NOVEMBER13,2025

TheRiseofAI:

ARealityCheckonEnergyandEconomicImpacts

MarkP.Mills

ExecutiveDirector,NationalCenterforEnergyAnalytics

DistinguishedSeniorFellow,TexasPublicPolicyFoundationDistinguishedFellow,HammInstituteforAmericanEnergy

ARESEARCHPARTNERSHIPBETWEEN:

TheNationalCenterforEnergyAnalytics

TheHammInstituteforAmericanEnergy

2

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

ExecutiveSummary

Totalprivatesectorspendingcommittedtoartificialintelligence(AI)isnowasignificantshareofU.S.GDP.Spendingonbuildingdatacentersalone—exceeding$50billionannuallyandrising—hasnowsurpassedspendingonallothercommercialbuildingscombined.

TheevidenceofanemergingstructuralshiftintheU.S.economycanbeseeninthecombinationoftheepicspendingonAI,rapidadoptionofAItools,theimplicationsfornationalsecurity,thevigorousdebatesovertheimpactofAIinsociety,and,aswell,thestockmarketenthusiasms.

ThecoreconsequenceofAIdeployedatscaleisinitspotentialforboostingproductivity,thefeatureofeveryeconomythatdrivesgrowthandprosperity.IfdemocratizingAIelevatesU.S.productivitygrowthonlytotheaverageofthepasthalf-century,itwilladdacumulative$10trillionmoretotheU.S.GDPthanisnowforecastoverthecomingdecade.

OftenignoredinAIforecasts:theadditionalwealthcreatedbyusingthenewtechnologyleadstobehaviorsthatusemoreenergy.Anextra$10trillionwouldleadtoincreasedoverallenergyuseequivalenttoaboutfivebillionbarrelsmoreenergyoverthenextdecade.Suchawealth-inducedincreaseinenergyconsumptionwillbefargreaterthanthequantityofenergyneededtopowerthewealth-creatingAI.Butunleashingthiseconomicandstrategicopportunityrequiresbuilding,andpowering,theAIinfrastructure.

AIdatacentersarenotuniqueinthatregard.Energyisthe“masterresource”neededforoperatingeverypartofcivilization.AsNvidiaCEOJensenHuangrecentlysaid:“AIisenergy,AIischips,themodels,andtheapplications....Andweneedmoreenergy.”Asithappens,datacentersaremeasuredandtrackedintermsofpowerinwatts,notdatainbits.

WhatisuniqueaboutAIdatacentersisthescaleandvelocityofpowerdemandsnowemerging.Someindividualdatacentersnowunderconstructionwillhavecity-scalepowerdemands,andhundredsarebeingbuiltorplanned.Therateofconstruction—especiallyincombinationwithreshoringmanufac-turingandreanimatingbasicdomesticindustries—hasendedthetwo-decadeinterregnumofflatelectricsectorgrowth.

Policymakers,investors,andbusinessesintheAIsupplychainareallinterestedindiscerningjusthowmuchadditionalelectricitywillbeneededspecificallyforpoweringdatacentersandhowitwillbesupplied.Forthisanalysis,we’verevieweddozensofdifferentindustryandtechnicalreportstolookforcluesandconsensus.

ThefactsandtrendssuggestAIdigitaldemandswillrequirebuildingno

lessthanabout75GW,possiblyasmuchas100GWofgenerationby2030.Andneitheroutcomeincludestheelectricitydemandsforexpandingtheancillarybutdirectlyrelatedtelecommunicationsnetworks,aswellasthatneededforreshoredchipfabricationfacilitiesthatwillmanufacturethelogicenginesinsidethedatacenters.

Tomeetthatmuchnewdemandby2030,underlyingengineeringrealitiesshowthatmostoftheadditionalelectricitygenerationwillnecessarilycomefromburningnaturalgas.Thatwill,inturn,requireabouta10%to20%increaseinoverallU.S.gasproduction.

ThatriseingasdemandwilloccurcontemporaneouslywithroughlythesameamountofnewdemandcomingfromadditionalLNGexportterminalsthatarealreadyunderconstruction.

Thenationiscapable,technically,ofmeetingsuchalevelofgrowthinnaturalgasproduction,pipelineinstallations,andpowerplantconstruction.Theprimaryimpedimentsareinstitutionalandregulatory.

Inthelonger-term,astheAIstructuralrevolutioncontinuestoplayoutpast2030,evengreaterenergydemandswillemergetopowerthenextphaseofgrowth.Thosedemandswilllikelybemetincreasinglyfromnuclearenergyandadditionalsolarcapacity.Buteachofthoserequireassociatedinfrastruc-tureexpansionsthat,inherently,takefarlongerthanthecurrenttorridpaceofAIconstruction.Nevertheless,ifrelevantpoliciesandprojectsarenotputinplacetoday,electricityplannersintheforeseeablefuturewillbeagaincaughtflat-footedinfailingtoprepareforgrowth.

Thechallengeisthatelectricity-relatedenergypoliciesnowinplacewereframedduringtherecentperiodoflowornogrowth,combinedwithmisguidedpursuitsofanenergytransitiontoreplaceconventionalenergysources.TheAIboomhasilluminatedthefactthatnewperiodsofgrowthareinevitable—evenifdifficulttopredict—andthatsuchperiods,predictably,leadtoincreasedenergydemandsrequiringadditionsto,ratherthanreplacementsof,existingenergysystems.

Googleobservedearlierthisyearthat“AIpresentstheUnitedStateswithagenerationalopportunityforextraordinaryinnovationandgrowth”butthatitrequires“expeditedefforttoincreasethecapacityof...U.S.energysystems.”AJuly2025WhiteHousedirective,WinningtheRace:America’sAIActionPlan,calledontheprivatesectortobuildthe“vastAIinfrastructureandtheenergytopowerit.”Thatispreciselywhatisunderwayinanhistoricconvergenceofthetechnology,energy,andfinancialsectors.

3

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

KeyTakeawayswithImplicationsforPolicymakers

ThegreatAIinflectionismarket-driven:

?TheracetobuildAIinfrastructuresandservicesisbeingfundedbytheprivatesector1i.e.1itisnotapolicy-driventransformation.

?ThepaceofAIadoptiondependsonintegrationchallengesforspecificapplications1andthematurityandefficacy(power)oftheAItools1i.e.1morepowerfulandusefulAItoolsarerapidlyemergingthatwilldrivemoreadoption.

Threekeymetricsilluminatethetransformation:

?TherateofprogressintheunderlyingAIcomputecapabilitiesfollowsawell-established1predictabletrendofexponentialgains.

?AIcomputationalperformancehasincreasedathousandfoldsince2018andiscontinuingatthatpace.

?ThescaleandvelocityofprivatecapitaldeployedtobuildAIinfrastruc-tureissettingrecordsbuthasnot(yet)exceededhistorichighsincomparableperiods.

?Over$1trillionorprivatecapitalisplannedorontracktobespenttobuildAIinfrastructure.

?ThescaleofenergydemandsforAIinfrastructureswillrequiresubstan-tialexpansionofbothelectricpowerandnaturalgasproduction.

?ThelikelypaceofAIwillrequire75–100GWofnewelectricitygenerat-ingcapacitytosupplyasmuchas11000terawatt-hoursayearmoreelectricityfordigitaldemandsbytheearly2030s.

?Thenewpowerplantswilldrivetheneedfora10%–15%overallincreaseinU.Snaturalproduction,contemporaneouswithasimilarincreaseindemandforLNGexports.

Akey1enablingrequirement:meettheunprecedentedscaleof

electricitydemandofindividualfacilities1atthevelocitybeingbuilt.

?Hundredsofplanneddatacentershavedemandexceeding300MWeach1manyover1GW1withconstructioncompletionsoftenintwoorthreeyears.Therearenoprecedentsinutilityhistoryforsuchscaleorvelocity.

?Thepaceandscaleareintensionwithsupplychainsandworkforceavailabilityinasectorthatisaccustomedtoflatgrowthwithregula-tions1policies1andconventionsadoptedinrecentyearstopursueanenergytransition.

?Paceandscalearealsointensionwiththeneedtonotcompromisegridreliabilityorimposecostsonresidentialratepayers.

Engineeringrealitieswilldictateviablesolutions.TherearetwotimeframesformeetingAIpowerdemand:

?Twotofiveyears:Variousclassesofnaturalgasturbinesandenginesdominate.

?Thetechnicalcapacityexiststomeetthedemandsforgasproduction,pipelines,andpower-producingengines.Bothgrid-integrationandprivategridapproachestoprojectsareunderway.

?Fiveto10years:Feasiblefornewnucleargenerationandexpandingtransmissiontoaccessutility-scalesolar/windfacilities(requiringgrid-scalebatteries).

?Bothfacesupplychainchallenges;solar,wind,batterieswithforeignorChinasourcedinputs;nuclearwithrestoringatrophiedinfrastruc-tures.Thereisnoclearpathtosignificantaccelerationoftransmissionconstruction(naturalgaspipelineconstructionisfarfaster).

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

4

Introduction

StructureofTechnologicalRevolutions

Ithasescapednoone’sattentionthateye-wateringamountsofcapitalarebeingdeployedintheprivatesectoronartificialintelligence(AI).Similarly,it’snowcommonknowledgethatthegreatAIbuild-outisinducingincreasesinelectricdemandnotseenfordecades.

Aswithalltechnologies,whatappearstobeanovernightphenomenoninfactcomesfromyearsofengineeringdevelopmentsandincrementalprogress.Historyshowsthatwhentherelevantcapabilitiesbecomegoodenoughandcostsarelowenough,atippingpointisreached,andaninflectionhappenswithrapidand“surprising”growth.

EconomichistorianJoelMokyr,theco-recipientofthisyear’sNobelPrizeinEconomicSciences,borrowedfromphysicstheterm“phasechange”todescribesuchinflections,whentransformationalinnovationstriggereconomicandsocialrevolutions.1Thechangeintheeconomyandindailylife—frompre-topost-railroad,fromanagrariantoindustrialsociety—wasaphasechangeasdramaticasaliquidbecomingasolid.

AsprofessorMokyrwroteinoneofhismanyseminalbooks,TheLeverofRiches:“Technologicalprogresshasbeenoneofthemostpotentforcesinhistoryinthatithasprovidedsocietywithwhateconomistscalla‘freelunch,’thatis,anincreaseinoutputthatisnotcommensuratewiththeincreaseineffortandcostnecessarytobringitabout.”2

Buttechnologies’economicandsocietalbenefitsaren’tunlockedinisolation.Mokyrisnotna?veinhis“freelunch”observationbutratherusesthataphorismtoframestructuralrevolutions,i.e.,howaphasechangeoccursintherealworld.

InMokyr’sframing,suchrevolutionsoccurattheintersectionofthreeforces:newknowledgeinstantiatedthroughtechnology,theavailabilityandsourcesofcapitaltodeploytechnologies,andtheroleofinstitutionsthatenableorconstraindeployment.

FewdoubtthatAIisconsequential,eveniftherearesomewhoaremoreanxiousthanexcitedaboutthepossibilities.ThenatureoftheopportunitiesandwhatitwilltaketoensurethattheUnitedStatescancapturethebenefitsfromAIcanberevealedbyansweringthreequestionsaboutthefuture,withalltheusualcaveatsaboutpredictions,aroundwhichthisreportisorganized.

1.IsAIastructuralshiftorabubble?

2.Whataretheupstreampowerimplications?

3.WhataretheconstraintstounleashingthenecessaryenergytofuelanAIboom?

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

5

1.TheStructuralShift

1.1PillaroftheBoom:ComputePower

PerformanceGainsinAISupercomputers

Source:KonstantinF.Pilzetal.,“TrendsinAISupercomputers,”arXiv,Apr.23,2025

EconomicPerformanceGainsofConventionalComputers

Source:HansMoravec,“WhenWillComputerHardwareMatchtheHumanBrain?”JournalofEvolutionandTechnology,vol.1

(1998).

ThereleaseofChatGPTonNovember30,2022,ledtowidespreadpublicrecognitionthatAIwasnowgoodenoughtobecomeincreasinglyuseful—andtodisruptbusinesses,institutions(andthuspolicymaking),andstockmarkets.Itwastheclimaxofdecades-longtechnologicalachievements.

Thearrivalofsupercomputerspowerfulenoughtoexecutethecomplexmathematicsandcodingof“machinelearning”arrived,predictably,becauseoftheinexorable,exponentialgainsincomputeperformance(measuredinFLOPS/second,orlogicoperationspersecond,whereinconventionalcomputerprogressisoftenmeasuredinthemoregeneralMIPS,millioninstructionspersecond[MIPS]).Aswiththefirstcomputerrevolution,therateofimprovementcontinuesunabatedandexponentially.

Theerawhenbinarylogiccomputersweredemocratized,circa1980,offersananalogyforthestateandfutureoftheemergenceofpracticalAI—atechnol-ogythathasbeeninexistencefordecades,justasdigitalcomputershadbeenby1980.

Theoriginofdigitalcomputersbasedonbinarylogictracesto1937,withClaudeShannon’sseminalMITmaster’sthesis,theMagnaCartaofthedigitalage.3AlanTuring,acolleagueofShannon,observedatthetimethattheColossuscomputer(builtduringWorldWarII,justbeforetheU.S.Eniac)couldperformtasksthatwouldotherwisehavetaken,asTuringissaidtohaveobserved,“100Britonsworkingeighthoursadayondeskcalculators100years”tocracktheGermancode.4SomefourdecadeswouldpassbeforetheageofthePCwouldarrivebecauseoftheinexorable,exponentialprogressincomputepower.

Similarly—nearlyfourdecadesbeforethereleaseofChatGPT—theconceptofalearningalgorithm,of“machinelearning,”tracestoaseminal1986paperco-authoredbyGeoffreyHinton,creditedasa“godfather”ofAI.ButthesiliconenginesthatcouldrealizeHinton’svisiondidnotemergeuntil1993,whenJen-Hsun“Jensen”Huangco-foundedNvidia.5

Aswithearlier,conventionallogic,ittookdecadesofadvancementforsiliconhardwaretobecomesufficientlypowerfulandinexpensivetodemocra-tizethemassivelyparallelfunctionsinherenttoAI.Thepasthalf-dozenyearshaveseenathousandfoldincreaseinAIcomputationalpower,unlockingusefulAI—apacethatisnotslowing;indeed,evidencepointstoanacceleration.6

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

6

1.2TheSpending:DataCenters

SpendingonConstruction:OfficesvsDataCenters

Note:Spendingrecordedintheyearconstructionstarts.

Source:LydiaDePillis,“TheA.I.SpendingFrenzyIsProppingUptheRealEconomy,Too,”TheNewYorkTimes,August27,2025.

BigTechCapitalSpending

Source:IanHarnett,TheAICapexEndgameIsApproaching,FinancialTimes,October3,2025.

TheJuly2025WhiteHousedirective1WinningtheRace:America’sAIActionPlan1summarizedthestateofplayregardingtheAIimperative1issuedacallto“buildandmaintainvastAIinfrastructureandtheenergytopowerit1”andofferedadirectivethatthenationshould“Build1Baby1Build!”7

Privatesectorspendingondatacentershadbeenrunningatroughly$10billionperyearratebeforethe2022releaseofChatGPT.Sincethen1spendingtookoff—datacenterconstructionthisyearsurpassedconstructionspendingonallotherU.S.commercialbuildings.Thespendrateexceedscurrenttotalspendingtobuildeithermanufacturingfacilitiesorpowerplants.8

Someanalystsworrythatthespending1andtheassociatedescalationinstockvaluationsofallthecompaniesintheAIecosystem1upstreamanddownstream1pointstoaclassicbubble.9However1BlackRockrecentlymadeabetonthefuturewitha$40billionacquisitionofAlignedDataCenters—the”largestdatacenterdealinhistory”—inaclearsignalofthatfirm1sexpectationofalong-runtrend1notabubble.10Ofcoursethereisacorrelationbetweenthetwodomainsofwhatinvestorsthinkastockisworthandwhatisbeingbuiltandplanned1andexpectationsaboutthefutureusesofAI.

TotalcapitalspendingonAIinfrastructuresisontracktoexceed$1trillionayear1withsomeanalystspredictingacumulative$5trilliondeployedby2030.11Whileonlyhindsightwillrevealwhetherornotprivateandpublicmarketinvestorsentimentswereoverexuberant1thereareanumberofbasicindica-torsthatcanrevealwhethertheAIboomisstructural1i.e.1asecularshiftintheeconomy1orastockfad.

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

7

1.3TheSpending:CurrentContext

U.S.BusinessSectorCapitalExpenditures

Source:FederalReserveBankofSt.Louis

CloudServices:GlobalRevenues

Source:Statista

OverallU.S.businessspendingoninformationinfrastructureshas1sincethestartofthe21stcentury1dominated1dominatedoverallspendingonthefourcoreareasofprivatesectorinvestment1substantiallyexceedingcapitaldeployedforindustrialequipment1transportationequipment1andstructures.Theshareofthatinformationinfrastructurespendinghasbeenincreasinglyshiftingfrompurchasesofon-sitecomputetoenterprisecloud-basedservices.ThearrivalofAIacceleratesthetrendbecausethedominantlocusofAIspendingisoncloudinfrastructure.

Thecurrentlevelofdatacentercapitaldeployed(annualspendingonconstruction1ratherthanstatedtotalcommitments)canbeputinthecontextoftheoverallscaleofnearly$1trillioninannualspendingbybusinesstousecloudservices.RoughlyhalfofglobalcloudservicesisprovidedbyU.S.businesses.12

Ofcourse1allinfrastructuresatscalenecessarilyconsumesignificantamountsofenergytooperate.Attentiontothatrealitywastriggeredbya2024FederalEnergyRegulatoryCommission(FERC)forecastthatdrewonarangeofestimatessuggestingtheU.S.wouldneedbetween50GWand130GWmoregeneratingcapacitythanhadbeenearlierimagined1endingthetwo-decadeinterregnumofnearlyflatelectricsectorloadgrowth.13

Forcontext1theannualenergyusedbyasingle1-GWdatacenterismorethantenfoldtheannualenergyusedbycarson11000milesofsuperhighway.Itcostsabout$10billiontobuildbotha1-GWdatacenterand11000milesofhighways.Andthecostoftheenergy-usingGPUsinsidethedatacenterisanother$20billion1essentiallythesameasthe$20billioncostofthequantityofcarsthat11000milesofhighwaysupports(atpeak).14

Thus1thebiggestquestionforthe21stcentury:Areweattheequivalentof1958inbuildingoutanewAI-infused“informationsuperhighway”infrastruc-ture1orareweapproachingtheequivalentoftheyear1992whenthelastsectionofthe401000milesofinterstatenetworkwascompleted?

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

8

1.4TheSpending:HistoricalContext

StockMarketConcentrationbySector

Source:GoldmanSachs,“WhyWeAreNotinaBubble…Yet,”GlobalStrategyPaperNo.73,October8,2025

ConstructionSpendingbyChemicalIndustries:1900to1930

Source:EstimateinterpolatedfromavailableCensusbenchmarks(seeendnote18)

Inthefaceofdisruptions,analystsandjournalistsseekillustrativeanalogieshopingsomeareusefultogaugetrends.TheAIraceisoftenframedasaManhattanProjectoramoonshot.15Botharecategoryerrorsontwocounts;bothwereanchoredingovernmentspendingandprograms,andbothinvolvedsingle-purposegoals,notinfrastructures.

AnalogizingtheAIbuild-outwiththeconstructionoftheU.S.InterstateHighwaySystemdoesilluminatetheenergyfeaturesofnation-scaleinfrastruc-tures,but,alsounlikeAI,thehighwaysystemwasfinancedwithpublicfunds.Theprivatelyfinancedcontinentalrailroadsystemhasalsobeenoffered;spendingtripledfromthe1840sto1870s,peakingatover5%ofthenation’sGDP.16However,acomparisonwithAIwouldbemoreanalogoustoconsid-eringtheoverallspendingonthegreattransportationtransformationmadepossiblebytheunderlyingtechnologyofthecombustionengine.

Theemergenceoftheinternetinfrastructureoffersamorerecentanalogy,thoughittoowouldbemorerelevantintermsofthecontemporaneousexpansionoftheinternetalongwithcellularnetworksandthePC.Whilethatbuild-outwasassociatedwiththeinfamousdot-combubbleof1999,itisobviousinhindsightthatalltheenthusiasms—andthecollateralneedforphysicalinfrastructures—wereallrealizedandexceededinthesubsequentyears.Thatboomwasprivatelyfinanced,whereinthecapitaldeployedreached1.25%ofGDP—alevelthattheAI/cloudbuild-outisabouttoreach.17

Perhapsthemostsuitableanalogyistherevolutionarydevelopmentofchemicalscienceattheendofthe19thcenturyandthesubsequentemergenceofthechemicalindustryintheearly20thcentury.Chemicalscienceallowedhumanitytomanipulatethebasicbuildingblocksofmoleculesandatomstoinventandfabricateentirelynewproductsandservices,frompharmaceuticalstothepolymersthatareessentialtonearlyeverymodernproduct.Similarly,theknowledgeunderlyingtheinventionofAI,machinelearning,andlargelanguagemodelsenablesmanipulationofbasicdatatocreate—hencethenewlocutionof“AIfactories”—entirelynewclassesofproductsandservicesforeverysectoroftheeconomy.

Notably,overthetwodecadespriorto1929,privatesectorinvestmenttobuildwhatwasthenrevolutionarychemicalfactoriesandinfrastructuresrosetoover5%ofGDP.18

TomatchthatshareofGDPby2030,AIspendingwouldhavetoreachnearly$2trillionayear,alevelintherangeofsomeforecasts.19

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

9

1.5TheImpact:ProductivityDeficit

LaborProductivityGrowthRate

Source:J.P.MorganAssetManagement,TheTransformativePotentialofArtificialIntelligence,2023.

U.S.GDP:TwoProductivityGrowthScenarios

Source:“RekindlingUSProductivityforaNewEra1”McKinseyGlobalInstitute,February16,2023

Inhisacceptancespeechforthe1987NobelPrizeinEconomicSciences1RobertSolowobservedthat“technologyremainsthedominantengineofgrowth1withhumancapitalinvestmentinsecondplace.”20Thecollectiveeffectoftechnologicalprogress—enabledbycapitalmarketsandfacilitatedbyinstitutions—yieldedanhistoricjumpinU.S.laborproductivityfollowingWorldWarIIwhich1inturn1producedanenormouswealthexpansion.

Technologyhas1overvariousperiodsinhistory1yieldedsimilargainsinproductivity.From1910to19301thelabor-hoursneededpercarmanufactureddroppednearlyfivefold1asdidthelabor-hourspertonofsteel.21Therailroad-eraproductivitygainsinshippingwerevisiblenotjustfromgreaterspeedthanhorse-and-wagontransportbutalsobecauseofa25-foldcollapseintheton-milecostofshippinggoods1withsimilargainsformovingpeople.

Lookingfurtherback1theflourishingMiddleAgescamefrommachineinventionsinan“ageofreasonandmathematics1”aseconomichistorianJeanGimpelframeditinhisbook1TheMedievalMachine.Backthen1manymachinescutman-hoursneededbyasmuchastenfold1farmorethanthemere10%to30%laborcutsnowbeingattributedtoAI.22

TheessenceofAI1spromiseliesinhow—acrossmanyandvariedapplica-tions—itwillyieldmeaningfuland1eventually1trulyconsequentialproductivi-tygains.Indeed1marketadoptionisfundamentallyonlyexplicablebyvirtueoftheopportunitytoimproveproductivity.

Initseconomicforecasts1theU.S.CongressionalBudgetOfficeassumesa1.4%annualproductivitygrowthrateforthenextdecade1aslightincreaseoverthepastdecade1sanemicandhistoricallylowaverageof1.2%.23WhilesomeanalystsbelieveAIcanboosttheU.S.economytoa4%productivitygrowthrate1thehighestlevelinthepostwarperiod—andthereisgoodevidencetosupportthatoptimism—considerinsteadtheimplicationsofmerelyrestoringtheratetothepostwarlong-runaverageof2.2%.Thatwould1arithmetically1induceacumulative$10trillionofgreatereconomicgrowthoverthecomingdecadethaniscurrentlyforecast.

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

10

1.6TheImpact:ProductivityPromise

ReportedImpactsfromUsingAI

Source:The2025AIIndexReport(StanfordUniversity),chap.4,p.49

GrowthinAIPatentsIssued(2018-2023)

Source:WIPOPatentLandscapeReport,2024

Theabsenceofanyrecent,dramaticincreasesincorporaterevenueshasbeenheldupasevidencefortheabsenceofrealbenefitsfromAI.24However,theproperplacetostartlookingisincostsavingsthatyieldthesameoutput,i.e.,theverydefinitionofproductivity.

Note,forexample,theconclusionfromasurveyofcorporateusesforAIthusfar,pertheStanfordUniversity’s2025AIIndexReport:“TheareaswhererespondentsmostfrequentlyreportedthattheiruseofAIhasresultedincostsavingswereserviceoperations(49%),supplychainandinventorymanage-ment(43%),andsoftwareengineering(41%).Forrevenuegains,thefunctionsthatmostcommonlybenefitedfromtheiruseofAIincludemarketingandsales(71%),supplychainandinventorymanagement(63%),andserviceoperations(57%).”25

Inotherwords,AIproductivitygainsarecomingfirstininformation-centricapplications,theproverbiallow-hangingfruitforapplyingsoftware.Applica-tionsinphysicalsystems—whichconstitute80%oftheGDP—willtakelongerbutlikelybemoreconsequential.

Becausebusinessesfilepatentsinadvanceofwidespreaddeployment,wecanusepatentsasaleadingindicatorofwherefutureproductivitygainsarelikelytoemerge.ThedatashownotonlyaboominAIpatentsissuedoverthepasthalf-dozenyearsbutalsothatmanyareinoperationalbusinessactivitieswhereproductivitygainsareconsequential.

TheRiseofAI:ARealityCheckonEnergyandEconomicImpacts

11

1.7Adoption:EarlyDays

DevelopmentStatusSurveyofBusinessesUsingAI

Source:“AIAdoptionAcrossIndustries:TrendsYouDon’tWanttoMissin2025,”CoherentSolutions,October30,

2025

SurveyofAIUsesbyIndustry

Source:“TheStateofAIin2025:Agents,Innovation,andTransformation,”McKinsey&Company,Nov.5,

Forallbusinesses,theopportunitytoincorporatenewtechnologies—whetherhardwareorsoftware—intoexistingproductsorprocessesinvariablyentailsawell-knownlearningcurve.Therearenoexceptions.

Considermanufacturing,notonlybecauseoftherecentpoliticalrediscov-eryofthefundamentalimportanceofthatsectorbutalsobecauseofU.S.goalsandpoliciesforreshoring.RockwellAutomation’s2025globalsurveyofover1,500manufacturersfoun