人工智能與全球經濟：對央行的影響（英）

BISBulletin

No130

Alandtheglobaleconomy:

AIandtheglobaleconomy:implicationsforcentralbanks

I?akiAldasoro,LeonardoGambacorta,EnisseKharroubiandMatthiasRottner

July2026

28July2026

BISBulletinsarewrittenbystaffmembersoftheBankforInternationalSettlements,andfromtimetotimebyothereconomists,andarepublishedbytheBank.Thepapersareonsubjectsoftopicalinterestandaretechnicalincharacter.TheviewsexpressedinthispublicationarethoseoftheauthorsanddonotnecessarilyreflecttheviewsoftheBISoritsmembercentralbanks.TheauthorsthankPabloHernándezdeCos,PhurichaiRungcharoenkitkul,OlivierSirelloandPhilipWooldridgeforhelpfulcommentsandsuggestions,AdamCap,AndrewLiandEmeseKurucforexcellentstatisticalassistanceandNicolaFaesslerforadministrativesupport.

TheeditoroftheBISBulletinseriesareGastonGelosandFrankSmets.

ThispublicationisavailableontheBISwebsite

()

.

?BankforInternationalSettlements2026.Allrightsreserved.Briefexcerptsmaybereproducedortranslatedprovidedthesourceisstated.

ISSN:2708-0420(online)

ISBN:978-92-9259-971-3(online)

BISBulletin1

I?akiAldasoro

Inaki.aldasoro@

LeonardoGambacorta

Leonardo.Gambacorta@

EnisseKharroubi

Enisse.Kharroubi@

MatthiasRottner

Matthias.Rottner@

AIandtheglobaleconomy:implicationsforcentralbanks

Keytakeaways

?TheAIboomisdrivingalarge,increasinglydebt-financedinvestmentsurgeandboostingtradeandequitymarkets,generatingsizeableterms-of-tradeandwealtheffectsthatdifferacrosscountries.

?TheproductivitypayofffromAI,thoughpotentiallylarge,remainsuncertainanduneven,acrossbothsectorsandcountries.

?Bysimultaneouslyaffectingdemandandsupply,AIblurscyclicalsignals,complicatingcentralbanks’assessmentofunderlyingeconomicconditionsandmonetarypolicycalibration.

Thesustainedboominartificialintelligence(AI)hascontributedtosurprisinglyresilientgrowth.AI’sfootprintoninvestment,tradeandassetpriceshasgrownlargeenoughtoshapetheglobaloutlookinrealtime.Moreover,ithasactedasapowerfulcountervailingforcetotradetensionsandgeopoliticalshocksthathaveraisedenergypricesandweighedongrowth.Inthisway,AIhassupportedaggregatedemand,trade,financialmarketvaluationsandoverallmacroeconomicresilience.

ThisBulletinassesseshowtheAIboomisshapingnear-termmacroeconomicdynamicsandtheattendantimplicationsformonetarypolicyandfinancialstability.Itdocumentsfoursetsoffindings.First,AIisdrivingalargeinvestmentandfinancingimpulse:spendingondatacentres,semiconductorsandrelatedinfrastructurehasreachedaround1%ofGDPinthemostexposedeconomies,andtheassociatedcapitalexpendituresincreasinglyrelyonpublicdebtmarketsaswellasprivatecredit.Second,theseforcesarereshapingtradeandequitymarkets,generatingsizeableterms-of-tradeandwealtheffectsthatdiffermarkedlyacrosscountries.Third,theproductivitypayoff,thoughpotentiallylarge,remainsuncertainandunevenandisaccompaniedbysomeearlysignsoflabourmarketsofteningintheeconomiesmostexposedtoAI.Finally,bymovingdemandandsupplyatthesametime,AIblursthecyclicalsignalsonwhichcentralbanksrely,therebycomplicatingmonetarypolicycalibration.

AIasadriverofshort-termmacroeconomicdynamics

AIaffectseconomicactivitythroughchannelsthatoperateviaaggregatedemandandsupply.Onthedemandside,AIisdrivingapowerfulinvestmentcycle,liftingthevaluationsofAI-relatedfirmsandreshapingtradeflows.Onthesupplyside,AIpromisestoraiseproductivityovertime.Thedemandsideeffectsarealreadylargeandobservable;thesupplysideeffects,whilepotentiallymoreconsequential,remainuncertaininscaleandtiming.Becausetheseforcesaffectgrowth,labourmarketsandinflationindifferent–andsometimesoffsetting–directions,AI’snetmacroeconomicimpactisunclearandwilllikelyvaryacrosscountriesandovertime.

Abroadinvestmentboomanditsfinancing.ThepromiseofAIhasunleashedaglobalinvestmentboom.AI-relatedinvestmentalreadyaccountsforamaterialshareofGDPinsomecountries(Graph1.A).Thisspendingspansdatacentres,specialisedchipmanufacturingfacilities,cloudinfrastructureandhardware,andhasalsobolsteredinvestmentininformationtechnology(IT)productsandsoftwaremore

2BISBulletin

broadly.

1

AvailableestimatespointtoasharpincreaseinAI-relatedinvestment:intheUnitedStatesandAustralia,wheremoregranularinvestmentdataareavailable,estimatedspendingondatacentres(includingtheITequipmenttheyhouse)andITmanufacturingfacilitiesrosesteeplyto0.8%andover1%ofGDP,respectively(purpleandbrownlines,right-handscale).IndustryparticipantsexpectglobalAI-relatedinvestmenttogrowfurther–from$500billiontodayto$3–4trillionby2030–whilecapitalexpendituresondatacentresareprojectedtotripleby2030(Graph1.B).

AIfirmsaregrowingrapidlyandsoareAI-relatedinvestmentsGraph1

A.AI-relatedinvestmentsarematerialrelativetoGDP1

Bandareprojectedtorisefurther2

C.Revisionsto2026GDPgrowthforecasts3

%ofGDP%ofGDP

61.0

50.8

40.6

30.4

20.2

10.0

151617181920212223242526IT(lhs):Datacentres(rhs):

USDtrn

4

3

2

1

02024202620282030

Capitalexpenditure:Datacentres

AIinfrastructure

_3SGUSKRJPEAGBVNMYIDINBRPH

GDPgrowthrevisionOct25–Jun26

Contributions:Investment

PrivateconsumptionNetexports

%pts

2

1

0

_1

_2

AEsEMEs

USEACN

USAU

1ForEA,estimatesfor2025.Seetheonlineannexfordetails.2ProjectionsareforglobalspendingonAIinfrastructureanddatacentres.Seetheonlineannexfordetails.3Revisionsto2026GDPgrowthforecastsandthecontributionsofsomeGDPcomponents.Forecastsaredisplayedbypublicationmonth;surveycutoffprecedesthisbyseveralweeks.

Sources:Aldasoroetal(2026);FederalReserveBankofStLouis;IMF;USCensusBureau;Dell’OroGroup;FocusEconomics;Gartner;GoldmanSachs;LSEGDatastream;Omdia;Reuters;WindEconomicDatabase;nationaldata;BIS.

AIinvestmenthasbeenasignificanttailwindtogrowth.AI-relatedcapitalexpendituressupportedaggregateinvestment,whichhasbeenanimportantdriverofupwardrevisionstogrowthforecastsacrossseveralcountries(Graph1.C).Inlinewiththis,businessinvestmenthasgrownfasterineconomieswithhigherscoresinanAIpreparednessindex(AIPI)thaninthosewithlowerscores(Graph2.A).

2

ThefactthattotalinvestmentisalsohigherforcountrieswithahighAIPIscoresuggeststhatbusinessinvestmentgrowthhasnotcomeattheexpenseoflessinvestmentelsewhere,iehousingorpublicsectorinvestments(Graph2.B).

TheAIinvestmentboomhasalsoreshapedtradethroughsupplychainlinkagesthatrunfromsemiconductorstodatastorageanddigitalinfrastructure.Theseshiftsareclearlyvisibleintradedata,wheregoodswithhighAIcontenthaverecordedthefastestratesofexportgrowth(Graph3.A).Althoughvolumeshaverisen,alargepartoftheincreaseinexportsreflectshigherprices.GranulardatafromtheUnitedStatesshowthatexportpricesforgoodswithhighAIcontentclimbedmarkedlyoverthepasttwoyears,evenaspricesforothergoodsfellorstayedbroadlyflat(Graph3.B).

1Ontheupstreamside,theAIboomhasalsorevivedinvestmentinenergyinfrastructure,inparticularelectricitygeneration.

2TheAIPIcapturesabroadrangeofAI-enablingconditionsin2023andisstronglycorrelatedwithemploymentsharesinAI-exposed,cognitivelyintensivesectors(GraphA.1intheonlineannex).Fordetails,seeCazzanigaetal(2024).

BISBulletin3

AIpreparednessraisesbusinessinvestmentandsupportstotalinvestmentatthe

lowendofthedistribution1Graph2

A.Businessinvestment2

%pts

5

0

_5

_102023202420252026

Averagedi?erenceinbusinessinvestmentbetweenhighandlowAIpreparedness

90%con?denceinterval

B.Totalinvestment-to-GDPdistribution3

Density

0.12

0.08

0.04

0.00

101520253035

Investment-to-GDPratio(%)HighAIPILowAIPI

1AIPI=AIpreparednessindex.“HighAIPI”referstocountrieswithAIPIscoresabovethesamplemedian.AIPIscoresreflectthecapacitytoadopt,developandbenefitfromAItechnologies.Seetheonlineannexfordetails.2ThesolidbluelineplotsthedifferenceincumulativebusinessinvestmentgrowthsinceQ42023betweencountrieswithhighandlowAIPIscores.ThedashedbluelineshowsprojectionsbasedonOECDforecasts.Countrysample:AU,CA,DE,DK,FI,FR,GB,JP,KR,NL,NO,NZ,SEandUS.3LinesplotthedistributionofgrossfixedcapitalformationasashareofGDP,basedonquarterlynationalaccountsdataforcountrieswithanAIPIscoreaboveandbelowthesamplemedian.Sample:32economies(24advancedeconomiesandeightemergingmarketeconomies)overQ42023–Q42025.

Sources:IMF;OECD;nationaldata;BIS.

StrongAI-relatedinvestmenthasputupwardpressureonAIinvestmentinputprices,notleastbecauseinvestmentbyAIfirmshasfarexceededthatbysemiconductorproducers.Asaresult,theimpacthasbeenunevenalongtheAIsupplychain.UpstreamexportersofAI-relatedgoods(egsemiconductors)

–suchasKorea,ChineseTaipei,MalaysiaandSingapore–haveseenexportpricesoutpaceimportcosts,deliveringsizeableterms-of-tradegains.ThishascausednationalincomeandpurchasingpowertoincreasebysubstantiallymorethanrealGDP,whichmeasuresproductionvolumes.

AIdeliversanexportboom,butwithpricepressuresbuildingup

Graph3

AHighAIcontentgoodsdrivesurgeinexportvalues1

.…

yoy,%

30

20

10

0

_10

CAAUNLBRKRESDEJPCNUSZAIDGBMXITMYTW2025:TotalHighAIcontentOther

2024:Total

Bbutstrongdemandispushingpricesup2

February2023=100160140120100

80

60

40202320242025

HighAIcontentOther

1Annualgrowthrateofexportsfordifferentcategoriesofgoodsandtotalexports.2LinesplottheUSexportpriceindicesforgoodswithhighAIcontentandothergoods,rebasedto100inFebruary2023.

Sources:Waugh(2026);UNComtrade;ChinaCustoms;FocusEconomics;BIS.

4BISBulletin

Thesegainshave,however,beennarrowlyconcentrated.InKorea,forexample,fivefirmsaccountedfor43%ofexportearningsinthefirstquarterof2026,upfrom27%twoyearsearlier.Bycontrast,countriesexpandingtheirdigitalinfrastructureoftenneedtoimportlargequantitiesofAI-relatedgoods,withthehigherpricesofthesegoodsweighingontheirtermsoftrade.Inlinewiththis,termsoftradehaveweakenedineconomieswithhigherAIdigitalinfrastructurescores(GraphA.2.Aintheonlineannex),partlybecausethecostofimportedAI-relatedequipmentandinfrastructurehasrisen.

FinancialmarketshaveamplifiedthedemandimpulsefromAI,althoughheretootheeffectshavebeenunevenacrosscountries.TheweightofAIfirmsinequitymarketshasrisensharply,supportedbystrongearnings.IntheUnitedStates,theirshareoftotalmarketcapitalisationclimbedfrom24%to40%between2022and2025,withsimilar,thoughlesspronounced,developmentsinothereconomiesintegratedintotheAIproductionchain(GraphA.3.Aintheonlineannex).Bycontrast,equitymarketswithmorelimitedexposuretoAIproductionhavegenerallyrecordedweakerperformance.Highervaluationshavegeneratedwealtheffectsforasset-holdinghouseholds,supportingconsumption(notablyintheUnitedStatesand,toalesserextent,theeuroarea).

Beyondequityvaluations,thescaleofAIinvestmentshasimplicationsforfinancialmarketsintermsoffinancing.AsAIfirms’capitalexpenditureshaveincreasedmarkedlyinrecentyears,freecashflowshaveshrunksignificantly(GraphA.3.Bintheonlineannex,redarrows).Thispromptedashifttodebtfinancing,withtwosourcesstandingout.Thefirstisprivatecredit,whichhasgrownparticularlyfastintheUnitedStates(Aldasoroetal(2026)).OutstandingprivatecreditloanstoAI-relatedfirmsgrewfromnearzeroin2016to$200billionin2025,whiletheirshareintotalprivatecreditloansrosefromlessthan2%to8%overthesameperiod(GraphA.3.Cinonlineannex).Thesecondfinancingsourceisbondmarkets.AIfirmsraised$243billionthroughbondsin2025,upfrom$79billionin2023(againsttotalnon-financialcorporatebondissuanceofaround$3.0trillionand$2.1trillion,respectively)(OECD(2026)).WhileUS-basedAIfirmsaccountformostofthis,issuanceisalsoincreasingelsewhere,albeitfromalowerbase.

Anuncertainproductivitypayoff.AIalsoaffectsactivitythroughproductivity,althoughthesizeandpersistenceoftheseeffectsremainhighlyuncertain.EarlymicroevidencesuggeststhatgenerativeAIcandeliversizeableproductivitygains,particularlybyautomatingcomponentsofnon-routinecognitivetasks.EvidencealsosuggeststhatAIcompressesperformancedifferences,aslessexperiencedworkersoftenbenefitmorethanseniorones(GraphA.4.Aintheonlineannex).Whetherthesemicro-levelgainswilltranslateintohigheraggregatetotalfactorproductivity(TFP)remainsuncertain.

3

Whiletherangeofmacroestimatesisrelativelywide,themedianTFPgrowthgainisaround0.5percentagepointsperyear(GraphA.4.Bintheonlineannex).Cross-countryevidencesuggeststhatsomeoftheseeffectsmayalreadybematerialising,withhigherAIpreparednessasof2023correlatingpositivelywithlabourproductivitygrowthoverthesubsequenttwoyears(GraphA.5.Aintheonlineannex).

Economicgrowthandthelabourmarket.ThegrowtheffectsofAIwillreflectboththedemandandsupplysidefactorsdescribedabove.Therefore,theycurrentlyvarysubstantiallyacrosscountries,reflectingdifferencesineconomicstructure,thescaleofAI-relatedinvestment,thecountries’capacitytoadoptanddeploythetechnology,orthepresenceofdemandbottlenecksthatmayconstrainthegrowthbenefitsofAI(BIS(2026)).Theavailableevidencesuggeststhatadvancedeconomies(AEs)arebetterpositionedtocapturethenear-termgains,whileoutcomesacrossemergingmarketeconomies(EMEs)aremoreheterogeneous(GraphA.5.Bintheonlineannex;Gambacortaetal(2025)).

BeyonditseffectsonGDPgrowth,AIcouldalsoreshapelabourmarkets.Twobroadforcesareatwork:complementaritybetweengenerativeAIandtasksthatbenefitfromhumaninput,andsubstitutionforroutinecognitivetasksthatcanbefullyorpartiallyautomated.Sofar,labourdisplacementhasbeenlimited,althoughsomeearlysignsareemerginginspecifictasksandoccupations(iecallcentres,businesscentres).Manyfirmsremainina“wait-and-see”phase,experimentingwithAItechnologiesbutstillfacing

3Translatingmicro-levelgainsintoaggregateproductivityisuncertain.Itdependsonreallocationacrossfirmsandmaybuildonlygradually,consistentwitha“productivityJ-curve”(Brynjolfssonetal(2021)).

BISBulletin5

uncertaintyregardingregulation,costs,organisationaladaptationandthereliabilityofAIsystems,consistentwiththecurrent“lowhiring,lowfiring”dynamicsobservedinseveraljurisdictions.

Evenso,corporatebehaviourpointstoanintensifyingimpact.Inrecentearningscalls,nearly80%offirmsindicatedplanstoautomateagrowingshareofproductionprocessesandincreaselaboursubstitution(Graph4.A).Consistentwiththesesoftindicators,employmentgrowthintheUnitedStateshasbeenweakerinsectorswithhigherAIexposure:betweenthethirdquarterof2023andthethirdquarterof2025,employmentinhighAI-exposuresectorsgrewaround0.8percentagepointslessthaninlowAI-exposuresectors(Graph4.B).Unemploymenttrendstellasimilarstory.CountrieswithahighAIPIscoresawunemploymentratesincreaseby0.75percentagepointsonaveragebetween2023and2025;thosewithlowAIPIscoressawflatunemploymentrates(Graph4.C).

AIcouldposesomeriskstolabourmarketsGraph4

A.Inearningscalls,morefirmssignalintenttoautomate1

%ofcompanies

60

40

20

0

2020202220242026

AIadoption

Automationandlaboursubstitution

Changes,%pts

1.0

0.5

0.0

_0.5

_1.0

_1.5

23Q224Q424Q225Q425LowAIPIHighAIPI

B.FirmsinAI-exposedsectorshireC.Unemploymentratesshiftupinlesslabour2countrieswithhigherAIPIscores3

_0.5

_1.0

_1.5Q324Q424Q125Q225Q325

Employment:Response

90%con?denceinterval

%

0.0

1Shareofcompaniesreportingthefollowingintheirearningscalls:earlystageoradvanceduseofAIwithinthebusiness,significantpositiveimpactofAIoncompanyproductivityandactualorpotentialreductioninlabourinputduetoAIorautomation.Basedonearningscallanalysisusingalargelanguagemodel.2Estimatesbasedonacross-sectoralperiod-by-periodregressionwherecumulativeemploymentgrowthisregressedontheQ32024logofemploymentandQ32024sectoralexposuretoAIintheUnitedStates.SectoralexposuretoAImeasuredasthefractionofdatascientistsinsectoralemploymentinQ32024.3Solidlinesplotthemedianchangeinunemploymentraterelativeto2023.Dottedlinesdisplaytherespectiveinterquartileranges.

Sources:KansalandRungcharoenkitkul(2026);BIS(2026),IMF;OECD;USBureauofEconomicAnalysis;USBureauofLaborStatistics;S&PGlobalMarketIntelligence;BIS.

AImayalsocreateskillsmismatchesbyincreasingdemandforworkerswithAI-relatedandcomplementaryskills.Whentheskillsofunemployedworkersdonotmatchthoserequiredfornewlycreatedjobs,labourmarketmatchingefficiencymaydecline.Consistentwiththismechanism,economiesthataremoreadvancedinAIadoptiontendtoexhibitasteepervacancy-unemploymentrelationshipthancountrieswithlowerAIpreparedness(GraphA.6intheonlineannex).

Impactoninflation.Thechannelsdiscussedabovepointtoopposingshort-runinflationpressures.Ontheonehand,AI-relatedinvestment,tradeandwealth-drivenconsumptioncanboostaggregatedemandandaddtopricepressures.Ontheotherhand,productivitygains,iftheyoccur,wouldexpandsupply,helpingtocontaininflation.ConcernsaboutAI-inducedjoblossescouldalsocurbdemand,wagesand,inturn,inflationarypressures.Therelativestrengthandtimingoftheseforces,however,remainuncertain.

Beyondthestrengthanddirectionofeffects,therearedifferencesintheirtiming.AIcouldexertupwardpressureoninflationbeforebecomingdisinflationaryasproductivitygainsbroadenandsupply

6BISBulletin

adjusts.Near-terminflationaryeffectsmayalreadybeemerging,albeitmodestly(Abecasis(2026)).Disinflationaryeffectsfromhigherproductivityorchangesinworkers’perceptionsoftheirbargainingpowerarelikelytoemergemoregradually.CanonicalmacromodelssuggestAItendstobeinflationary,ashouseholdsboostconsumptioninanticipationofstrongerfutureincomegrowth.AndeveniftheydonotanticipatetheAIboom,financialmarketsmay,inturn,lifthouseholdconsumptionthroughwealtheffects.Theeffectsoninflationaremoremuted(possiblyevenreversed)iftheAI-ledproductivityboomexceedsexpectationsorifhouseholdsrestrainconsumptionagainstpotentiallabourdisplacement.

Implicationsforcentralbanks

TheconsiderableuncertaintysurroundingtheeffectsofAIraisesseveralchallengesformonetarypolicyandfinancialstability.Forone,AIsimultaneouslyaffectsdemandandsupply,inbothcyclicalandstructuralways.Moreover,theeffectsdifferacrosssectors,complicatingtheassessmentofunderlyingtrends.Strongactivitymayreflecttemporarydemandfactors(eginvestmentboomsandwealtheffects)ratherthansustainedincreasesinpotentialoutput,whileproductivitygainsareunevenandhardtomeasure.TheAIboommaythusalterkeyunobservablebenchmarks,includingthe