生成式AI鸿沟：2025年商业人工智能现状报告（英文版）.docxVIP

TheGenAIDivide

STATEOFAIINBUSINESS2025

July2025

1CONTENTS

1.ExecutiveSummary

2.TheWrongSideoftheGenAIDivide:HighAdoption,LowTransformation

3.WhyPilotsStall:TheLearningGapBehindtheDivide

4.CrossingtheGenAIDivide:HowtheBestBuildersSucceed

5.CrossingtheGenAIDivide:HowtheBestBuyersSucceed

6.Conclusion:BridgingtheGenAIDivide

2EXECUTIVESUMMARY

Despite$30-40billioninenterpriseinvestmentintoGenAI,thisreportuncoversasurprisingresultinthat95%oforganizationsaregettingzeroreturn.Theoutcomesaresostarkly

dividedacrossbothbuyers(enterprises,mid-market,SMBs)andbuilders(startups,vendors,consultancies)thatwecallittheGenAIDivide.Just5%ofintegratedAIpilotsareextractingmillionsinvalue,whilethevastmajorityremainstuckwithnomeasurablePLimpact.Thisdividedoesnotseemtobedrivenbymodelqualityorregulation,butseemstobe

determinedbyapproach.

ToolslikeChatGPTandCopilotarewidelyadopted.Over80percentoforganizationshaveexploredorpilotedthem,andnearly40percentreportdeployment.Butthesetools

primarilyenhanceindividualproductivity,notPLperformance.Meanwhile,enterprise-

gradesystems,customorvendor-sold,arebeingquietlyrejected.Sixtypercentof

organizationsevaluatedsuchtools,butonly20percentreachedpilotstageandjust5

percentreachedproduction.Mostfailduetobrittleworkflows,lackofcontextuallearning,andmisalignmentwithday-to-dayoperations.

Fromourinterviews,surveys,andanalysisof300publicimplementations,fourpatternsemergedthatdefinetheGenAIDivide:

?Limiteddisruption:Only2of8majorsectorsshowmeaningfulstructuralchange

?Enterpriseparadox:Bigfirmsleadinpilotvolumebutlaginscale-