2025年技術(shù)趨勢報告：機(jī)遇大于風(fēng)險 -TECH TRENDS 2025 -Big risks Bigger possibilities

25個INTRODUCTIONINFO-TECHRESEARCHGROUPOFPROBABLEFUTURESThesummerblockbusterandOscar-winningfilmOppen-heimerdetailstheclassifiedManhattanProjectandeffortsTledbyJ.RobertOppenheimertodevelopthefirstatomicbombinaracetowinWorldWarII.Whatitdoesn'tdetailaretheeffortsofscientistsJohnvonNeumannandStanislawUlamastheycreatedanewstatisticalmethodnecessaryforthesafetyofthescientistsduringthetestsconductedinthedesert–amemorablesceneinthemovie.Thescientistsneededtoidentifywhatmate-rialswouldprotectthemfromtheradiationoftheblastwhiletheyobserveditfromadistance.Thisrequiredanestimationofhowneutronspropelledbytheblastwouldshootthroughdifferentmaterials.Calculatingthiswithexistingdeterministicmodelswouldn'twork–therewerejusttoomanyneutrons.So,thescientistscreatedabrilliantnewapproachthatcalculatedtheprobabilitybasedonarandomsubsetofneutronsinstead.Aswitheverythingdevelopedduringtheclandestineproject,thenewapproachwasveiledinsecrecywithacodename–theMonteCarlomethod.ThescientistssawsimilaritiesbetweentheprobabilisticmethodandthegamesintheMonteCarlocasinoinMonaco.Notonlywasitaneffectivewaytosimulatethemovementofneutrons(noscientistswereharmedduringthebombtests),butit'sprovenausefulstatisticalsimu-lationapproachinmanyindustries,fromfinancetomedicalusesandsupplychainoperations(VirginiaTech).DevelopingthesimulationcapabilityprovidedtheManhattanProjectonemoreslightedgeoveritscompetitors.Thatadvantagewasmaintainedbytheapproachtoknowledgeassurancetakenduringthehighlysecretmilitaryproject.Oppenheimerdetailsthemeasurestheprojectwenttoinordertomaintainsecrecy–buildingatemporarytowninthedesertsoscientistsandtheirfamiliescouldbecontainedwithoutcontactwiththeoutside.Allcommunicationinandoutofthetownwasmonitoredandcontrolled,andevenwithintheprojectitselfinformationwascompartmentalizedonaneed-to-knowbasis.TherearetwolessonstodrawfromthisforITleaderslookingaheadto2025.Historically,chiefinformationofficershavebeenaccount-ablefortherecordkeepingattheirorgani-zations.Likearesidenthistorian,CIOsmaintainedtheintegrityofanorganization'spast,makingitverifiableandauditable.Withdigitaltransformation,CIOswereaskedtodomoretoreportonthecurrentstateoftheorganization–inasreal-timeaspossible.Thebusinessintelligenceandanalyticsrequiredtodrivedecision-makingcouldn'tbebasedonoldinformation.Now,asfirmspushtheirinvestmentintoartificialintelli-gence(AI)andmorespecificallygenerativeAI,thefocusshiftstosimulatingthefuture.Inafast-changinganduncertainworld,AIpredictsdifferentscenariosbasedontheprobabilityoftheoutcome.GenerativeAIprovidesanoutputthataimstosimulateahumanresponsepromptedbyaninput.EvolutionofITfromrecordkeepingtofore-castingprobablefuturesIn2025,CIOsneedtograduatefrombein-grecordkeeperstobeingforecastersofprob-ablefutures.InTechTrends2025,Info-Techwillexplorethreetrendsalongthethemeof"simulatedfutures":>QuantumAdvantageTheimperativeforITtoprotectknowledgecreationandcurationAtthesametime,CIOsmustpushtodevelopforecastsandadvancetheknowledgeoftheirorganizationsinawaythatcarefullymanagestheintegrityoftheknowledgecreationandcurationprocess.LiketheManhattanProject,CIOsneedtocreateatrustedenvironmentthatensuresknowledgeassurance.Hereweexplorethreeothertrends:>Post-QuantumCryptographyAltogether,oursixtrendsreflecttheoppor-tunitiesaheadfororganizationstoseizeuponwithemergingtechnologycapabilitiesandtheriskstheymustmitigatealongtheway.JustlikeaplayerputtingabetdownattheMonteCarlo'sroulettetableknowstheymaywinorlosebasedontheprobabilities,thesethreethemesrepresentbothrisksandrewards:>Pre-QuantumFoundations-BrianJackson,ReportAuthorandPrincipalResearchDirectoratInfo-TechResearchGroup.TECHTRENDS2025//0425個INTRODUCTIONINFO-TECHRESEARCHGROUPPleaseestimatethetotalheadcountofyourentireorganization.(n=820)percentcount0-25026.46%217251-1,00024.02%1971,001-2,50013.17%1082,501-5,00015.00%123Info-Tech'sTechTrends2025reportisbasedontheresultsofitsPleaseestimatethetotalheadcountofyourentireorganization.(n=820)percentcount0-25026.46%217251-1,00024.02%1971,001-2,50013.17%1082,501-5,00015.00%123ExpertinterviewswereconductedbetweenMarchandJuly2024andprovideadditionalcontexttothetrendsaswellasspecificcasestudyexamplesofhoworganizationsarerespondingtothetrends.Viewtheexpertcontributorssectiontoseeacompletelistofexternalcontributors.Intotal,tenexpertcontributorsarelisted.Morethan5,00021.34%175Inaddition,theFutureofITsurveyandTechTrends2025reportweredevelopedthroughdiscussionswithmanyInfo-Techresearchadvisors,practiceleads,executives,workshopfacilitators,andMorethan5,00021.34%175Whattitlebestdescribesyourposition?(n=820)FurtherfirmographiccontextontheFutureWhattitlebestdescribesyourposition?(n=820)percentcountOwner/President/CEO6.22%CIOorotherC-levelofficer20.49%168VP-level8.66%Director-level22.80%187Manager21.34%175Seniorindividualcontributor11.83%97Entry-levelindividualcontributorContractor/consultant5.37%44Inwhichcountryorregionisyourorganization'sprimaryheadquarters?(n=820)countcountpercentUnitedStates44.51%365Canada14.76%121Australia6.46%53Africa6.10%50CentralAmerica0.12%1SouthAmerica2.32%195.00%Other(Pleasespecify)10.12%83Throughoutthereport,analysiswilluseITmaturityasanindependentvariabletocomparethemostmaturegrouptotheaveragegroup.Thiscomparisonwillbedonebycomparingrespondentsthatchoose"ITtransformsthebusiness"(Transformers)toacombinationofrespondentsthatselect"ITsupportsthebusiness"and"IToptimizesthebusiness"(Average).WhatbestdescribesyourcurrentlevelofITmaturity?(n=702)percentcountITtransformsthebusiness16.10%113ITexpandsthebusiness12.25%86IToptimizesthebusiness29.06%204ITsupportsthebusiness34.47%242ITstrugglestosupportthebusiness8.12%57Whatisyourorganization'sprimaryindustry?(n=820)percentcountArts&Entertainment(includingsports)0.61%5ConstructionEducation9.39%FinancialServices(includingbanking&insurance)10.49%86Gaming&HospitalityGovernment/PublicSector19.39%159Healthcare&LifeSciences8.90%Manufacturing7.44%61NotforProfit(includingprofessionalassociations)4.76%39Media,Information,Telecom10.61%87ProfessionalServices8.17%6726Transportation&WarehousingUtilities2.20%18RealEstateandPropertyManagementNaturalResources16Other(Pleasespecify)6.83%56TECHTRENDS2025//0625個INTRODUCTIONINFO-TECHRESEARCHGROUPOpportunitiesRiskMitigation 020101030304050506TECHTRENDS2025//0825個EXPONENTIALAIINFO-TECHRESEARCHGROUP/imagineanAI-powered,top-down3Dsimulationtourofanoptimizeddepartmentstorewithawireframeoverlay,dramaticlighting--styleraw--v6.0--ar17:11AI'saccelerationintomoreaspectsoftheorganization'soperations.SincegenerativeAIenteredintothebusinesslexiconin2022,thetechnologyhasworkeditswayintovariousenterprise-scalesolutionsfrommostofthemarket-leadingvendors.FromsolutionsthathelporganizationsbuildwithgenerativeAI,marryingtheirowndataandprocesswithalargelanguagemodel,tofeaturesembeddedinlargerproductivitysuites,AIseemstohavespreaditstentaclesintomanydifferentbusinessfunctionsinashortamountoftime.Lastyear,wesawartificialintelligence(AI)ormachinelearning(ML)wasthefastestgrowingtechnologyintermsofnet-newinvestmentamongallorganizations.Thattrendcontinuesthisyear,withAIorMLactu-allygrowingtwopointsfasterthanlastyear.AIisnowinvestedinbyalmosthalfofallorganizations,soitremainsinour"emergingtech"quadrant.Withagrowthscoreof64,AIistheclearleaderinthisquadrantintermsofgrowth.Yet,itremainsbehindothermoreentrenchedtechnologiesthatcontinuetoseemoreinvestmentplanned:cybersecuritysolutions,cloudcomputing,anddatamanagementsolutions.Newthisyear,we'veadded"hardwaretoaccelerateAItraining"toourtechnologieslist.GiventhemarketdemandforGPUsandneuralprocessingunits(NPUs)tosupportAItrainingandinferenceoperations,wewantedtogetasenseofhowmanyorgani-zationsareinvestinginthisarea.Slightlymorethanone-thirdoforganizationshavealreadycommittedinvestment,andslightlymorethanhalfplantoinvest.Thisplacesthetechnologyjustintotheemergingcate-goryaswell.TECHTRENDS2025//01025個EXPONENTIALAIINFO-TECHRESEARCHGROUPWefurtherclarifiedwithrespondentsinvestedorplanninginvestmentinAIiftheywereplanningtoincludeinvestmentingenerativeAI.OfallAIinvestors,87%arealreadyinvestedorplanningWefurtherclarifiedwithrespondentsinvestedorplanninginvestmentinAIiftheywereplanningtoincludeinvestmentingenerativeAI.OfallAIinvestors,87%arealreadyinvestedorplanningtoinvestingenerativeAI.100%Artifi9cialintelligencerfrormachinelea320rning%87%13%100%Methodologynote:Datacollectedfromresponsesacrosstwoquestionscoveringinfrastructureandhardwaretechnologiesaswellassoftwareandplatformtechnologies.Respondentsindicatediftheywerealreadyinvested,iftheyplannedfutureinvestment,orneither.Indexscoreswerecreatedbysubtractingnegativesentimentfrompositivesentimentonbothcurrentinvestment(x)andgrowth(y)dimensions.*GenerativeAIisasubsetofAIoGiventhetrendofarapidriseinAIinvest-ment,AIisonthevergeofexitingoutoftheemergingquadrantandintoourtransform-ativequadrant,andthatinvestmentisbeingdrivenprimarilybygenerativeAI.We'dexpectthatinnextyear'sreportAIwillcrossthatthresholdtojoinmoreestablishedenterprisetechnologies.Whetheritcontin-uestoseegrowthatthesamepaceornotwilllikelydependonhowmuchvalueorgan-izationsseeinreturnfromtheirinitialeffortswithAIandiftheyfeelconfidentinovercomingthenewrisksassociatedwithAIadoption.?100%EMERGINGGenerativeAI*GenerativeAI*DatamanagementDatamanagementservicesTRANSFORMATIVECybersecurityCybersecurityArtificialintelligence(AI)ormachinelearning(ML)CloudcomputingCloudcomputingNo-code/No-code/ApplicationprogrammingHardwaretoHardwaretoaccelerateAItrainingInternetofThings(IOT)Roboticautomatorintelliautomatlow-codeprocessgentprocessinterfaces(APIs)00%?5In-memorysolutions0%In-memorysolutionsRoboticsRoboticsand/ordrones%(Mixedrealityaugmentedorvirtualreality)(On-premisesOn-premisesserversrivatecellularrivatecellular5GorLTE)BlockchainBlockchain%%QuantumQuantumcomputingNICHEENTRENCHEDTECHTRENDS2025//1225個EXPONENTIALAIINFO-TECHRESEARCHGROUP01/imagineahumanoidfacemadeofmillionsoftinyTECHTRENDS2025//14glowingvoxelswithdatatendrilsflowingfromthebrain,3Drender,darkpurple--styleraw--v6.0TECHTRENDS2025//1425個EXPONENTIALAIINFO-TECHRESEARCHGROUPTREND01|EXPERTMODELSInvestmentinAIisslightlymorepronouncedamongITdepartmentswithhighermaturity,with80%oftransformerssayingtheyarealreadyinvestedorwillbebytheendof2025.Seventy-twopercentofaverageITdepartmentssaytheywillbedoingthesame.Forthemostpart,theinvestmentisalreadymade–onlyslightlymorethanone-quarterofallorganizationssaytheyaren'tinvestedyetbutplantoinvestbytheendof2025.%%InvestafternextyearinvestHigherHighermaturityITfirmsaremoreoptimisticaboutsolvingthechal-lengesassociatedwithunlockingAIvalue.Transformersarealmosttwiceaslikelytosaythatfast-paceddevelopmentoftechnologieslikegenerativeAIwillleadtoITgeneratinganexponentialincreaseinvalue.Thesehighmaturityfirmsareencouragedbygoodearlyresults.WhileothersmaystillbelookingfortherightproofofconcepttoapplyAI,transformersarefurtheraheadandhavediscoveredlucra-tivedeploymentsthroughseveralkeytactics.Developingexpertmodels:FoundationmodelsreleasedbyleaderslikeOpenAI,Meta,Cohere,orAnthropicprovideexcellentbreadthofknowledgebutoftendisappointwhenseekingspecializedanaly-sis.Ofcourse,theyalsolackcontextaboutspecificindustriesandorganizationsduetoholesinthetrainingdata,eventhoughitisquitevast.OrganizationsharnessingAIcansolvethisthroughdiffer-entroutes:leveragingamodeldevelopedbyanindustry-focusedvendor,collaboratingwithindustrypeerstotrainamodel,orcompletingadditionalpretrainingandfine-tuningonmodelsseques-teredinanorganization'sinfrastructure.Thoseadoptingmorecustomizedmodelsfindoutputsaremorereliableandrelevanttousers,overcomingsomecommonchallengesorganizationshavefacedwithAIaccuracy.(TechCrunch,202Augmentingexistingprocesses:RatherthancreatinganewprocessdedicatedtorenderingAIoutputsortryingtoentirelyautomatetaskswithAI,firmsfindtheyarebetteroffkeepingthehumanintheloopandaugmentingtheirexistingworkflowswithAI.Largelanguagemodelsexcelatpatternrecognitionandrapidanalysisoflargevolumesofinformation,whichcanenhancethespeedandqualityofworkforadopters.Focusingondatamanagement:Astheoldadagegoes,garbagein,garbageout.AIdoesn'thaveanyintuitiononwhatisrightorwrongandonlyknowswhatit'sbeentold.Organizationsloadingbadorredundantdataintotheirfine-tuningwithLLMswillseemoreerrorsintheoutput.Organizationswithhealthydatahygienewillbebestpreparedtotakethenextstepwithmodeltraining,andthosethatcanmakedatamoreaccessibletoworkersstandanevengreaterchancetoreapthebenefits.(InterviewwithMarinellaProfi)DemocratizingAI:Alongwithaccesstodata,theabilitytouseAIandintegrateitintotheirworkflowsisneededfornon-technicalworkersorcitizendevelopers.AIsolutionprovidersofferno-codeenvironmentswherenaturallanguagecanbeusedtosetupspecificautomationstailoredtoindividualjobs.(InterviewwithAlOrganizationsbullishonAIseeitfittingintothenextwaveofdigitaltransformation.Itcanaugmentmanydifferentbusinessprocesses,anditalsopromisestoupendsomebusinessmodelsentirely,demand-ingnewwaystointeractwithcustomersandlikelyraisingexpecta-tionsevenhigher.Aswithdigitaltransformation,manyorganizationsmayfinditchallengingtosucceeddespitemakinganinvestment(HarvardBusinessReview,July2023).CompaniesaredrawntothepromisedbenefitsofgenerativeAI,withitspowertosummarizeandAverage(n=428)Transformers(n=104)SuccessfulfirmswillnotonlygraspthetechnicalcomplexityofAIitself,butsolveseveralrelatedchallengestointegrateitAverage(n=428)Transformers(n=104)0%0%LessvaluethanbeforeSamevalueasbeforeIncrementalincreaseinvalueExponentialincreaseinvalue28%0%0%LessvaluethanbeforeSamevalueasbeforeIncrementalincreaseinvalueExponentialincreaseinvalue28%%50%01020304050607%>IntegrateAIintoexistingbusinessprocessesinawaythataugmentsthem.7%5%>HireortraintherequisiteAItalent.5%>Architectahigh-qualityandspecializeddatapipelinethatcan64%beusedtofine-tuneandpretrainfoundationmodels.64%>Navigatemyriadtechplatformsandinteroperabilityissues.Thosepersistentenoughtosolvetheseproblemswillreaptherewards.AsAIcreststhewaveoftechnologyadoptionandpushesthroughtheemergingphaseandintothetransformativephase,bestpracticestounlockvaluewillbecomemoreevident.Thoseridingtheleadingedgeofthatwavedemonstratesomeofthebestwaystoseethatreturnoninvestmentmadeasearlyaspossible.TECHTRENDS2025//1625個EXPONENTIALAIINFO-TECHRESEARCHGROUPTREND01|EXPERTMODELSGenerativeAIisprovidingvaluetotechcompanies,householdconsumerbrands,governments,andpharmaceuticalcompaniesasoftoday.Modernahasreleasedoneproducttomarketthusfar:thelife-savingCOVID-19vaccineSpikevax.Asasequel,Modernaplanstorelease15productstomarketoverthenext15years.TheaccelerationplanincludesinvestmentinChatGPTEnterprise,allowingemployeestocreatetheirownchatbots.Already,750havebeenmadeacrossthefirm.DoseIDreviewsyearsofresearchandthou-sandsofpagesofdatatomakeoptimaldoserecommendations.Researcherscanreviewtherationalefortherecommendationsandvisualizedataused.Anotherbot,ContractCompanion,isemployedinthelegaldepart-menttosummarizelegalcontractsandanswerquestionsaboutthem.(TheWallStreetJournal,April2024)DataanalyticsplatformSASViyahelpedabankingcustomertoapplynaturallanguageprocessing(NLP)tocustomercomplaintsreceivedacrossvariouschannels.SASViyaenablesbankstointegrateandgovernLLMswithintheirexistingsystemstogenerateapersonalizedresponsetothecustomerwhileaccountingfordataprivacy.Ahumanintheloopreviewedtheresponsebeforesendingit.Theprocesshelpsthebankaddresscustomercomplaintsfasterandwithbetterresponses.Organizationswantingtogetbetterandgo fasterwithintheirexistingmissionislike"90%oftheconversationsI'vebeenhavingwithcustomersinthepast18months[aboutsoratSAS.“AlessonlearnedsofaristhatLLMsalonedon’tDataanalyticsplatformSASViyahelpedabankingcustomertoapplynaturallanguageprocessing(NLP)tocustomercomplaintsreceivedacrossvariouschannels.SASViyaenablesbankstointegrateandgovernLLMswithintheirexistingsystemstogenerateapersonalizedresponsetothecustomerwhileaccountingfordataprivacy.Ahumanintheloopreviewedtheresponsebeforesendingit.Theprocesshelpsthebankaddresscustomercomplaintsfasterandwithbetterresponses.Organizationswantingtogetbetterandgo fasterwithintheirexistingmissionislike"90%oftheconversationsI'vebeenhavingwithcustomersinthepast18months[aboutsoratSAS.“AlessonlearnedsofaristhatLLMsalonedon’tsolvebusinessproblems.Youneedhumansandaplatformthatenables theintegrationofLLMsinthebusiness’exi ingsystemsinsafeways.Inthecustomercomplaintsusecase,theLLMonlygenerates thereplyofthecomplaint.Theend-to-endprocessisenabledbyagovernanceplatform.” (InterviewwithProfi)Theconsumerpackagedgoodsbranddevel-opedanin-houseAItoolcalled"Brahma."Itpullsdatafromthefirm'sconsumerinsightsstudiesacross80,000respondentsin11differentcountries.Brahmahelpsacceleratethetimetosharenewproductideaswithcustomers,shorteningthecyclefromamatterofmonthstomeredays.Marsalsodoubleditsclick-throughratesandincreasedsalesliftby70%byanalyzingcustomerjour-TheCanadianretailerusesasystemdubbed"Tetris"tooptimizeitsstorelayouts,suggest- ingtheoptimalproductplacementwithin storestomaximizeprofits(TheLogElsewhere,aCeeTeeshoppingassistanthelpscustomersinthemobileappfindtheright tiresfortheircarandpurchasethem(Retail Insider,2024).opinganAIusecaseforclientScribeWire,amediaaccessibilitysolutionsfirm.Recogniz-ingthatmanualtranscriptionwastime-con-sumingandpronetomistakes,theteamiscreatinganapproachthatwouldaugmenthumantranscribersbyprovidingthemwith"Alotofpeoplejustwanttoridethehype,tousethesefoundationmodelstodefinethebusiness.IthinkweshouldnotjustlookforaproblemtosolvewhenwehaveaTECHTRENDS2025//1825個EXPONENTIALAIINFO-TECHRESEARCHGROUP02/imagineAvastdigitalcitywithmanylargecorporateofficetowerswithflagsmadeofmillionsofglowingTECHTRENDS2025//20voxelsonaflagpole,purplesandpinks,3Drender--styleraw--vTECHTRENDS2025//2025個EXPONENTIALAIINFO-TECHRESEARCHGROUPTREND02|AISOVEREIGNTYButAIthreatensbusinessmodelsbeyondcreativeindustries.Itismakingwavesincustomerservice,education,finance,legal,softwaredevelopment,marketing,sales,andmore.ITleadersrankthelikelihoodofAIdisruptingtheirbusinessatabout3.5outof5.ThisisButAIthreatensbusinessmodelsbeyondcreativeindustries.Itismakingwavesincustomerservice,education,finance,legal,softwaredevelopment,marketing,sales,andmore.ITleadersrankthelikelihoodofAIdisruptingtheirbusinessatabout3.5outof5.Thisisaclosesecondonlytoatalentshortagecausingdisruption.AIismoreofaconcernthancyberse-curityincidentsorgovernment-enactedpolicychanges,amongotherfactors.customerspurchasingnewrecordedmusicfromartists,butnewstyleorgenrebasedonauserprompt.>ThenewsindustryreliesoncustomerstovisittheirwebsitestoshowthemArtificialintelligenceads,ortheycancollectasubscriptionfeefromtheircustomers.ButnewArtificialintelligenceAI-generatedanswersprovidedbynumberofvisitsthesewebsitesreceive,divertingadrevenueandthreateningsubscriberrevenuebycuttingoffasourceofnewprospects.ThoseinvestedinexistingbusinessmodelsintheseindustriesarefightingbackbytakingtheAIcompaniestocourt.TheRecordingIndustryAssociationofAmericaandmusiccept,RawStory,andAlternetareamonginfringement(TheNewYorkTimes,Feb.2024).Thedecisionscouldholdmajorimpli-cationsforthefutureofcreativeAIposesathreat,yetorganizationsareinvestinginittopursueitsbenefits.Thiscreatesatensionbetweentheorganizationsadoptingthetechnologyandthevendorsprovidingit;onethat'smanagedbycarefulattentiontogovernanceandcontrolmechanisms.OrganizationswishtomaintaintheirownAIsovereigntyastheyleveragepowerfulfoundationmodels.Aswe'llseeinthenextsection,maintainingcontroloverdataconfidentiality,performancestandards,andcostsareattheheartofthematter.TECHTRENDS2025//22TECHTRENDS2025//2225個EXPONENTIALAIINFO-TECHRESEARCHGROUPTREND02|AISOVEREIGNTYQualityControlTrustedEnvironment65.43%PrivacyandsecurityDigitaltransformationinvitedmoreadop-tionofcloudcomputinginorganizations,andITlearnedtoenablethatwhileensuringprivacyandsecurityalongthewayas65.43%Privacyandsecurity59.57%57.71%59.57%57.71%returnoninvestmentAccuracyandreliability24.47%PrivacyandsecurityisthemostimportantfactorwhendeterminingwhatgenerativeAIsolutionsorganizationsinvestin,selectedby65%ofrespondents.Thenextmostimpor-tantfactoriscost-effectiveness,withsixoutoftenleadersselectingit.Closebehindthatisaccuracyandreliability,at58%.returnoninvestmentAccuracyandreliability24.47%industry-specificknowledgeFlexibilityofusefordifferenttasks20.74%CostOptimizationEmbeddingincomprehensivevendorsolutionThesethreefactorsstandoutaboveallotherfactorsindetermininginvestment.Organi-zationswanttoharnessAI'scapabilitieswithoutgivingupcontrolovertheirbusiness,andjustascloudadoptionneededtocoFlexibilityofusefordifferenttasks20.74%CostOptimizationEmbeddingincomprehensivevendorsolutionAfterChatGPTAfterChatGPTwasfirstreleasedasaconsumertool,manyorganizationsissuedpoliciesaskingemployeestonotuseitoratleastavoidinputtingsensitiveinformation.ThetermsofserviceclearlystatedthatuserdatawasfairgamefortrainingfutureAImodels.Theconcernoflosingcontroloverorganizationdataremains.Whilevendorshavemadestridesinprovidingtrustedinfrastructureandcontractualpromisestonotusecustomerdatafortrainingtheirownmodels,ITleadersmustmitigateriskbyminimizingdatausageingoodaccesscontrolsareinplace,andprotectingagainstknownintrusiontacticsusedbybadactors.ManyfoundationmodelschargeforaccessbyAPIataper-tokenrate.Thisisaffordableatlowvolumebutcanbecomeexpensiveatscale.Runningmodelslocallyincurscapitalcostsforassociatedinfrastructure,andconductingtrainingonnewmodelsoradditionalpretrainingonexistingmodelsalsocomeswithinfrastructurecosts.EthicalsourcingoftrainingdataTherearebothtechnicalandqualityfactorstoconsiderwithperformance.TechnicalModelexplainabilityconsiderationsincludethesizeoftheModelexplainabilitycontextwindowandlatencyofreceivingaresponse.ContextwindowsinlargeVendorassumeslegalrisk(indemnification)8.51%foundationmodelshavegrownsignificantlyoverthelast18monthsandVendorassumeslegalrisk(indemnification)8.51%forinputsequaltomanythousandsofpages.Butdifferentmodelsofferdifferentsizesof5.32%Multimodalinputwindow(Google,2024).Accessingmodelsinthecloudyieldresponseswithinafew5.32%Multimodalinput010203040506070seconds,butinsomecasesthatwaittimeistoo010203040506070near-zeroaspossible,movingthemodelclosertothesourceofinputsandreducingTECHTRENDS2025//2425個EXPONENTIALAIINFO-TECHRESEARCHGROUPPursuingAIsovereigntycanlookdifferentbasedontheorganizationalcontext.Wewouldn’texpectagovernmentintelligenceagencytohavethesamerequirementsforcontrollevelasaretailerthatsellstires.Let'sexaminesomespecificinstanceswhereorgan-izationsfoundtherightbalancebetweenharnessingAIandmaintainingacomfortablelevelofcontrol.Likemanyotherorganizations,thebankingandinvestmentfirmstartedbybanningtheconsumerversionofChatGPToncorporatedevices.Butitdidn'tstopthereandeventu-allydevelopedacentralithatfacilitatesinteractionbetweenamenuofmarket-leadingfoundationmodelsandpropri-etaryorg