systemsanddataencryptiontechnologiesbasedontrafficflowsstruggletomeetthesecurityandtrustrequirementsofAIscenarios.However,emergingtechnologieslikeblockchainandquantumencryptionofferinnovativesolutionsfortrustworthydatacirculation:blockchainprovidesimmutable,end-to-endtraceabletrustmechanismsforAIdataflowsthroughdistributedledgersandsmartcontracts;quantumencryptionleveragesbreakthroughslikequantumkeydistributiontofundamentallyenhanceanti-eavesdroppingcapabilitiesfordatatransmission.MANsmustintegratetheseinnovativemechanismstoestablishatrustedfoundationforlarge-scaleurbanAIdeployment,providingcriticalinfrastructuresupportforthewidespreadimplementationofmetropolitanAIservices.

12

ChapterII

AI-DrivenRequirements

forMAN

13

2.1AIapplicationsannovationaontinuestoaccelerate

Inearly2025,DeepSeekspearheadedatransformativewaveingenerativeAI,drivenbyitsexceptionalperformanceandindustry-leadingcostefficiencyinLLMtrainingandinference,acceleratingthecommercializationofAItechnologies.Today,AIapplicationshaveenteredthestageofscaleddeployment,servingdiversescenariosacrosshome(toH),consumer(toC),andbusiness(toB),withpenetrationintomultipleverticalindustriesincludingmedia,legalservices,education,andmanufacturing.

2.1.1AItoHscenarios

AIsignificantlyenhancestheprofessionalism,interactivity,andpersonalizationofhomeservices,enrichinghomescenarios.Currently,theindustryisgraduallyreachingaconsensusonbuildinganintegratedsmarthomeecosystemthatcombinesconnectivity,computingpower,andintelligence.Throughcloud-network-edge-devicecollaboration,providingbroadbanduserswithansmartcloudservices,supportingvariousAItoHscenariosincludingsmarthomeandhomeassistants:

lSmarthome:SmartTV,smartrefrigeratorsandothersmarthomeproducts,

utilizingAItechnologieslikevoicerecognitionandcomputervision,nowsupport

intelligentcapabilitiesincludingnaturallanguageinteraction,userhabitlearning

andcontextualadaptation.Thesesmarthomeproductscandynamicallyadjust

lighting,temperatureandhumiditybasedonuserpreferences,whileemploying

facialrecognitionandbehavioranalysistechnologiestoenhancehomesecurity.

lHomeassistants:Smarthomeassistantproducts,includingsmartspeakersanddomesticrobots,employnaturallanguageprocessingandotherAItechnologiestoenable

harmonious

human-machinedialogue.Theseproductsachievepreciseintentunderstandingtoexecutetasksincludingscheduleremindersandinformationretrieval,whileenablingcontextualizedservicessuchasappliancecontrolandsecuritymonitoringthroughseamlessIoTinteroperability.

14

2.1.2AItoCscenarios

AIrevolutionizestheinteractionsbetweenconsumerandservice,drivingenhanceduserexperiencesandfosteringmarketinnovation.TheAIinnovationlandscapeiswitnessingrapidproliferationofvariousverticalapplications.MajorindustryplayersareactivelydeployingAItoCsolutionsacrosssmartterminals,personalizedservices,anddigitallifestyledomains,leveragingmetropolitanAIservicestoenhanceuserexperienceandretention.ThecurrentAItoCapplicationsprimarilyencompassthefollowingcategories:

lProductivityEnhancement:AIapplicationssuchasintelligentsearch,automatedsummarization,contentgeneration,andcodeassistancehavesignificantlyimprovedefficiencyforbothindividualsandorganizations.Theseapplicationsstreamlinecomplexworkflows,enablinguserstofocusonhigher-valuestrategicinitiativeswhilefosteringinnovationandcompetitiveadvantage.

lCreativeGeneration:AIapplicationsincludingdesignautomati