ArtificialIntelligence’sEnergy

Paradox:

Balancing

ChallengesandOpportunitiesTransformationofIndustries

intheAgeofAIW

H

IT

E

PA

P

E

RJA

N

UA

RY

2

0

2

5IncollaborationwithAccentureAIGovernance

AllianceImages:Getty

ImagesContentsReading

guide3Foreword

4Executivesummary5Introduction61

ElectricityconsumptionofAI

71.1TheAI

life

cycle71.2The

roleof

data

centres81.3Opportunitiesto

reduceAIsystemelectricityconsumption92

AI-enabledenergytransition112.1Non-exhaustiveexampleopportunitiesforAI-enabled11electricityreduction2.2Sample

use

cases123

Primarychallengesandecosystemenablers143.1

Infrastructurechallenges

143.2

Environmentalchallenges

143.3Overviewofecosystem

enablers153.4

Regulatoryand

policyenablers163.5

Financial

incentiveenablers

163.6Technological

innovationenablers173.7

Marketdevelopmentenablers174

FutureoutlookofAIenergy

impact184.1Thedeploymentandcollaboration

landscape184.2AIandenergy–2024to

2025

outlook22Conclusion

23Contributors24Endnotes26DisclaimerThisdocumentispublished

bytheWorld

Economic

Forumas

a

contribution

to

a

project,

insight

area

or

interaction.Thefindings,interpretationsandconclusionsexpressed

hereinarearesultofacollaborative

processfacilitated

and

endorsed

bytheWorld

Economic

Forumbutwhoseresultsdo

not

necessarilyrepresent

the

views

of

the

World

Economic

Forum,nor

theentirety

of

its

Members,

Partners

or

other

stakeholders.?2025World

Economic

Forum.All

rights

reserved.

No

partofthispublicationmaybe

reproduced

or

transmitted

in

anyformorbyanymeans,

including

photocopying

and

recording,or

by

any

information

storage

and

retrieval

system.Artificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities2AsAIcontinuestoevolveat

an

unprecedentedpace,each

paper

inthisseriescaptures

a

unique

perspectiveonAI–

includingadetailed

snapshot

ofthe

landscapeatthetimeofwriting.

Recognizing

thatongoingshiftsandadvancements

are

already

in

motion,theaim

istocontinuouslydeepen

andupdatethe

understandingofAI’s

implications

andapplicationsthroughcollaborationwiththecommunityofWorld

Economic

Forum

partnersandstakeholdersengaged

inAIstrategy

and

implementationacrossorganizations.Together,these

papersofferacomprehensive

viewofAI’scurrentdevelopmentand

adoption,

aswellasaview

of

itsfuture

potential

impact.Each

papercan

be

readstand-aloneoralongside

theothers,withcommonthemesemergingacross

industries.TheWorld

Economic

Forum’sAITransformationof

Industries

initiativeseekstocatalyse

responsibleindustrytransformation

byexploringthestrategicimplications,opportunitiesandchallenges

ofpromotingartificial

intelligence

(AI)-driven

innovation

across

businessandoperating

models.ReadingguideThiswhite

paperseriesexploresthetransformative

roleofAIacross

industries.

It

provides

insightsthrough

both

broadanalysesand

in-depthexplorationsofindustry-specificand

regional

deep

dives.Theseries

includes:AIGovernanceAllianceIncollaborationwithAccentureTransformationofIndustriesintheAgeofAIAI

in

Action:BeyondExperimentationtoTransform

IndustryF

LAGSH

I

P

W

H

ITE

PA

PE

R

S

E

RI

ES

JANUA

RY2025Incollaborationwith

PwC

IncollaborationwithMcKnsey&CompanyTransformationofIndustresntheAgeofAIIntelligentTransport,Greener

Future:AIasa

CatalysttoDecarbonizeGlobal

LogisticsW

H

IT

E

PA

P

ERJAN

UARY2025IncollaborationwithBostonConsultingGroupTransformationofIndustriesintheAgeofAITheFutureof

AI-EnabledHealth:

LeadingtheWayWH

IT

E

PA

P

E

RJANUA

RY2025AIGovernanceAllianceIncollaborationwiththeGlobalCyberSecurity

CapacityCentre,UniversityofOxfordTransformationofIndustriesintheAgeofAIArtificialIntelligence

andCybersecurity:Balancing

Risksand

RewardsWH

IT

E

PA

P

E

RJANUA

RY2025Media,entertainmentand

sport

Healthcare

TransportImpactonindustries,sectorsandfunctionsImpactonindustrialecosystemsIndustryorfunctionspecificArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities3FrontierTechnologies

inIndustrialOperations:TheRiseof

ArtificialIntelligence

AgentsIntelligentTransport,

Greener

Future:AIas

aCatalysttoDecarbonizeGlobalLogisticsArtificialIntelligence’s

EnergyParadox:BalancingChallenges

andOpportunitiesLeveragingGenerative

AIfor

Job

AugmentationandWorkforceProductivityBlueprintto

Action:

China’sPathtoAI-PoweredIndustry

TransformationArtificialIntelligence

andCybersecurity:

BalancingRisksandRewardsImpacton

regionsArtificialIntelligencein

Media,Entertainment

and

SportRegionalspecificAIin

Action:Beyond

Experimentationto

TransformIndustryTheFutureofAI-EnabledHealth:Leadingthe

WayCrossindustryAdvancedmanufacturingandsupplychainsUpcomingindustryreport:TelecommunicationsUpcomingindustryreport:ConsumergoodsAIGovernanceAllianceIncollaborationwithAccentureTransformationofIndustriesintheAgeofAIArtificialIntelligence’sEnergyParadox:

BalancingChallengesandOpportunitiesWH

IT

E

PA

P

E

RJANUA

RY2025AIGovernanceAllianceIncollaborationwithAccentureTransformationofIndustriesintheAgeofAIArtificialIntelligencein

Media,EntertainmentandSportWH

IT

E

PA

P

E

RJANUA

RY2025AIGovernanceAllianceIncollaborationwithAccentureTransformationofIndustriesintheAgeofAIBlueprinttoAction:China’sPathtoAI-Powered

IndustryTransformationWH

IT

E

PA

P

E

RJANUA

RY2025ArtificialIntelligence

inFinancialServicesAdditionalreportstobeannounced.AIGovernanceAllianceIncollaborationwithAccentureTransformationofIndustries

intheAgeofAIArtificialIntelligence

inFinancialServicesW

HIT

E

PA

P

E

RJA

N

UA

RY202

5Leveraging

Generative

AIforJobAugmentationandWorkforceProductivity:Scenarios,CaseStudiesandaFrameworkforActionINS

IG

HT

R

E

PORTN

OV

EM

BE

R

2

024FrontierTechnologiesinIndustrialOperations:The

RiseofArtificialIntelligenceAgentsW

HIT

E

PA

P

E

RJA

N

UA

RY202

5ConsumergoodsFinancialservicesTelecommunicationsTransformationofIndustries

intheAgeofAIIncollaborationwithBostonConsultingGroupachievementofefficiencygains.To

achievethis,it

s

pivotalto

understand

innovative

mitigationstrategiesandsolutionsthatcan

effectivelyfacilitate

this

balance.Overthe

pastyear,theWorld

Economic

Forum

sAIGovernanceAlliance

has

united

industryandgovernmentwithcivilsocietyand

academia,establishingaglobal

multistakeholderefforttoensureAIservesthegreatergoodwhile

maintaining

responsibility,

inclusivityandaccountability.

Players

fromacrosstheAIvaluechainareconvenedtocultivate

meaningfuldialogueonemergingAI

issues.WithAccentureasa

knowledge

partner,thealliance

sAI

Energy

ImpactCommunity(composed

ofover40global

members)

hasfacilitated

cross-

industrydiscoursetowardsconsensusandsurfacedapplied

usecasesonAI

s

energy

impact.This

paper

highlightscross-industry

insights

fromadiversestakeholder

groupto

outline

mitigationstrategies:Identifyingelectricity

use

reductionstrategies

forAIsystemsTouching

uponAI

s

potentialforthewider

energytransitionOutlining

key

partnerships,frameworksand

policiestosupportsustainableAIadoptionTheincreaseinAIadoption,alongside

other

market

factorsiscontributingtoincreasedelectricity

use.Annualglobalelectricitydemandgrowthis

nowforecastedtoreachnearly3.5%

inthecomingyears.3,4

Thischallengeisamplified

by

globalcompetitionforAIprojectsacrossregions.Thiswillrequirestakeholdersacrossthevaluechaintonavigatemarketpressuresforcomputing

power,whilebalancingsustainabilitytargets,gridconstraints

andcommunityimpacts.Intoday

seconomy,artificial

intelligence(AI)

systems

offer

bothchallengesandopportunities.As

integral

componentsofdigitalinfrastructure,thedatacentres

thatenableAIsupportavarietyof

applications,fromcloudcomputingtocomplexdata

processing.

AI

s

rapidexpansion,however,isaccompanied

bygrowingelectricitydemand,withthelargestfacilitiesintheworldusingthesameamount

of

power

assmallcitiestoensureuninterrupted

operation.

Data

centrescomeinvaryingsizeshowever,

rangingfrom

large,hyperscalefacilitieswith

morethan

1gigawatt

(GW)ofpowercapacity,

to

smaller,

micro

edgedeploymentsthat

maydraw

lessthan

10

kilowatts

(kW)

of

power.1Oneestimatenowexpects

data-centre-relatedelectricityconsumptiontogrowfromapproximately

1%ofglobalelectricity

demandto

over

2%

by2026,

potentially

reaching3%

by2030

ifforecasted

growthcontinues.2

Such

projections

have

raisedconcernsaboutsupportingthisdemandwhile

also

meeting

net-zerocommitments.Simultaneously,AI

can

bea

powerfultoolto

positivelysupportwider

energysystemtransformation.

Forexample,

it

isalready

being

usedto

improveenergyefficiency

across

industries,accelerate

renewableenergy

integrationand

make

powergrids

more

resilient.This

istheAIenergy

paradoxbalancingthese

challengesagainstAI-enabledopportunities.However,currentestimatesofAI

senergy

impact

vary,andthe

magnitudeofelectricity

demandgrowth

remains

unclear.Other

issues

includea

lackofstandardizedtaxonomiesanddefinitions.Theextenttowhichelectricitydemand

growthwillbeoffset

byefficiencygainsfromadvancementsintechnologies(e.g.chips,algorithms

etc.),

datacentredesignandchanging

regional

dynamicsisalso

uncertain.Whilea

near-term

rise

inAI

selectricityconsumption

isexpected,thefuturemagnitudeofthisgrowth

maydecline

duetotheArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunitiesForewordCathy

LiHead,AI,

Dataand

Metaverse;Deputy

Head,Centrefor

the

Fourth

IndustrialRevolution;

Member,ExecutiveCommittee,World

Economic

ForumRoberto

BoccaHead,Centre

for

Energyand

Materials;

Member,ExecutiveCommittee,World

Economic

ForumJames

MazurekManaging

Director,

US

Utilities

Strategy

Lead,AccentureArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities4JeremyJurgensManaging

Director,WorldEconomic

ForumJanuary2025Artificial

intelligence(AI)

is

facilitating

a

new

eraofinnovation,withnearlythreeinfour

companies

usingAIforatleastone

businessfunction.5Thisinnovationbringsmany

benefits,

includingenhancedproductivity,newwaysofworkingandrevenuegrowth.AI-relatedelectricityconsumptionisexpectedtogrowbyas

muchas

50%

annuallyfrom2023to2030.AIdatacentreconsumption,whilegrowingrapidly,is

projectedto

remaina

small

fractionofglobalelectricitydemand,startingatjust0.04%in2023(see

Figure

4).

However,

whencombinedwithothermarketfactors(suchasgrowing

electricitydemandfortransport,buildingsandmore),

AI’sacceleratedadoptioncouldpotentially

increase

thestrainonpowergridsand

electricity

providers.However,suchprojectionscanvary.6

UncertaintyremainsaroundhowprofoundAI’soverall

energyimpactwillbeandwhichstrategiescould

mitigate

challengesthatariseorenablenew

solutionopportunities.

Inthiscontext,it’sessentialtoassess

howAIcouldacceleratetheenergytransitionin

line

withnet-zerogoals,aswellaswhichsupportingecosystemenablerscansupportthis.ThispaperfocusesonAI’selectricityimpactswhileaddressing

thebroaderenergylandscape,

includinggeneration

andfuelsourcessupportingAI.Work

undertheAIGovernanceAlliance(AIGA)AI

Energy

Impact

Initiative

hassurfaced

key

insights

onthesetopics.The

initiativecollaborateswithover40globalorganizationsacross

more

than

nine

industriesdrivingAIadoption.Thisanalysis

highlights

keyfindings

relevanttothree

distinctareas

relatedtoAI’s

role

intransformingenergysystems:1.ElectricityconsumptionofAI:ReviewingtheAIlifecycle,strategiesforreducingits

consumption

andnewopportunitiesforprocessdigitalization–AIadoptionvaries

bysector,with

electricity

demandexpectedto

risesharply.

However,

projections

remain

uncertain,

underscoring

a

needforongoingassessment.–OptimizingAI’sconsumption

includesharnessingtechnological

innovationssuch

asenergy-efficientAIchip

hardwareandAI-

optimizedcoolingsolutions.–Companiesare

reducing

datacentreelectricityconsumptionthrough

operationalstrategies

likeAI-drivenenvironmentalcontrols,servervirtualization

andworkloaddistribution.2.

AI-enabled

energy

transition:Exploringinnovative,emergingcompany

usecasesandthe

potentialforscalingacross

industries–Existing

usecases

demonstrate

reduced

energyconsumptionof

10-60%

insome

instances,with

potentialfor

furtheroptimization.–AI

is

helpingelectricity

providersoptimize

operationsviaenergystorage,enhanced

batteryefficiencyand

smart

grid.–

AIcansupportdecarbonization,

helpingto

loweremissions,

reducewasteand

improve

resource

use.3.

Primarychallengesandecosystemenablers:Analysing

regulation,

policyand

partnerships

necessaryforsustainableAIadoptionat

scale–Enabling

sustainableAI

requires

amultifacetedapproachspanning:

regulation

and

policy,financial

incentives,technological

innovationand

marketdevelopment.–Regulatory,

policy

andfinancialenablerscan

incentivize

responsibleAI

throughcomplianceframeworksand

funding

mechanisms.–Technological

innovationand

marketdevelopmentfoster

research,collaboration

andsustainableAIadoption.Thiswhite

paper

isa

preliminaryexplorationof

AI’senergy-related

impact,andoutlinesthe

key

challengesandopportunitiesthatemerge

asAI

adoptiongrowsacross

industries.

Itconcludes

bysharingfourareasto

monitorfor

continued

understandingofAI’sevolvingenergy

impact:–

AIdeploymentfordecarbonization–TransparentandefficientAI

electricity

use–Innovation

intechnology

and

design–Effectiveecosystem

collaborationExecutivesummaryArtificialintelligencepresentsenergyopportunities

andchallenges–strategic

mitigationcan

helpto

maximize

benefitswhile

reducing

burdens.Artificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities5FIGURE1Electricitydemandgrowth

by

end

use

inthe

Stated

Policies

Scenario

(STEPS)

2023-2030,ingrowingelectricitydemand,

but

predicting

AI-specificenergy

impacts

remainscomplex.Introductionemergedasa

powerfultransformationalcatalystandthe

risingadoptionofdigitaleconomysolutions.

capableofautomatingtasksand

reinventingAI

is

revolutionizing

industries,

resultingGrowing

demand

for

AI

Overall

electricity

demandOther

Heavy

industryacross

industries

growth

driversinnovation,

increasingefficiencyandchanging

howother

growth

drivers

include

industrial

shifts

towardstoenablingcomplex

problem-solving,AI

isdrivingtheelectrificationof

bothtransport

and

buildings,societyoperates.

In

particular,generativeAI

haselectric

motors,

urbanization,

populationgrowthArtificial

intelligence

(AI)

istransformingseveralSeveral

marketfactorscontributeto

increasedperformanceandcompetitiveness.7

however,astechnologicaladvancementsandAI-relatedelectricitydemandgrowth

plays

inthe

contextofglobalenergy

trends.aspectsofdaily

life.

Fromautomatingsimpletasksglobalelectricitydemand.

Aside

from

AI

andprocessesacrossvaluechains,therebyenhancingProjectingAI-specificgrowth

is

challenging,Data

centres

6760

TWhArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities6Source:

International

EnergyAgency(IEA).(2024).

WorldEnergyOutlook.differingadoption

ratescomplicate

predictions.While

Figure

1givessome

indication,furtherresearch

is

neededtoelucidatethe

rolethatanddatacentresensitivitycasesElectricitydemandgrowth,2023-30ElectricvehiclesOthertransportOther

buildingsSpace

heatingSpacecoolingOther

industryDesalinationElectricityconsumption

of

AIModeldeployment

isAI

smostenergy-intensive

stage(accountingforapproximately

60%)innovativestrategiescan

mitigateconsumption.*Insufficient

data

available

for

estimationSource:

Electric

Power

Research

Institute(EPRI).(2024).PoweringIntelligence:

Analyzing

ArtificialIntelligenceandDataCenter

EnergyConsumption.

InternationalEnergy

Agency(IEA).(2023).

Tracking

Data

Centres

and

Data

Transmission

Networks.https://www.iea.org/energy-system/buildings/data-centres-and-data-transmission-networks;

D.

Patterson

et

al.(2022).The

Carbon

Footprint

of

Machine

Learning

Training

Will

Plateau,Then

Shrink.

Computer,vol.

55,

no.

7,

pp.

18-28.

https://ieeexplore.ieee.org/document/9810097.Furtherresearchisneededto

estimate

consumption

forstages

1and5,howeverestimates

existfor

stages

2-4.Withinthesethreestages,modeldeploymentisthemostenergy-intensive(approximately60-70%

ofcombinedelectricityconsumption),butwilllikelycontinuegrowinginthelongterm.

Modeltraining

isthenextmostenergy-intensive,accountingfor20-40%ofconsumption,followedbymodeldevelopment

at

upto

10%.9

Theseestimateshowever,will

likelyvaryacrossdifferingAImodeltypes.TheAI

lifecycle

beginswith

planninganddatacollection,duringwhichdataisgathered,

processed

andstored.8

Next,the

modeldevelopment

phaseincludesdesign,

problemanalysisand

datapreparation.

Modeltrainingthenoptimizesthe

modelthrough

iterativedataexposure.

Model

deploymentsubsequentlyopensthemodel

for

real-worldapplication.

Lastly,

monitoringand

maintenancesupportongoing

refinement.1.1FIGURE2Stage5:Monitoringandmaintenance*</>Stage4:Deployment60%Stage

1:Planningand

datacollectionsonnature*The

AI

life

cycleArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities7ElectricityconsumptionacrosstheAIlifecycleStage

2:Modeldevelopment10%Stage3:Modeltraining30%1C41253RacksSecurityTheroleofdata

centresHarnessing

powerfulservers,specialized

hardware

andadvanced

networkingcapabilities,datacentres

enablethehigh-speedcomputationsand

dataprocessing

requiredforAI.Withindatacentres,electricity

consumption

includesthree

maincomponents:10–ITequipment

(40-50%),

including

servers,

storageandnetwork

systems.Exampledatacentrelayout–Coolingsystems(30-40%)to

maintain

optimal

temperatures.–Auxiliarycomponents(10-30%),

including

power

supplies,securityand

lighting.Notethatthese

proportionswillevolveovertime

as

AI

use

becomes

more

prevalent.1.2FIGURE3

Enginegenerators

*UninterruptiblepowersupplySource:Vianova.(n.d.).Data

Center

offer.https://www.vianova.it/en/data-center/. Fire

system

HoldandcoldaislesArtificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities8UPS*

Cooling

DSource:

International

Energy

Agency(IEA);Goldman;Accenture.Enabling

a

more

energy

efficient

AI

system

includesexploringopportunitieswithindatacentrestoreduce

electricityconsumption.Accordingly,anon-exhaustive

inventoryofexamplestrategiesareexploredbelow.DatamanagementstrategiesWithinAI’sfirststage(planningand

data

collection),

“digitaldecarbonization”techniquescanaddressNon-AIdemand(TWh)

AI

demand

(TWh)Note:This

is

an

extrapolated

scenario

that

extends

the

IEA’s

forecast

from

2023to2026

through2030

using

a

combination

of

2021-2023

historical

growth

andtheir

proposed

growth

rate

from

2023-2026.“darkdata”,whichoccupiesserver

space

and

consumeselectricitywithout

providingvalue.Forsomeorganizations,darkdata

may

account

forasmuch

as

60-75%

ofstored

data.11Digitaldecarbonizationstrategiescan

identifyand

eliminatedarkdata,

reducingstorageand

electricity

consumption.Opportunities

mayalsoexisttorepurposedarkdatato

generatevalue.FIGURE4Data

centre

demand

over

timeDatacentredemand(TWh):

Non-AI

versusAITABLE1Featured

data

management

use

caseLoughboroughUniversity:automotiveindustrycollaboration:

unlockingdarkdataforsustainable

industrial

maintenance1.3Opportunities

to

reduce

AIsystemelectricityconsumptionResultsIntotal,

10-20%of

dark

datawastransformed

intoactionable

knowledge,improvingfaultanalysisand

maintenance,enhancingdata

reliability,

reducingdowntime,

loweringtheenvironmentalfootprint

andhighlightingwaste

data.ApproachA

knowledge

managementsystemwithdatascrapingand

enrichmenttechniqueswasdevelopedto

integrateand

structuredarkdata,organizing

it

intovaluable

datasetsfordecision-making,andwaste

categoriesfor

disposal.This

increasedenergy

intensity,

however,

isaccompanied

bytheadditional

benefitsthatcapabilities

likegenerativeAIcan

provide,

including

theabilityto

perform

morecomplexworkandtoenableexpandedvalueopportunities.14001200100080060040020002023Datacentreconsumption

includes

bothAIand

non-AIelements.AI

processing,

particularly

forgenerativeAI,

is

moreenergy-intensivedueto

large

modelcomplexity,

longertraining

durationsandsubstantialdata

processing.Situation/context“Darkdata”

remained

instorage,

underuseddueto

poorlystructuredformats.2024

2025

2026

2027

2028

2029

2030Artificial

Intelligence,s

Energy

Paradox:BalancingChallengesandOpportunities9Source:Communityconsultation.SAP:Aiming

for“green”data

centresApproachSAPdatacentres

track

resource

use

andminimiz