《德勤：2025年技術、媒體與電信（TMT）預測報告（英文版）（132頁）.pdf》由會員分享，可在線閱讀，更多相關《德勤：2025年技術、媒體與電信（TMT）預測報告（英文版）（132頁）.pdf（132頁珍藏版）》請在三個皮匠報告上搜索。

1、TMT Predictions 2025:Bridgingthe gapsDeloitte predicts 2025 will be a“gap year”forgen AI and the TMT sector,marked by eightcritical gaps that need to be bridged for todayspotential to be realizedARTICLE16 MINUTE READAs we look ahead to 2025 and beyond,its clear that TMT is on the verge of a signi?ca

2、nt leap forward,largely powered by rapid generative AI adoption.But to get there,the industry will need to close gaps,including:balancing gen AI infrastructure investments with monetization,addressing gender disparities ingen AI usage,managing the energy consumption of gen AI data centers,tackling t

3、rust concerns surroundingdeepfake content,discovering how best to use gen AI in media and gaming,and harnessing the power ofgen AI agents to manage and act in real time.Further gaps exist in streaming video and cloud spending.Plus,there are non-gap predictions,around new smartphones and PCs with gen

4、 AI chips on them,newstadiums and other sports infrastructure leveling up the fan experience,and telco consolidation,speci?callyof wireless players.Overcoming these hurdles will be important to help businesses and industries thrive.This blank space is caused due to formatting limitations.Text contin

5、ues on next page.Closing the gap for a brighter futureWe are at a pivotal moment in the history of human invention.Future generations will certainly look back on the choices wemake today.Deloi?es prediction that 2025 will be a“gap year”for generative AI underscores this signi?cant in?ection point.Th

6、ese gapsspanning infrastructure investment,gender disparities,energy consumption,trust de?cits,and the capabilities ofgen AI agentsare not just challenges for the industry but societal imperatives.How we collectively address these gaps willde?ne the legacy we create.Beyond gen AI,advancements in clo

7、ud computing and telecommunications are expected to bring unprecedented e?ciencies,new business models,and augmented consumer experiences.Investments in sports infrastructure and the increasingprominence of womens sports can act as catalysts for economic and social development.These trends reinforce

8、 the industrysrole in fostering innovation that enhances businesses,consumers,and broader communities.The 2025 TMT Predictions represent opportunities to create lasting impact.By navigating the path forward with trust,inclusivity,and sustainability at the forefront,industry advancements can bene?t n

9、ot only the current generation,but all those who follow.Together,we can rise to the occasion and bridge the gap to a brighter tomorrow.-Lara Abrash,chair,Deloitte USEight gaps that mark 2025 as a“gap year”for TMT1.The gen AI infrastructure and monetization gap.As we predicted last year,companies are

10、 spending tensof billions of dollars on chips and further hundreds of billions to build gen AI data centers for trainingand inference of gen AI models.While some companies o?ering gen AI enterprise software are seeingincremental revenues,the investment is 10 times(or more)higher than the return,at l

11、east for now.Thosespending the most might suggest that the risk of underinvesting in gen AI is higher than the risk ofoverinvesting.But the gap persists and seems to be widening.2.The gen AI data center electricity and sustainability gap.Proposed gen AI data centers requireunprecedented amounts of p

12、ower,preferably low carbon,which is creating a gap between their needsand the capacities of electrical grids,and companies sustainability targets.Much is being done to close itby hyperscalers,chip companies,and utilities around the world,but the gap is expected to remain in2025.3.The gen AI gender g

13、ap.Women are less likely than men to use gen AI tools for both work and play.Some of this is due to lack of trust,but womens usage of gen AI is expected to catch up to mens usage within the year in some markets.4.The gen AI deepfake trust gap.The proliferation of deepfake gen AI content(images,video

14、,and audio)is making it harder for consumers,as a society,to trust their own eyes and ears.That gap needs to bebridged by the gen AI ecosystem comprehensively and immutably labeling gen AI content,as well asreliably and accurately detecting fake images in real time.The marginal cost of creating conv

15、incing deepfakes is falling,and the cost of detection needs to fall at an equivalent pace to help close the gap.5.The studio gen AI usage gap:Many expect large studios to be using gen AI for content production,andsome are,but there is a gap between those expectations and reality.Many are cautious ab

16、out challengeswith intellectual property inherent to generative content,but they are keen to gain enterprise capabilitiesthat can reduce time,lower costs,and expand their reach.6.The autonomous gen AI agent gap.The prospect of autonomous bots that can consistently and reliablycomplete discrete tasks

17、 and orchestrate entire work?ows is tantalizing.Agentic AI pilots are launching in2024will they reach widespread adoption in 2025?7.The streaming video gap.Many media and entertainment companies assumed consumers would“buyand hold”multiple subscriptions.Instead,customers are looking to cut costs by

18、bundling their favoritesubscriptions and dropping others.We now see the number of services per household not merelystagnating but shrinking,and streamers increasingly relying on bundling to help?ll the growth gap andusing other parties to aggregate and distribute their content.8.The cloud spending g

19、ap.One of the original selling points of using the cloud was the claim that it wascheaper,but in reality,spending is often decentralized and poorly controlled.Some buyers are leaninginto FinOps to bridge the gap between promised cost savings and current spending to manage theircloud spending and pot

20、entially save billions.New this yearThis year,were introducing two new sections containing 11 additional mini predictions between them.OurUpdatessegmentrevisits seven topics from previous TMT Predictions reports to ask:“Howd we do?”while also exploring the latest predictions onthose subjects.(Spoile

21、r:We did really well!)Next,in ourRising trendssection,we unveil four cu?ing-edge topics in TMT.While these emergent themes may not have made it into mainstream forecasts just yet,we believe they could be the hiddengems of tomorrows industry conversations.2025 topicsHeres a quick look at our main top

22、ics,plus those from our two new sections:Women and generative AI:The adoption gap is closing fast,but atrust gap persistsFor women to reap the full rewards of gen AI,tech companies should work to increase trust,reduce bias,and strive for more representative workforcesDeloitte predicts that womens us

23、e of generative AI will equal or exceed that of men in the United States bythe end of 2025.Although their use of gen AI was half that of mens in 2023,their pace of adoptionsuggests theyll reach parity in the United States within the next year and in many European countrieswithin one to two years.Des

24、pite accelerating adoption,women express less trust that gen AI providers willkeep their data securewhich may inhibit their full engagement and their gen AI spending.To helpovercome this,tech companies should enhance data security and data management practices,mitigate AIbias,and improve womens repr

25、esentation in their AI ranks.As generative AI asks for more power,data centers seek morereliable,cleaner energy solutionsThe tech industry should optimize infrastructure,rethink chip design,and collaborate with electricityproviders to help secure a more sustainable future for data centersAI-driven d

26、ata center power consumption will continue to surge,with Deloitte predicting data centers willonly make up about 2%of global electricity consumption by 2025.This growth is expected to emerge fromhigh-density data center infrastructure to support massive computing power and cooling needs.Butregulator

27、y,infrastructure,and cost issues are posing challenges for electricity generation and the grid tokeep pace with data centers unprecedented demand for 24/7 reliable energy.Technology and electric powerindustries can jointly address these issues by increasing the use of carbon-free sources,improving e

28、nergye?ciency in gen AI chips and algorithms,and re-balancing compute-intensive AI workloads.Ambitious stadium projects aim to bridge public-privateinvestment goalsSports owners transform stadiums into destinations,driving socioeconomic growth,communityengagement,and revenue diversi?cationThe sports

29、 industry has repeatedly demonstrated its ability to act as a catalyst for economic and socialdevelopment,with stadiums largely at the epicentres of their communities.Investment in sportsinfrastructure is seeing an upward trend,as these developments often instigate wider returns to both thepublic an

30、d private sectors.With growth as a common goal between the public and private sectors,governments and communities can work with sports investors to provide supplementary investments ininfrastructure and supercharge the socioeconomic impact of sports,enhance fan engagement,and diversifyrevenue stream

31、s for the organization.In 2025,we expect to see new stadium developments continue atpace,with almost half of these new projects expected to take place across North America and Europe.Autonomous generative AI agents:Under developmentAutonomous gen AI agentsagentic AIcould increase the productivity of

32、 knowledge workers and makework?ows of all kinds more e?cient.But the“autonomous”part may take time for wide adoption.Deloitte predicts that in 2025,25%of companies that use gen AI will launch agentic AI pilots or proofs ofconcept,and this?gure will grow to 50%in 2027.Agentic AI has the potential to

33、 complete complex tasksautonomously,improving the productivity and e?ciency of knowledge workers.Some of todays mostpromising applications include software development,customer support,cybersecurity,and regulatorycompliance.The pace of improvement for agents is accelerating,but like most new technol

34、ogies,widespread use will take time.That said,some agentic AI applications,in some industries,and for some usecases,may see actual adoption into existing work?ows in 2025.Deepfake disruption:A cybersecurity-scale challenge and itsfar-reaching consequencesAs the e?ort to detect and combat fake conten

35、t escalates,the costs of maintaining a credible internet mayfall on consumers,creators,and advertisers alikeAs AI generates increasing amounts of online images and video,questions around content authenticity andthe potential harms of fake content grow more urgent.Online platforms,tech companies,and

36、mediaplayers are taking two complementary approaches:using technology(often AI)to detect and?ag fakes andusing cryptographic metadata to assure provenance of authentic media assets.Deloitte predicts that thismarket will follow a similar pattern as that of cybersecurity,with bad actors?nding ways to

37、thwartdetection tools and industries collaborating to con?rm the credibility of at least some online content.Cloud gets lean:FinOps makes every dollar work harderEnterprise cloud spend is growing,and using FinOps strategies can make each dollar work harder.Companies can save money,boost value,and bu

38、ild cross-functional cohesion.Global cloud spend is set to top US$825 billion in 2025,1 but ask an organizations leaders what theyspend,and it might be di?cult for them to answer.However,in 2025 Deloitte predicts more companiesthan ever will turn to“FinOps”,a set of tools and strategies to measure a

39、nd optimize cloud spend,to savean estimated US$21 billion.Companies can start simple,acting to reduce cloud waste,take advantage ofdiscounts,and proactively right-sizing compute,network,and storage.But advanced companies could alsodrive cultural change,such as making business units?nancially account

40、able for their portion of the cloudbill.The goal is a“cloud unit economics”modellinking each dollar of spend to the business value itgenerates,so that companies can make more e?ective decisions about IT.On-device generative AI could make smartphones more excitingif they can deliver on the promiseWit

41、h specialized chips and extensive mobile OS integration,smartphones could become smartevenintelligent.Will users embrace the new approach?Deloitte predicts that in 2025,global smartphone shipments will see a modest lift to around 7%,up fromabout 5%annual growth in 2024.Some of this lift will be due

42、to the device upgrade cycle,which has beendown the past two years,and some will be from early adopters seeking new generative AI capabilities.Smartphones with on-device generative AI capabilities will test the value of features like intelligentassistants and conversational interfaces;the capabilitie

43、s of small models running on-device;and the businessmodels seeking economic value from the capital intensity of the generative AI buildout.There is excitementabout generative AI,but can the technology deliver on its promises,and will users embrace a new way ofinteracting with the most widely used co

44、nsumer device?Large studios will likely take their time adopting generative AIfor content creation.Social media isnt hesitating.Hollywood(and others)may be cautious about using gen AI for content creation,but they will likely bequicker to adopt it for operations and distributionIn 2025,Deloitte pred

45、icts that the biggest TV and?lm studiosespecially those in the United States andEuropean Unionwill be cautious in adopting generative AI into their creative work?ows,with less than3%of their production budgets going to these tools.But we also predict that operational spending willexpand by 10%to int

46、egrate generative AI enabled tools for more bread-and-butter functions like contractand talent management,permitting and planning,marketing and advertising,and localization and dubbingof content that can expand their reach into diverse global markets.This approach can help studios slow thepotential

47、disruptions that gen AI can pose to talent and content,while more quickly adopting gen AI toolsthat can help reduce costs and accelerate performance across their businesses.Reevaluating direct-to-consumer:The shift toward videoaggregatorsVideo content creators may need more distributors to reach the

48、ir total addressable marketConsumers are expected to continue to stack streaming video services and have more than one standalonesubscription at a time in 2025,but Deloitte predicts that the stack is about to get shorter.According toDeloitte surveys,after reaching more than four services per consume

49、r in the United States and over two inmost European markets in 2023 and 2024,we appear to have passed the peak,and the number ofstandalone services will slowly decline in most markets.Instead of consumers directly subscribing to eachcontent providers service,there will likely be increased aggregatio

50、n,where intermediariesranging fromtelcos to grocery stores to tech platforms to streamers themselvescombine multiple content sources in asingle package.This trend is likely a win for many of the players,keeping costs under control and creating astable and sustainable streaming ecosystem for 2025 and

51、 beyond.Wireless telecom consolidation speeds up where regulatorsallowIn many markets,smaller wireless telecoms see slow growth,low pro?ts,and have debt to repay.M&A,speci?cally combining assets or even entire consumer-facing companies,may help where it gets approved byregulators.In some markets,esp

52、ecially Europe and Asia,there is an increasing perception that wireless markets are toofragmented,subscale and unsustainable,with the smaller third and fourth place operators not able to investin networks over the long term.Although these markets have historically kept the number of operatorshigh,re

53、cent conversations have seen opportunity to allow or even encourage consolidation.Deloittepredicts that although it is expected to be a slow process,and regulators will have their conditions,therewill be an increased pace of consolidation,beginning in 2025 and continuing,creating a more viable andsu

54、stainable wireless ecosystem,especially in smaller markets.UpdatesThis year were looking at seven previous predictions to see how we did,and what the latest developmentsare:Generative AI comes to the enterprise edge:On-prem AI is aliveand wellOwning their own servers gives companies a more private,s

55、ecure,?exible,and possibly cheaper ITenvironment for AIDeloitte predicts that although gen AI via cloud will continue to be the dominant option in 2025,abouthalf of the enterprises worldwide will add AI data center infrastructure on premises,primarily to protecttheir IP and sensitive data,comply wit

56、h data sovereignty or other regulations,and save costs.Deloittes2024State of Generative AI in the EnterpriseQ2 survey noted 80%of companies with“very high”AIexpertise reported spending more on AI in the cloud but 61%are investing more in their own hardware.Enterprise gen AI will likely be a hybrid a

57、pproach,with enterprises doing some in the cloud,and some onpremise.(Re)defining the investment case for womens sportsRising womens sports revenue fuels investor interest and valuation recordsThe increasing professionalization and commercialization of womens sports around the world is garneringthe a

58、ttention of fans,sponsors andcriticallyinvestors.In 2024,we predicted the womens elite sportsmarket would generate over$1 billion in revenue.In North America,clubs are recognizing recordvaluations,from Angel City FC in the National Womens Soccer League at a valuation of$250 million2 toLas Vegas Aces

59、 in the Womens National Basketball Association at$140 million.3Elsewhere,organizationsare creating innovative structures to channel investment into their womens teams and emphasizing strategicgrowth,independent leadership,and commercial opportunities.In 2025,we expect to see an expandinggroup of inv

60、estorsincluding institutional investors,private equity and high net worth individualstakemore note.Fixed wireless access:Contrary to popular opinion,adoptionmay continue to growWith US FWA net adds likely being slightly lower than last year,and some markets not expected to take o?until 2026 there ma

61、y be pockets of unrealized or potential growth out there,both in the US and globallyFixed wireless access(FWA)when consumers and enterprises get their home broadband over a?xedcellular device(mainly 5G)rather than via wireshas beenthe5G growth story over the last few years inthe United States,with w

62、ell more than 10 million homes expected to be connected by the end of 2024.However,growth is slowing,with the?rst quarter of 2024 net additions lower than Q1 2023,and apotential slowdown anticipated in 2025.Despite this,Deloitte predicts global FWA net additions will riseby 20%annually in 2025 and 2

63、026(in line with our 2022 Prediction on FWA),driven by growth in newenterprise FWA and markets that are not as large as the United States or India individually but still add upto millions in new subscriptions annually.5G standalone appears to be at a standstill:Will 6G run late?Telecoms reassess inv

64、estments in 5G standalone and delay 6G progress amid ROI concernsThe deployment of 5G standalone networks is progressing more slowly than expected.Telecom companiesmay be hesitant to invest heavily in this next generation of 5G in part due to underwhelming returns ontheir existing 5G investments,mak

65、ing the rollout of 6G seem further away than ever.In 2022,DeloitteGlobal predicted that the number of telcos investing in 5G SA networks would double from more than 100operators in 2022 to at least 200 by the end of 2023,but that has not happened:as of March 2024,only 49operators(out of 585 who have

66、 launched 5G globally)have deployed,launched,or soft-launched 5G SAnetworks.4In 2025,Deloitte predicts that fewer than 20 additional networks will be upgraded tostandalone,keeping 5G SA at around 12%of all 5G deployments.Open RAN mobile networks and vendor choice:Single vendornow,multivendor when?Op

67、en RANs journey toward a diverse,multivendor ecosystem is marked by slow growth and complexchallengesOpen Radio Access Network(Open RAN)aims to democratize networks by providing mobile networkoperators(MNOs)who build RANs with greater choice and more?exibility.In 2021,Deloitte predictedthat global a

68、ctive Open RAN deployment would double from 35 to 70.We were too optimistic:as ofMarch 2024,the ongoing Open RAN deployments and trials stand at 45,with only two networks globallybeing multivendor Open RAN.5The transition towards a diverse,multivendor ecosystem is proving slowerand more complex than

69、 initially anticipated,and realizing true multivendor Open RAN may take a while.Deloitte predicts that no additional multivendor Open RAN networks will be deployed or announced in2025.Despite quantums slow start,dont be slow to start yourdefense against itQuantum drug discovery and?nancial modeling

70、are likely several years away,but the time needed toupgrade cyber defenses for the quantum age likely necessitates prompt actionAs Deloitte predicted in past reports,quantum computers remain works in progress,with few real-worlduse cases where they o?er a computing advantage,at least for now.But the

71、 threat of“harvest now,decryptlater”attacks,where threat actors gather encrypted data,store it for years,and then unlock itwithcryptographically relevant future quantum computers at some point has reached a tipping point.Deloitte predicts that the number of companies,and the dollars spent,working on

72、 implementing postquantum cryptography standards will quadruple in 2025 compared with 2023.Post-quantumcryptographic solutions are expected to span the gamut in 2025,from enterprise and hyperscalers toconsumer smartphones and messaging services.RISC-V:Closing the geopolitical gen AI loopholeAn open-

73、source alternative to proprietary chip design is gaining popularity among CPU designers.Itspotential role in gen AI for markets under export restriction is a new twist for a new technology.As predicted in Deloittes2022 Global TMT Predictionsreport,open-source RISC-V(pronounced risk?ve)CPU chips reve

74、nues are expected to be close to US$1 billion by 2024.And RISC-V based SoC shipmentscould be close to two billion units.But in a new development,there are discussions in the United Statesabout restricting RISC-V chips exports,as it involves potential national security risks,given how China is“making

75、 signi?cant investments in RISC-V chip design architecture”to undermine US export controls and“leapfrog”US technological leadership in chip design.Gen AI servers need both GPUs(already undervarious export restrictions)and CPUs to control data?ows.Closed-source CPU architectures are alreadyunder US e

76、xport restrictions,and therefore,restricting RISC-V based designs may close an apparentalternative.Rising trendsKeep your eye on these newly emerging trends.We predict they could soon become the center of attention,transforming the conversation and shaping the future of the industry:Generative AI an

77、d cyber:Big risks,but big opportunities tooRecognizing generative AIs potential for enabling both threats and cyber solutions,cybersecurityprofessionals are exploring ways to harness its power to counter emerging risks and help fortify thetechnology environmentFrom the 2024 Deloitte-NASCIO Cybersecu

78、rity Study,nearly three quarters of security experts surveyedsaid the cyber threat from AI was high.Gen AI-based cyberattacks look like they will have doubled ortripled in 2024,and Deloitte predicts they will grow again in 2025,used by threat actors writing maliciousphishing emails,deepfakes,or soft

79、ware code for malware attacks.Tech companies who make gen AI toolswill likely develop guardrails to prevent malicious use in 2025.While gen AI tools can be used by threatactors for malicious purposes,the same tools can also be used by defenders to help improve securityprocesses,monitoring,and risk m

80、anagement.Silicon building blocks:Chiplets could move Moores LawforwardChiplets promise to deliver more?exible,scalable,and e?cient systems for AI and high-performancecomputing environments,at higher yieldsChipletsa heterogeneous technological architecture to develop and package semiconductorsenable

81、 high-speed data transfers,reduce latency,and help optimize PPA(power,performance,and area).Deloittepredicts worldwide advanced packaging revenue based on chiplets will more than double from an estimatedUS$7 billion in 2021 to reach US$16 billion in 2025.Chiplets are already used and explored in som

82、e of thefast-growing markets such as AI accelerators(especially generative AI),high performance computing,andtelecommunications applications.Theyre enabling the semiconductor industry to continue increasingperformance and yield.B/OSS:Telcos modernize their business and operational supportsystems sof

83、twareTelcos back-end business and operations software market is growing slowly but modernizing itbyadopting SaaS and microservices architecture,moving to the cloud and moreis a hot spot of growth forsoftware vendors and an opportunity for telcos to do more with 5G,?ber,and AIHistorically,telcos main

84、tained separate IT systems:business support systems(BSS)for customer orders,customer relationship management(CRM),and billing,and operational support systems(OSS)for ordermanagement,network inventory,and operations,often custom-built and hardware de?ned.These systemswere typically on-premises and co

85、mposed of vertical,siloed solutions.However,by 2025,many telcos areexpected to modernize and integrate these systems,driven by evolving customer expectations and newdigital revenue streams.Deloitte predicts that the global B/OSS market will reach$70 billion by 2025,growing at 5%annually.Cloud-based

86、solutions and software-as-a-service o?erings are expected to growsigni?cantly faster,at 22%and 18%annually.Most of the growth in the next few years is expected tocome from the Americas,Middle East,North Africa,and emerging Asia-Paci?c regions.Silicon photonics:Gen AI communicates at lightspeedPropel

87、led by the demanding requirements of gen AI,optical devices on silicon are stepping out of researchlabs and into the limelight of data centersDeloitte predicts that sales of silicon photonics chips used as optical transceivers will grow at acompounded annual growth rate of 25%from 2023 to 2025 to re

88、ach US$1.25 billion in 2025.These chipsallow gen AI data centers to communicate at lightspeed,use components that are smaller and cheaper,consume less energy,and produce less heat than the traditional alternatives.In 2025,the main driver ofsilicon photonics adoption is expected to be in data center

89、applications,speci?cally for those running genAI training and inferenceespecially where data needs to travel anywhere from 10 cm to 10 metersbetween chips,trays,and racks.ByAriane BucailleFranceKevin Westco?United StatesGillian CrossanUnited StatesLara AbrashUnited StatesEndnotes1.Gartner,“Gartner f

90、orecasts worldwide public cloud end-user spending to surpass$675 billion in 2024,”press release,May 20,2024.2.Angle City FC,“Willow Bay and Bob Iger to become Angel Citys new controlling owners,”July 17,2024.3.Josh Sim,“Las Vegas Aces valued at US$140m as average WNBA team hits US$96m,”SportsPro,Jun

91、e18,2024.4.GSA,“5G-GSA Market Snapshot March-2024,”March 4,2024.5.TeckNexus,“Current State of Open RAN Countries&Operators deploying&trialing Open RAN,”March 10,2024.AcknowledgementsWe wish to thank Duncan Stewart,Je?Loucks,and Paul Lee,plus the entire team,for their work on the Predictionsreport.Co

92、ver image by:Jaime Austin;Ge?y Images,Adobe StockWomen and generative AI:Theadoption gap is closing fast,buta trust gap persistsFor women to reap the full rewards of gen AI,tech companies should work to increase trust,reduce bias,and strive for more representativeworkforcesARTICLE9 MINUTE READDeloit

93、te predicts that the experimentation with and use of generative AI by women will equal or exceedthat of men in the United States by the end of 2025.1 Although womens use of gen AI was half that ofmens in 2023,their pace of adoption suggests theyre likely to reach parity within the next year.2 While

94、thisparity prediction is for the United States,the gen AI gender gap is a global phenomenon:In Europeancountries,where the use of gen AI has been surveyed,our analysis not only identi?ed signi?cant genderadoption di?erences but also revealed that women are making up ground rapidly.3 These countries

95、willlikely close the adoption gender gap within the next two yearsand the global challenges and opportunitiesfor adoption will likely mirror the US?ndings.Despite accelerating their gen AI adoption,women express less trust than men that gen AI providers willkeep their data secure.4 This“technology t

96、rust gap”could inhibit womens regular use of the technologyand full participation in new gen AI applications,as well as slow down their future purchasing of gen AIproducts and services.To help overcome this trust gap,tech companies should enhance their data security,implement clearer data management

97、 practices,and provide greater data control.AI model bias can also have a negative impact on trust.5 Women constitute less than one-third of the AIworkforce,6 and most AI workers feel that AI will produce biased results as long as their?eld continues tobe male dominated.7 Increasing womens presence

98、in the?eld can help reduce gender bias in AI,as well asgive women a greater role in steering the future of the technology.The gen AI adoption gap is closing rapidlyRecent Deloitte research has highlighted a gender gap in generative AI adoption across various geographies.For the past two years,the De

99、loitte Connected Consumer Survey has investigated the adoption of gen AI byUS consumers as part of its research into digital life.8 Our analysis revealed that women in the United Stateshave been lagging in taking up this emerging technology(?gure 1):In 2023,womens adoption of gen AIwas roughly half

100、that of men(11%of women reported experimenting with gen AI or using it for projectsand tasks beyond experimentation,vs.20%of men).In 2024,the same survey revealed that gen AIadoption overall had more than doubled,but the gender gap remained:Thirty-three percent of womensurveyed reported using or exp

101、erimenting with gen AI,vs.44%of men.The gen AI gender gap has been noted in other geographies too:Deloitte UKs 2024 Digital ConsumerTrends survey of UK consumers reported that 28%of women were using gen AI,vs.43%of men.9 Analysisof this study,as well as Deloitte UKs European study on gen AI and trus

102、t,revealed double-digit di?erencesbetween womens and mens adoption of gen AI in 12 additional European countries.10In the United States,women are rapidly closing the adoption gap.In the past year,the proportion of USwomen surveyed who have adopted gen AI tripledoutpacing the 2.2x rate of growth for

103、men.11 Analysisof current adoption levels and these rates of growth allows Deloitte to predict that the proportion ofwomen experimenting with and using gen AI for projects and tasks will match or surpass that of men in theUnited States by the end of 2025.12Full engagement may be harder to achieveWhi

104、le the trend is encouraging,reaching adoption parity wont automatically ensure that women willincorporate gen AI into their everyday work?ows.Indeed,among gen AI users surveyed in Deloittes 2024Connected Consumer Survey,34%of women say they use the technology at least once a day,vs.43%ofmen.13 And a

105、mong gen AI users who reported using it for professional tasks,41%of women currently feelthat gen AI substantially boosts their productivity,vs.61%of men.14 Tech companies and otherorganizations looking to bene?t from using gen AI should heed these di?erences and take active steps toimprove womens e

106、ngagement.The contrasts between genders may stem partly from a striking di?erence in perspective on trust.15 Aswomen progress from familiarity with gen AI into experimentation and use,negative emotions ofuncertainty,anxiety,fear,and confusion diminish,while positive feelings of fascination,excitemen

107、t,surprise,and trust grow(?gure 2).16 However,at both the experimentation and project and task use levels,womens feelings of trust toward the technology are signi?cantly lower than mens,and their feelings ofuncertainty remain higher.Indeed,only 18%of women surveyed who are experimenting with or usin

108、ggenerative AI indicated having“high”or“very high”trust that the providers of the gen AI capabilities theyuse will keep their data securewhereas,for male adopters,that number has reached 31%.17The trust gap is not unique to gen AI,but extends to broader tech services and interactions:While 54%ofwome

109、n surveyed in Deloittes 2024 Connected Consumer Survey agree that the bene?ts they get from onlineservices outweigh their data privacy concerns(an improvement from 46%in 2023),more men agree(62%).18 Last year,we reported that women are more wary than men about how their personal data is usedand prot

110、ected and that it was a?ecting their willingness to share data,particularly when it comes tosensitive health and?tness metrics.19 Women may perceive the potential consequences of a security breachor data misuse as more signi?cant.20The growing popularity of generative AI may exacerbate these longsta

111、nding issues around data privacy andtech.21 As users interact with gen AI,the systems may feed users data back into their AI modelsandexperts say its not necessarily clear or easy to opt out of having ones data used for AI training.22 Asconsumers start to converse with gen AI for advice on sensitive

112、,personal topics,the data privacy andsecurity stakes grow.Indeed,the trust gap around data privacy and security may underpin the di?erenceswere seeing between womens and mens levels of interest in having a variety of gen AI experiences in thefuture(?gure 3).23 Surveyed women are somewhat less intere

113、sted than men in interacting with gen AI onless-sensitive topics(such as travel,shopping,?tness,and nutrition),but they are substantially lessinterested than men in engaging with gen AI around more sensitive topics(such as personal?nances andrelationships,and medical or mental health issues).The tru

114、st gap may also contribute to less excitement among women to purchase new gen AI technologies.Tech companies are beginning to sell laptops,tablets,and smartphones with embedded AI chips designed toimprove functionality(for example,summarizing information in real time,generating photos and videos,and

115、 instantly translating foreign languages).24When Deloittes 2024 Connected Consumer Survey askedwhether new AI functionality will have any e?ect on their plans to upgrade devices,fewer women saidtheyre likely to upgrade their devices sooner compared to men.25 For example,while 43%of men withsmartphon

116、es said embedded AI would make them very or somewhat likely to upgrade their phone soonerthan planned,only 32%of women said the same(conversely,58%of women said it would have no e?ecton their upgrade plans,vs.50%of men).And when it comes to laptops,41%of men said on-device AIwould make them very or

117、somewhat likely to upgrade those devices sooner,vs.28%of women.Withwomen controlling or in?uencing an estimated 85%of consumer spending,their lower enthusiasm forupgrading to devices with AI could pose an issue for tech providers.26The trust gap is not the only factor holding women back from maximiz

118、ing their use of gen AI.Women genAI users surveyed are less likely to feel that their company actively encourages their use of the technology atwork(61%of women users feel this way,vs.83%of men).27 And while 49%of women gen AI users saythat their company invests in training employees on how to use g

119、enerative AI,that falls short of the 79%ofmen reporting the same.Whether these numbers re?ect di?erences in perception or actual experiences withaccess to training programs and encouragement in the workplace,companies should pay heed and work toclose the gaps.Women in tech are forging ahead with gen

120、 AIbut betterrepresentation is neededIn the tech industry,there is a di?erent story about gen AI adoption entirelyand women working in techmay hold clues for fostering greater gen AI engagement by women overall in the future.Not surprisingly,the industry creating AI products and services has higher

121、levels of gen AI adoption among its employees:InDeloittes 2024 Connected Consumer Survey,70%of women and 78%of men working in the tech industryreported experimenting with gen AI or using it for projects or tasksfar outpacing nontech women(32%)and men(40%).28 What may be more surprising is that women

122、 working in the tech industry appear to bemoving beyond gen AI experimentation and into using it for projects and tasks faster than their malecounterparts(44%vs.33%).And both groups are anticipating greater bene?ts:About 7 in 10 women andmen working in tech expect their use of gen AI to“substantiall

123、y boost”their productivity at work a yearfrom now.29Whats more,theres no notable trust gap between tech women and men.Both groups have greater trust ingenerative AI than adopters overall:More than 40%of tech women and men using or experimenting withgen AI reported having“high”or“very high”trust that

124、 gen AI providers will keep their data secure.30 Inboth groups,75%of those surveyed agree that the bene?ts they get from online services outweigh theirprivacy concernsvs.just 54%of women and 60%of men working outside tech.31 Its likely that womenin the tech industry have a better understanding of ho

125、w gen AI works than nontech workers,and that theirheavier professional use of gen AI has increased their comfort level and shown them how they can bene?tfrom the technology.Moreover,most tech women who use gen AI reported that their companies encourageits use(84%)and provide training(72%)in contrast

126、,among women using gen AI in other industries,farfewer reported that their companies encourage its use(55%)or provide training(45%).32Despite the greater adoption of AI by women in the tech industry,theres a relative lack of women workingin AI roles.Women only make up about 30%of the AI-related work

127、force,which is comparable to theirrepresentation in STEM?elds overall.33 This underrepresentation of women in AI could have seriousimplications for the development and deployment of AI systems across various domains and sectors.One of the major challenges posed by the relative lack of women in the A

128、I workforce is the risk ofperpetuating gender bias against women in AI applications.34 As many as 44%of AI systems acrossindustries exhibit gender bias,which can negatively a?ect outputs from AI systems in ways that continue tomarginalize and underrepresent women.35 For instance,gender bias in AI ca

129、n lead to bias in hiringpractices,36 lower quality health care,37 and reduced access to?nancial services for women.38 And Deloitteresearch has shown that bias in AI models can erode employee and customer trust.39 Bringing more womeninto AI jobs can be crucial for achieving gender equality and ensuri

130、ng that AI bene?ts society.40Bottom lineThere are several reasons why tech companies should work toward increasing womens engagement withgen AI.First,with women controlling or in?uencing most consumer purchasing,failing to get women onboard with frequent gen AI use could increase the risk that AI pr

131、oducts and services wont achieve theirexpected potential.Second,if women dont engage with gen AI tools as fully as male employees,companiescould risk not achieving the productivity gains they might expect to see after investing in gen AI.And,because gen AI depends upon collecting and building upon i

132、nteraction data,the underrepresentation ofwomens interactions could exacerbate biases in AI models.41 Finally,if women dont participate inemerging gen AI use cases as fully as they could,that may keep them from maximizing future tech bene?ts(for example,advantages of chatbot interventions in medical

133、 or mental health)and deepen existinginequities.42To help bolster womens trust in gen AI,tech companies should work to address the potential risksassociated with the technology.Deloittes 2024 Connected Consumer Survey found that earning trust maydepend at least partially on improving the transparenc

134、y of tech companies data privacy and securitypolicies,as well as making it easier for consumers to control their personal data.43 Tech companies shouldconsider prioritizing robust data security measures and communicating their data-handling practices moree?ectively.Making it simpler for consumers to

135、 understand what data gets collected and how its used,alongwith providing easier ways to control that use(such as prompting users at appropriate points to makeinformed choices about the use of their data)may not only build trust but could also confer a competitiveadvantage.But its not just tech comp

136、anies that should pay heed to potential gen AI risks:Eighty-fourpercent of survey respondents believe that governments should do more to regulate the way companiescollect and use consumer data.44Across industries,companies that want to achieve full use of gen AI by men and women workers shouldtake c

137、are to encourage the use of gen AI capabilities.Beyond various popular professional use casesdocument editing,web searches,summarizing materials,and research assistancecompanies can embraceindustry-speci?c ways to use generative AI.45 Maximizing the use of gen AI by employees may requireestablishing

138、 training programs.Striving for full consumer engagement in generative AI is a commendable objective,but it may be moredi?cult to achieve without equitable representation among the people who develop generative AItechnologies.To increase the diversity and inclusion of women in AI roles,companies sho

139、uld considerfocusing on creating workplaces that meet the needs of those they employ.For example,a study of womenin AI noted that work/life balance is the most important factor for their job satisfaction,which includeselements such as having a?exible working schedule or being able to work remotely.4

140、6 Women also reportedlooking for jobs with women in leadership,transparency around pay and promotions,and zero-tolerancepolicies for harassment and abuse.47 Attracting more women to the?eld may also involve providing moreeducation and training opportunities for women to learn AI skills and competenc

141、ies.It could also bebene?cial to create more mentorship and networking programs that allow women in AI to share theirexperiences and support one another,and to provide funding for more women to participate in AI researchand innovation projects.As womens role in developing gen AI grows,its likely tha

142、t there will beapplications and systems that engage all women more.BySusanne HupferUnited StatesBree MathesonUnited StatesGillian CrossanUnited StatesAriane BucailleFranceJe?LoucksUnited StatesEndnotes1.To understand consumer attitudes toward digital life,the Deloitte Center for Technology,Media&Tel

143、ecommunications surveyed 3,857 US consumers in the second quarter of 2024 and 2,018 USconsumers in the second quarter of 2023.These 2024 and 2023 Connected Consumer Surveys collecteddata on consumers reported adoption of generative AI,including experimentation and use for projectsand tasks(beyond ex

144、perimentation).By analyzing longitudinal adoption data and calculating the rate ofchange in adoption from 2023 to 2024 for men and women,we are able to project that women willclose the adoption gap by the end of 2025;see:Jana Arbanas et al.,Earning trust as gen AI takes hold:2024 Connected Consumer

145、Survey,5th edition,Deloitte,December 3,2024;Jana Arbanas,Paul H.Silverglate,Susanne Hupfer,Je?Loucks,Prashant Raman,and Michael Steinhart,“Balancing act:Seeking just the right amount of digital for a happy,healthy connected life,”Deloitte Insights,Sept.5,2023.2.Ibid.3.Our analysis was conducted from

146、 August to October 2024,based on data from Deloitte UKs 2023 and2024 Digital Consumer Trends surveys,as well as a 2024 Deloitte UK survey of European consumers onthe topic of generative AI;see:Paul Lee and Ben Stanton,“Generative AI:7 million workers andcounting,”Deloitte,June 25,2024;Jonas Malmlund

147、,Frederik Behnk,and Joachim Gullaksen,“Generative AI is all the rage,”Deloitte,2023;Roxana Corduneanu,Stacey Winters,Jan Michalski,Richard Horton,and Ram Krishna Sahu,“Europeans are optimistic about generative AI but there ismore to do to close the trust gap,”Deloitte Insights,Oct.10,2024.4.Analysis

148、 based on Deloittes 2024 Connected Consumer Survey;see:Arbanas et al.,Earning trust as genAI takes hold:2024 Connected Consumer Survey.5.Don Fancher,Beena Ammanath,Jonathan Holdowsky,and Natasha Buckley,“AI model bias candamage trust more than you may know.But it doesnt have to.”Deloitte Insights,De

149、c.8,2021.6.World Economic Forum,“Global gender gap report 2023,”June 2023.7.Deloitte AI Institute,“Women in AI,”accessed November 2024.8.Jana Arbanas et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey,5th edition,Deloitte,publishing December 3,2024;Arbanas,Silverglate,Hupfer,Lo

150、ucks,Raman,and Steinhart,“Balancing act.”9.Deloitte,“Generative AI:7 million workers and counting,”accessed November 2024.10.The Digital Consumer Trends study conducted in various countries in 2024 revealed gen AI adoptiongaps of 17 points in Denmark;12 points in Sweden,Italy,and the Netherlands;11

151、points in Belgium;and 10 points in Norway.Additional analysis of a Deloitte European gen AI study revealed gen AIadoption gaps ranging from 10 to 15 points in 11 European countries studied(Belgium,France,Germany,Ireland,Italy,the Netherlands,Poland,Spain,Sweden,Switzerland,and the United Kingdom);se

152、e:Deloitte,“Generative AI”;Deloitte,“Generative AI is all the rage,”accessed November 2024;Corduneanu,Winters,Michalski,Horton,and Sahu,“Europeans are optimistic about generative AI butthere is more to do to close the trust gap.”11.Analysis based on 2024 and 2023 Deloitte Connected Consumer Surveys;

153、see:Arbanas et al.,Earningtrust as gen AI takes hold:2024 Connected Consumer Survey;Arbanas,Silverglate,Hupfer,Loucks,Raman,and Steinhart,“Balancing act.”Deloitte,“Generative AI.”12.Ibid.13.Arbanas,et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.14.Ibid.15.Ibid.16.Ibid.17.Ib

154、id.18.Ibid.19.For example,only 43%of women we surveyed in the Deloitte 2023 Connected Consumer Survey whoowned smart watches or?tness trackers said that they share the data collected by those devices withtheir health care provider,vs.57%of men;see:Susanne Hupfer,Jennifer Radin,Paul H.Silverglate,and

155、Michael Steinhart,“Tech companies have a trust gap to overcomeespecially with women,”DeloitteInsights,Nov.8,2023.20.These fears may be warranted.Consider that most health appsalong with the data they gather andtransmitare not covered by the Health Insurance Portability and Accountability Act,which m

156、eans thedata may be shared or sold to third parties;see:Steve Alder,“Majority of Americans mistakenly believehealth app data is covered by HIPAA,”The HIPAA Journal,July 26,2023.21.Ina Fried,“Generative AIs privacy problem,”Axios,March 14,2024;Federal Trade Commission,“AIcompanies:Uphold your privacy

157、 and con?dentiality commitments,Jan.9,2024.22.Ibid;Matt Burgess and Reece Rogers,“How to stop your data from being used to train AI,”Wired,April 10,2024.23.Arbanas,et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.24.Baris Sarer,Ricky Franks,Cheryl Ho,and Jake McCarty,“AI and

158、the evolving consumer deviceecosystem,”The Wall Street Journal,April 24,2014;Sam Reynolds,“AI-enabled PCs will drive PC salesgrowth in 2024,say research?rms,”Computer World,Jan.11,2024;Clare Conley,“Generative AI in2024:The 6 most important consumer tech trends,”Qualcomm,Dec.14,2023.25.Arbanas,et al

159、.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.26.Monique Woodard,“Unlocking the trillion-dollar female economy,”TechCrunch,May 21,2023.27.Arbanas,et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.28.Ibid.29.Across industries,51%of women workers using gen

160、AI anticipate it would substantially boost theirproductivity at work a year from now,vs.64%of men;see:Arbanas,et al.,Earning trust as gen AI takeshold:2024 Connected Consumer Survey.30.Tech women and men are statistically tied:Forty-two percent of tech women who use or experimentwith gen AI have“hig

161、h”or“very high”trust that gen AI providers will keep their data secure,andanother 40%report moderate trust,while 47%of tech men report“high”or“very high”trust andanother 30%report moderate trust;see:Arbanas,et al.,Earning trust as gen AI takes hold:2024Connected Consumer Survey.31.Ibid.32.Greater pr

162、oportions of men in the tech industry who use gen AI report that their employers encourage itsuse(93%)and provide training(91%).While theres still a gender gap in these views among workers inthe tech industry,the gap is signi?cantly smaller than among men and women working in otherindustries;see:Arb

163、anas,et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.33.World Economic Forum,“Global gender gap report 2023.”34.Deloitte,“Generative AI.”35.Genevieve Smith and Ishita Rustagi,“When good algorithms go sexist:Why and how to advance AIgender equity,”Stanford Social Innovation R

164、eview,March 31,2021.36.Charlotte Lytton,“AI hiring tools may be?ltering out the best job applicants,”BBC,Feb.16,2024.37.Carmen Niethammer,“AI bias could put womens lives at risk-A challenge for regulators,”Forbes,March 2,2020.38.Ryan Browne and MacKenzie Sigalos,“A.I.has a discrimination problem.In

165、banking,the consequencescan be severe,”CNBC,June 23,2023.39.Fancher,Ammanath,Holdowsky,and Buckley,“AI model bias can damage trust more than you mayknow.But it doesnt have to.”40.World Economic Forum,“Global gender gap report 2023.”41.Smith and Rustagi,“When good algorithms go sexist.”42.Hyun-Kyoung

166、 Kim,“The e?ects of arti?cial intelligence chatbots on womens health:A systematicreview and meta-analysis,”Healthcare,Feb.23,2024;Sheryl Jacobson and Jen Radin,“Can FemTechhelp bridge a gender-equity gap in health care?”Deloitte,Oct.5,2023;Karen Taylor,“Why investing inFemTech will guarantee a healt

167、hier future for all women,”Deloitte UK,June 23,2023.43.Arbanas,et al.,Earning trust as gen AI takes hold:2024 Connected Consumer Survey.44.Ibid.45.Deloitte AI Institute,“The generative AI dossier:A selection of high-impact use cases across six majorindustries,”April 3,2023.46.Women in AI,“WAI at wor

168、k:Shaping the future of work for women in AI,”2022.47.Ibid.AcknowledgementsAuthors would like to thank Duncan Stewart,Paul Lee,Ben Stanton,Vipul Mehta,Roxana Corduneanu,MichaelSteinhart,Michelle Dollinger,Je?Stoudt,Catherine King,Elizabeth Fisher,Andy Bayiates,Prodyut Borah,MollyPiersol,Deloi?e Insi

169、ghts team.Cover image by:Jaime Austin;Ge?y Images,Adobe StockAs generative AI asks for morepower,data centers seek morereliable,cleaner energysolutionsThe tech industry should optimizeinfrastructure,rethink chip design,andcollaborate with electricity providers to helpsecure a more sustainable future

170、 for data centersARTICLE14 MINUTE READAI-driven data center power consumption will continue to surge,but data centers are notin factthat biga part of global energy demand.Deloitte predicts data centers will only make up about 2%of globalelectricity consumption,or 536 terawatt-hours(TWh),in 2025.But

171、as power-intensive generative AI(genAI)training and inference continues to grow faster than other uses and applications,global data centerelectricity consumption could roughly double to 1,065 TWh by 2030(?gure 1).1 To power those datacenters and reduce the environmental impact,many companies are loo

172、king to use a combination ofinnovative and energy-e?cient data center technologies and more carbon-free energy sources.Nonetheless,its an uphill task for power generation and grid infrastructure to keep pace with a surge inelectricity demand from AI data centers.Electricity demand was already growin

173、g fast due to electri?cationthe switch from fossil-fueled to electric-powered equipment and systems in the transport,building,andindustrial segmentsand other factors.But gen AI is an additional,and perhaps,an unanticipated source ofdemand.Moreover,data centers often have special requirements as they

174、 need 24/7 power supply with highlevels of redundancy and reliability,and theyre working to have it be carbon-free.Estimating global data centers electricity consumption in 2030 and beyond is challenging,as there aremany variables to consider.Our assessment suggests that continuous improvements in A

175、I and data centerprocessing e?ciency could yield an energy consumption level of approximately 1,000 TWh by 2030.However,if those anticipated improvements do not materialize in the coming years,the energyconsumption associated with data centers could likely rise above 1,300 TWh,directly impacting ele

176、ctricityproviders and challenging climate-neutrality ambitions.2 Consequently,driving forward innovations in AIand optimizing data center e?ciency over the next decade will be pivotal in shaping a sustainable energylandscape.Some parts of the world are already facing issues in generating power and m

177、anaging grid capacity in theface of growing electricity demand from AI data centers.3 Critical power to support data centers mostimportant componentsincluding graphics processing unit(GPU)and central processing unit(CPU)servers,storage systems,cooling,and networking switchesis expected to nearly dou

178、ble between 2023 and2026 to reach 96 gigawatts(GW)globally by 2026;and AI operations alone could potentially consumeover 40%of that power.4 Worldwide,AI data centers annual power consumption is expected to reach 90terawatt-hours by 2026(or roughly one-seventh of the predicted 681 TWh of all data cen

179、ters globally),roughly a tenfold increase from 2022 levels.5 As such,gen AI investments are fueling demand for so muchelectricity that in the?rst quarter of 2024,global net additional power demand from AI data centers wasroughly 2 GW,an increase of 25%from the fourth quarter of 2023 and more than th

180、ree times the levelfrom the?rst quarter of 2023.6 Meeting data center power demand can be challenging because data centerfacilities are often geographically concentrated(especially in the United States)and their need for 24/7power can burden existing power infrastructure.7Deloitte predicts that both

181、 the technology and electric power industries can and will jointly address thesechallenges and contain the energy impact of AImore speci?cally,gen AI.Already,many big tech andcloud providers are investing in carbon-free energy sources and pushing for net-zero targets,8 demonstratingtheir commitment

182、to sustainability.Hyperscalers plan massive expansion of gen AI datacenters to help support growing customer demandThe surge in electricity demand is largely due to hyperscalers plans to build out data center capacity,globally.9 As AI demandspeci?cally gen AIis expected to grow,companies and countri

183、es are racing tobuild more data centers to meet that demand.Governments are also establishing sovereign AI capabilities tomaintain tech leadership.10 The data center real estate build-out has reached record levels based on selectmajor hyperscalers capital expenditure,which is trending at roughly US$

184、200 billion as of 2024,andestimated to exceed US$220B by 2025.11Moreover,Deloittes State of Generative AI in the Enterprise survey noted that enterprises have been mostlypiloting and experimenting until now.12 But as they experiment with getting value from gen AI,respondentsare seeing tangible resul

185、ts and so intend to quickly scale up beyond pilots and proofs of concept.If usagegrows as the technology matures,hyperscalers and cloud providers capital expenditure will most likelyremain high through 2025 and 2026.Two broad areas drive most of the electricity consumption in a data center:computing

186、 power and serverresources like server systems(roughly 40%data center power consumption)and cooling systems(consume38%to 40%power).These two are the most energy-intensive components even in AI data centers,andthey will continue to fuel data centers power consumption.Internal power conditioning syste

187、ms consumeanother 8%to 10%,while network and communications equipment and storage systems use about 5%each,and lighting facilities usually use 1%to 2%of power(?gure 2).13 With gen AI requiring massiveamounts of power,data center providersincluding the hyperscalers and data center operatorsneed toloo

188、k at alternate energy sources,new forms of cooling,and more energy-e?cient solutions when designingdata centers.Several e?orts are already underway.Gen AI is contributing to increased electricity demandData centers energy consumption has been surging since 2023,thanks to exploding demand for AI.14De

189、ploying advanced AI systems requires vast numbers of chips and processing capacity,and trainingcomplex gen AI models can require thousands of GPUs.Hyperscalers and large-scale data center operators that are supporting gen AI and high-performancecomputing environments require high-density infrastruct

190、ure to support computing power.Historically,datacenters relied mainly on CPUs,which ran at roughly 150 watts to 200 watts per chip.15 GPUs for AI ran at400 watts until 2022,while 2023 state-of-the-art GPUs for gen AI run at 700 watts,and 2024 next-generation chips are expected to run at 1,200 watts.

191、16 These chips(about eight of them)sit on blades placedinside of racks(10 blades per rack)in data centers,and are using more power and producing more heat persquare meter of footprint than traditional data center designs from only a few years ago.17 As of early 2024,data centers typically supported

192、rack power requirements of 20 kW or higher.But the average powerdensity is anticipated to increase from 36 kW per server rack in 2023 to 50 kW per rack by 2027.18Total AI computing capacity,measured in?oating-point operations per second(FLOPS),has also beenincreasing exponentially since the advent o

193、f gen AI.Its grown 50%to 60%quarter over quarter globallysince the?rst quarter of 2023 and will likely grow at that pace through the?rst quarter of 2025.19 But datacenters dont only measure capacity in FLOPS,they also measure megawatt hours(MWh)and TWh.Gen AIs multibillion parameter LLMs and the mul

194、tibillionwatts they consumeGen AI large language models(LLMs)are becoming more sophisticated,incorporating more parameters(variables that enable AI learning and prediction)over time.From the 100 to 200 billion parameter modelsthat were released initially during 2021 to 2022,current advanced LLMs(as

195、of mid-2024)have scaled upto nearly two trillion parameters,which can interpret and decode complex images.20 And therescompetition to release LLMs with 10 trillion parameters.More parameters add to data processing andcomputing power needs,as the AI model must be trained and deployed.This can further

196、 accelerate demandfor gen AI processors and accelerators,and in turn,electricity consumption.Moreover,training LLMs is energy intensive.Independent research of select LLMs that were trained onmore than 175 billion parameters of data sets demonstrated that they consumed anywhere between 324MWh and 1,

197、287 MWh of electricity for each training run and models are often retrained.21On average,a gen AIbased prompt request consumes 10 to 100 times more electricity than a typicalinternet search query.22 Deloitte predicts that if only 5%of daily internet searches,globally,use gen AIbased prompt requests,

198、it would require approximately 20,000 servers(with eight specialized GPU cores ineach of the servers)that consume 6.5 kW on an average per server to ful?ll the prompt requests,amountingto an average daily electricity consumption of 3.12 GWh and annual consumption of 1.14 TWh23whichis equivalent to a

199、nnual electricity consumed by approximately 108,450 US households.24Data center demand could present challenges andopportunities for power sector transitionThe electric power sector was already planning for rising demand:Many in the industry predicted as muchas a tripling of electricity consumption

200、by 2050 in some countries.25But that trajectory has recentlyaccelerated in some areas due to burgeoning data center demand.Previous forecasts in many countries haveprojected rising power demand due to electri?cation as well as increasing data center consumption andoverall economic growth.But recent

201、sharp spikes in data center demand,which could be just the tip of theiceberg,reveal the growing magnitude of the challenge.26Round-the-clock,carbon-free electricity that many tech companies seek can be hard to come by,especiallyin the short term.This comes against the backdrop of a multi-decade powe

202、r industry transformation,as electric companiesbuild,upgrade,and decarbonize electric grid infrastructure,and digitalize systems and assets.In manyareas,electric companies are also hardening assets against increasingly severe weather and climate eventsand protecting networks from rising cybersecurit

203、y threats.27 Electric power grids in some countries arestruggling to keep up with demand,especially for low-or zero-carbon electricity.In the United States,datacenters are anticipated to represent 6%(or 260 TWh)of total electricity consumption in 2026.28 TheUnited Kingdom may witness a sixfold growt

204、h in electricity demand within a period of just 10 years,largelydue to AI.29 In China,data centersincluding the ones that power AIwill likely make up 6%of thecountrys total electricity demand by 2026.30 Data centers could also add to Chinas pollution problem,since the countrys power generation is do

205、minated by coal,which accounted for 61%of its energy use and79%of its carbon dioxide emissions in 2021.31Some countries that are facing the rising demand for electricity from data centers are responding withregulations.For instance,in Ireland,existing data centers consume a?fth of the countrys total

206、 electricityconsumption and this is only expected to grow further as AI-driven data centers spring up more;households are even lowering their power consumption.32Temporarily,Ireland halted the construction ofnew data centers connected to the grid,but has since reversed that position.33 Like Ireland,

207、even the city ofAmsterdam halted new data center construction to support sustainable urban development.34 Singaporeannounced new sustainability standards for data centers that require operators to gradually increase theoverall operating temperatures of their facilities to 26C or higher.Higher operat

208、ing temperatures reducethe demand for cooling and lower power consumption,but at the cost of shortening the lifespan of thechips.35The urgency and geographic concentration of data center demandand the requirement for 24/7 carbon-free energycan further complicate the challenge for tech companies and

209、electricity providers,in additionto new demand from electri?cation,manufacturing,and other sources.The largest data center marketglobally is in northern Virginia,36 and the local utility,Dominion Energy,expects power demand to grow byabout 85%over the next 15 years,with data center demand quadruplin

210、g.37 The round-the-clock,carbon-free electricity that many tech companies seek can be hard to come by,especially in the short term.Electricity providers are exploring multiple avenues to help meet demand while maintaining reliability anda?ordability.In addition to new renewable energy and battery st

211、orage,many electricity providers have alsoannounced plans to build natural gas?red power plants,which are not carbon-free.38 This couldpotentially make it more challenging to meet utility,state,and even national decarbonization targets.39Despite being poised to consume massive amounts of clean energ

212、y,AI could also potentially help hasten theclean energy transition:Some utilities are already using AI to enable electric grids to operate more cheaply,e?ciently,and reliably through improved weather and load forecasting,enhanced grid management andrenewable asset performance,faster storm recovery,b

213、etter wild?re risk assessment,and more.40Data center cooling is water-intensiveNext-generation CPUs and GPUs have higher thermal density properties than their predecessors.At thesame time,some server vendors are packing more and more power-hungry chips into each rack in anendeavor to cater to the gr

214、owing demand for high-performance computing and AI applications.But denserracks will demand more water,especially to cool the gen AI chips.AI data centers freshwater demandcould be as much as 1.7 trillion gallons(at the higher end)by 2027.41 A hyperscale data center that intendsto manage excess heat

215、 with air-based cooling and evaporated drinking water would require over 50 milliongallons of water every year(or roughly what it takes to make 14,700 smartphones).42 This water cannot bereturned to the aquifer,reservoir,or water supply where it came from.43Air-based cooling alone uses up to 40%of a

216、 typical data centers electricity consumption.Therefore,datacenters are looking at alternatives to traditional air-based cooling methods,mainly into liquid cooling,as itshigher thermal transfer properties could help cool high-density server racks and enable them to reducepower usage by as much as 90

217、%when compared with air-based methods.44 As liquid cooling directlydelivers cooling to server racks,it can support dense power racks on the order of 50 kW to 100 KW ormore.45 Moreover,it may help eliminate the need for chillers,which were traditionally used for producingthe cooling water.However,des

218、pite liquid cooling technologys promise to help save energy across the data center stack,46 itsstill in its early days and is yet to be widely adopted or integrated into AI data centers,globally.47 Moreover,water is a?nite resource,and therefore its cost and availability will likely a?ect future dec

219、isions about itsusage.The tech industry is moving toward more sustainablesolutions and carbon-free sourcesTo help expedite the move toward using carbon-free sources to power AI data centers,tech industry majorscontinue to be aggressive in their pursuit of renewable energy by way of power purchase ag

220、reements,orlong-term contracts with renewable energy providers.48 These deals have helped bankroll renewable energyprojects by enabling them to secure?nancing.In some cases,technology companies are working withelectricity providers and innovators to help test and scale promising energy technologies,

221、includingadvanced geothermal,advanced wind and solar technologies,hydropower,and even underwater datacenters.In some areas,local grid constraints and long interconnection times for new renewable and battery storagefacilities are causing delays in connecting these resources to the electric grid.49 Th

222、ese delays,which can be aslong as?ve years in the United States,are often due to high demand and insu?cient transmissioninfrastructure.As a result,tech companies are increasingly pursuing onsite,sometimes o?-grid,energysolutions.50 Additionally,they are investing in new technologies such as long-dur

223、ation energy storage andsmall modular nuclear reactors to help address these challenges.In some cases,tech companies and utilitiesare planning to coordinate to bring innovative clean energy technologies to scale,which could eventuallybene?t other organizations and help decarbonize the electric grid

224、faster.51 Many of these research anddevelopment programs,pilots and other clean energy investments may take years before reapingbene?ts,demonstrating return on investment,and becoming commercially viable.52 For example,small,modular nuclear reactors are still in early development stages,and may not

225、be a near-term zero-carbonsolution.53The technology sector consistently dominates US corporate renewable procurement and accounted for morethan 68%of the nearly 200 deals of associated contracted capacity tracked over the 12 months prior toFeb.28,2024.54 Similarly,hyperscalers and data center operat

226、ors in India are increasingly using solar topower their data centers in the region.55 Without these purchase commitments,many renewable energyprojects would not be built.56As such,the tech industrys role in bankrolling clean energy technologies to help bring them to scale willcontinue to be valuable

227、.In some cases,theyre working directly with innovators and renewable energyproducers,and in other cases,theyre partnering with utilities.57 Importantly,the way tech companies injectcapital to help advance the clean energy transition is critical,as neither the innovators nor the powerindustry would t

228、ypically have the level of?nancial resources that the tech industry possesses.Bottom lineWhat should the broader tech industry,hyperscalers,data center operators,utilities,and regulators do tohelp meet gen AI demand sustainably?Several considerations for hyperscalers and the broader tech industryare

229、 more or less in line with what we presented in Deloitte Globals 2021 prediction on cloudmigration.58Though demand drivers may have changed and the pace of change has accelerated,theindustry is working to achieve balance between sustainability and expedited time to market,while keepingdata centers r

230、ising energy demands under control and?nding more sustainable ways to power AIspeci?cally gen AI.Do we need to keep racing to build bigger and bigger foundational models(for example,more than atrillion parameter models),or are smaller models su?cient,while being more sustainable?1.Make gen AI chips

231、more energy-e?cient:Already,a new generation of AI chips can perform AI trainingin 90 days,consuming 8.6 GWh.This would be less than one-tenth the energy that the previous generationof chips takes to do the same function on the same data.59 Chip companies should continue to work withthe broader semi

232、conductor ecosystem to help intensify focus on improving FLOPS/watt performance,suchthat,future chips can train AIs several times larger than the largest AI systems currently available whileusing less electricity.2.Optimize gen AI uses and shift processing to edge devices:This includes assessing whe

233、ther its energy-e?cient to do training and inference in the data center,or on the edge device,and accordingly rebalancedata center equipment needs.Edge not only can support applications where response times are critical,butalso for those use cases where sensitive data is involved,or privacy needs ar

234、e high.It also helps savenetwork and server bandwidth,rerouting gen AI workloads to local and near-location or co-locationdevices,while only transmitting select AI workloads to data centers.603.Implement changes in gen AI algorithms and rightsize AI workloads:Do we need to keep racing to buildbigger

235、 and bigger foundational models(for example,more than a trillion parameter models),or are smallermodels su?cient,while being more sustainable?Already,startups are developing on-device multimodal AImodels,which do not require energy-intensive computations in the cloud.61 Customers should?ne-tune anda

236、djust the size of their AI workloads and go for targeted gen AI models(including preexisting models andtraining only when needed)based on real business needs,which can minimize energy use.Additionally,depending on speci?c needs with AI inferencing(for example,doing inference in real time and when la

237、tencyis critical),CPUs can be more advantageous and e?cient.624.Form strategic partnerships to serve local and cluster-level AI data center needs:For several small tomidsize organizations(including universities)that may?nd it hard to tap into gen AI data center capacity,those organizations should wo

238、rk with specialized data center operators and cloud service providers thatfocus on delivering high-performance computing solutions for smaller high-performance computing GPU-cluster co-locations.63 A corollary:Data centers can then actively track usage and availability for potentialopportunities and

239、 demand pockets to help deliver near-term co-location services.5.Collaborate with other stakeholders and sectors to make an overall positive environmental impact:Thevarious ecosystem playersincluding hyperscalers,their customers,third-party data center operators andco-location service providers,elec

240、tricity providers,the local regulators and municipalities,and the realestate?rmsshould have ongoing conversations around whats feasible and viable for the business,environment,and society.64 That collaboration should encompass multiple aspects including determiningpotential strategic co-location nee

241、ds(where a data center company rents computing and server resources toone or more companies),assessing cooling needs such as adequate temperatures in liquid cooling systems,identifying solutions to manage heat and wastewater,and?guring out recycling needs.For example,inEurope,data center operators a

242、re directing waste heat to warm swimming pools in the vicinity.65 Electricityproviders should consider working more closely with the tech industry to understand how to meet datacenter energy demand,while identifying ways the tech companies could potentially help fund and scale newenergy technologies

243、,which is a vital step to bring more clean energy to the grid.The holistic e?orts of hyperscalers and electricity providers to help increase the use of carbon-free sourcesto power data centersincluding the ones being built exclusively for gen AImay bear fruit in the longerterm.ByKarthik Ramachandran

244、IndiaDuncan StewartCanadaKate HardinUnited StatesGillian CrossanUnited StatesEndnotes1.Deloitte analysis based on publicly available information sources and conversations with industryspecialists.We used base electricity consumption data from the US Energy Information Administrations(EIA)Internation

245、al Energy Outlook 2023 data on total electricity usage across residential,commercial,industrial,and transportation end uses(a reference to US Energy Information Administration,“Table:Delivered energy consumption by end-use sector and fuel,”accessed Nov.4,2024)to arrive at estimatesand prediction val

246、ues for global data centers electricity consumption(TWh)between 2022 and 2030.Our estimates and projections for data centers percent electricity consumption of global total are basedon our research of multiple publicly available sources including SemiAnalysis,EPRI,Goldman Sachs,Bloomberg,and Latitud

247、e Media,and further validated based on our conversations with subject matterspecialists in the areas of technology,energy,and sustainability.Total energy consumption by end-usesector and fuel(as noted from the aforementioned table from EIAs International Energy Outlook 2023data),globally,is estimate

248、d and forecast at 26,787 TWh in 2025,27,256 TWh in 2026,and 29,160TWh in 2030 increasing from 25,585 TWh back in 2022.2.As noted in endnote 1 above,we arrived at 2022 to 2030 data,estimates,and predictions based on acombination of in-depth secondary research of multiple publicly available sources,an

249、d validated furtherfrom our discussions with subject matter specialists.Also,see Prof.Dr.Bernhard Lorentz,Dr.JohannesTrby,and Geo?Tu?,Powering arti?cial intelligence,Deloitte Global,November 2024.3.One-?fth of Irelands electricity is consumed by data centers,and this is expected to grow,even ashouse

250、holds are lowering their electricity use.To read further,see:Chris Baraniuk,“Electricity grids creakas AI demands soar,”BBC,May 21,2024.4.Dylan Patel,Daniel Nishball,and Jeremie Eliahou Ontiveros,“AI data center energy dilemma:Race forAI data center space,”SemiAnalysis,March 13,2024.5.Ibid.6.Data ce

251、nter BMO report,Communications Infrastructure,“1Q24 data center leasing:Records are madeto be broken,”April 28,2024;Moreover,due to strong demand from cloud providers and AIworkloads,the data center primary market supply in the United States alone was up 26%year over yearto 5.2 GW in 2023,and more a

252、re under construction.See further:CBRE,“North America data centertrends H2 2023,”March 6,2024.7.Lisa Martine Jenkins and Phoebe Skok,“Mapping the data center power demand problem,in threecharts,”Latitude Media,May 31,2024.8.Based on our analysis of multiple publicly available information and reports

253、 from what companies self-report,and further validated from third-party sources.9.For context,hyperscalers are large cloud service providers and data centers that o?er huge amounts ofcomputing and storage resources typically at enterprise scale.See:Synergy Research Group,“Hyperscaleoperators and col

254、ocation continue to drive huge changes in data center capacity trends,”Aug.7,2024.10.Yifan Yu,“AIs looming climate cost:Energy demand surges amid data center race,”Nikkei Asia,June12,2024.11.Data center BMO report,Communications Infrastructure,“1Q24 data center leasing:Records are madeto be broken,”

255、April 28,2024.Further,Deloitte analysis based on information from select techcompanies publicly available sources such as earnings releases and DellOro Groups market researchdata on data center IT capital expenditure shows that if we consider the capital expenditure spending ofother data center prov

256、iders,including third-party operators and outsourced cloud service providers,datacenters aggregate capital expenditure spending could be at least US$250 billion in 2025.See:BaronFung,“Market research on data center IT capex,”Delloro Group,accessed Nov.4,2024.12.Nitin Mittal,Costi Perricos,Brenna Sni

257、derman,Kate Schmidt,and David Jarvis,“Now decides next:Getting real about generative AI,”Deloittes State of Generative AI in the Enterprise quarter two report,Deloitte,April 2024.13.Deloitte analysis based on publicly available research reports including:Wania Khan,Davide De Chiara,Ah-Lian Kor,and M

258、arta Chinnici,“Advanced data analytics modeling for evidence-based data centerenergy management,”Physica A 624,2023;Kazi Main Uddin Ahmed,Math H.J.Bollen,and ManuelAlvarez,“A review of data centers energy consumption and reliability modeling,”inIEEE Access 9,2021:pp.152536152563.14.Tom Dotan and Asa

259、 Fitch,“Why the AI industrys thirst for new data centers cant be satis?ed,”TheWall Street Journal,April 24,2024.15.Noam Brousard,“Examining the impact of chip power reduction on data center economics,”Semiconductor Engineering,March 12,2024.16.Based on our analysis of multiple publicly available sou

260、rces including:Michael Studer,“The energychallenge of powering AI chips,”Robeco,Nov.6,2023;Agam Shah,“Generative AI to account for1.5%of worlds power consumption by 2029,”HPCwire,July 8,2024.17.From our study and analysis of select gen AI data center chip solutions o?ered by major AI chip vendors,fu

261、rther corroborated with publicly available third-party sources including:Beth Kindig,“AI powerconsumption:Rapidly becoming mission-critical,”Forbes,June 20,2024.18.Jones Lang LaSalle,“Data centers 2024 global outlook,”Jan.31,2024;Doug Eadline,“The gen AIdata center squeeze is here,”HPCwire,Feb.1,202

262、4;Per IDC,besides graphics processing unit,servers,data centers also need to grapple with a corresponding growth in storage capacity,which is likely todouble between 2023 and 2027 to reach 21 zettabytes in 2027.See:John Rydning,“Worldwide GlobalStorageSphere forecast,2023 to 2027:Despite decreased p

263、etabyte demand near term,the installed baseof storage capacity continues to grow long term,”IDC Corporate,May 2023.19.Patel,Nishball,and Eliahou Ontiveros,“AI data center energy dilemma.”20.Sean Michael Kerner,“What are large language models?”TechTarget,May 2024;Yu,“AIs loomingclimate cost.”21.Alex

264、de Vries,“The growing energy footprint of arti?cial intelligence,”Joule 7,no.10(2023):pp.21912194.22.Eren am,Zoe Hungerford,Niklas Schoch,Francys Pinto Miranda,and Carlos David Yez de Len,“Electricity 2024:Analysis and forecast to 2026 report,”International Energy Agency,accessed Nov.4,2024.23.Deloi

265、tte analysis based on publicly available reports and sources including:de Vries“The growingenergy footprint of arti?cial intelligence,”pp.21912194.24.Deloitte analysis based on data related to energy use and electricity consumption in homes in the UnitedStates.See:US Energy Information Administratio

266、n,“Use of energy explained,”accessed Dec.18,2023.25.Darren Sweeney,“Utility execs prepare for tripling of electricity demand by 2050,”S&P Global,April19,2023.26.Robert Walton,“US electricity load growth forecast jumps 81%led by data centers,”Utility Dive,Dec.13,2023.27.Aaron Larson,“How utilities ar

267、e planning for extreme weather events and mitigating risks,”POWER,March 13,2024.28.am,Hungerford,Schoch,Miranda,and de Len,“Electricity 2024.”29.Baraniuk,“Electricity grids creak as AI demands soar.”30.Yu,“AIs looming climate cost.”31.Data on Chinas energy use and CO2 emissions sourced from Internat

268、ional Energy Agency,accessedSeptember 25,2024.See:International Energy Agency,“Chinas energy use,”accessed Nov.4,2024;International Energy Agency,“Chinas CO2 emissions,”accessed Nov.4,2024.32.Baraniuk,“Electricity grids creak as AI demands soar.”33.Paul ODonoghue,“Build it and they will hum:What nex

269、t for Ireland and data centers?”The Journal,Sept.2,2024.34.Hosting Journalist,“City of Amsterdam puts halt to new data center construction,”Dec.21,2023.35.With every 1C increase,operators could save 2%to 5%on the energy they use for cooling equipment.To read further,see:Inno Flores,“Singapore unveil

270、s green data center road map amid AI boom thatstrains energy resources,”Tech Times,May 30,2024.36.Julie R.Peasley,“Ranked:Top 50 data center markets by power consumption,”Visual Capitalist,Jan.10,2024.37.Whitney Pipkin,“Energy demands for Northern Virginia data centers almost too big to compute,”Bay

271、Journal,June 18,2024.38.Zach Bright,“Southeast utilities have a very big ask:More gas,”E&E News,Jan.22,2024.39.Ibid.40.Robert Walton,“AI is enhancing electric grids,but surging energy use and security risks are keyconcerns,”Utility Dive,Oct.23,2023.41.Karen Hao,“AI is taking water from the desert,”T

272、he Atlantic,March 1,2024.42.Deloitte analysis based on publicly available information sources including:Jennifer Billock,“Photos:How much water it takes to create 30 common items,”North Shore News,Jan.19,2023.43.Hao,“AI is taking water from the desert”;One case in point is Chinawhere its data center

273、s annualwater consumption is expected to increase from around 1.3 billion cubic meter as of 2023 to over 3billion cubic meter by 2030.To read further,see:Yu,“AIs looming climate cost.”44.Eadline,“The gen AI data center squeeze is here.”45.Diana Goovaerts,“Data center operators want to run chips at h

274、igher temps.Heres why.”FierceNetwork,June 11,2024.46.Scott Wilson,“Is immersion cooling the answer to sustainable data centers?”Ramboll,Dec.13,2023.47.David Eisenband,“100+kW per rack in data centers:The evolution and revolution of power density,”Ramboll,March 13,2024;Direct-to-chip cooling(also kno

275、wn as cold plate liquid cooling or directliquid cooling)cools down servers by distributing heat directly to server components,while,immersioncooling involves submerging servers and components in a liquid dielectric coolant that also helps preventelectric discharge.48.Based on Deloittes analysis of d

276、evelopments and announcements from select major cloud hyperscalersand tech companies and information gathered from publicly available sources(time period:2023 and?rst three quarters of 2024).49.Joseph Rand,Nick Manderlink,Will Gorman,Ryan Wiser,Joachim Seel,Julie Mulvaney Kemp,Seongeun Jeong,and Fri

277、tz Kahrl,“Queued up:2024 edition,”Lawrence Berkeley National Laboratory,April 2024.50.Based on Deloittes analysis of developments and announcements from select major cloud hyperscalersand tech companies and information gathered from publicly available information sources betweenthe?rst quarter of 20

278、23 and the third quarter of 2024.51.Julian Spector,“Duke Energy wants to help Big Tech buy the24/7clean energy itneeds,”CanaryMedia,June 11,2024.52.For example,its not easy to submerge and drop a 1,300-ton data center unit underwater,especially sinceit demands special equipment to withstand pressure

279、 and corrosion caused by seawater.Moreover,thereare concerns related to its impact on marine life.53.David Schlissel and Dennis Wamsted,“Small modular reactors:Still too expensive,too slow,and toorisky,”Institute for Energy Economics and Financial Analysis,May 2024.54.Deloittes analysis of data and

280、information gathered from multiple reports from S&P Global MarketIntelligence,published during March and August 2024.55.Manish Kumar,“Indias data center boom opens up a fresh segment for green developers,”Saur EnergyInternational,July 1,2024.56.Naureen S.Malik and Bloomberg,“With AI forcing data cen

281、ters to consume more energy,software thathunts for clean electricity across the globe gains currency,”Fortune,Feb.25,2024.57.Based on Deloittes analysis of developments and announcements from select major cloud hyperscalers,tech companies,and power and utility playerson publicly available informatio

282、n sources during 2023and the?rst three quarters of 2024.58.Duncan Stewart,Nobuo Okubo,Patrick Jehu,and Michael Liu,“The cloud migration forecast:Cloudywith a chance of clouds,”Deloitte Insights,Dec.7,2020.59.Wylie Wong,“Nvidia launched next-generation Blackwell GPUs amid AI arms race,”Data CenterKno

283、wledge,March 19,2024;For instance,Nvidia notes that it can train a very large AI model using2,000 Grace Blackwell chips in 90 days,consuming 4 MW power.In comparison,it would take as muchas 8,000 of the previous generation chips to do the same work within the same time,consuming 15 MWpower.60.To rea

284、d further,see section“Generative AI comes to the enterprise edge:On prem AI is alive and well”in our 2025 TMT Predictions chapter on“Updates”;Additionally,see:SabuzimaNayak,RiponPatgiri,LilapatiWaikhom,and ArifAhmed,“A review on edge analytics:Issues,challenges,opportunities,promises,future directio

285、ns,and applications,”Digital Communicationsand Networks 10,no.3(2024):pp.783804.61.Yu,“AIs looming climate cost.”62.Luke Cavanagh,“GPUs vs.CPUs in the context of AI and web hosting platforms,”Liquid Web,Aug.20,2024.63.Eadline,“The gen AI data center squeeze is here.”64.Goovaerts,“Data center operato

286、rs want to run chips at higher temps.Heres why.”65.Baraniuk,“Electricity grids creak as AI demands soar.”AcknowledgementsThe authors would like to thank Dilip Krishna,Marlene Motyka,Jim Thomson,Adrienne Himmelberger,ThomasSchlaak,Freedom-Kai Phillips,Johannes Truby,Clement Cabot,Negina Rood,Ankit Dh

287、ameja,Suzanna Sanborn,andAkash Cha?erji for their contributions to this article.Cover image by:Jaime Austin;Ge?y Images,Adobe StockAmbitious stadium projects aimto bridge public-privateinvestment goalsSports owners transform stadiums intodestinations,driving socioeconomic growth,community engagement

288、,and revenuediversi?cationARTICLE10 MINUTE READInvestment in sports infrastructure such as stadiums,playing grounds,and training facilities,is seeing anupward trend,as these developments often instigate broad social and economic returns to both the publicand private sectors.This trend has seen a ren

289、ewed focus in recent years,as sports teams in regions such asNorth America,Europe,and Asia Paci?c invest heavily in developing infrastructure.With growth as acommon goal,governments and communities can work with sports investors to provide supplementaryinfrastructure like transportation links,and co

290、mmunity resources can help supercharge the socioeconomicimpact of sports.In the next year,multiple infrastructure projects are expected to take place that couldincrease economic returns for communities,as well as sports becoming further embedded in the heart ofculture and society.In 2025,Deloitte pr

291、edicts that over 300 global sports stadiums will have begun renovations or new builds.Almost 50%of these new stadium infrastructure projects are expected to take place across North Americaand Europe,according to Deloitte analysis of sports infrastructure developments taking place.The growthof stadiu

292、m investment across Europe,primarily focused on soccer stadiums,attempts to attract a new waveof fans and could provide revenue diversi?cation opportunities for the organizations setting out on theseprograms.In doing so,stadium developments can help reach targets for maximizing return on investmentf

293、or private investors as well as accelerating socioeconomic objectives for the public sector.Stadiuminvestment is also likely to increase across multiple global regions,as sports-led regeneration projects takeplace and fans increasingly demand innovative touchpoints both inside and outside the stadiu

294、m.A community-centric approach to venue developmentSports organizations can bring communities together,improve feelings of civic pride and cohesion,andfurther diversify the cultural o?erings of a city.Delivering a sports-led regeneration program using sportsvenues as anchors often requires collabora

295、tion with governments and other key stakeholders to help deliverstrategic initiatives;engage with the community;and develop a sustainable,thriving development wherepeople want to live and visit.The development of sports stadiums may no longer be able to hold focus on a single clubs interests.Indevel

296、oping new and enhanced stadiums,community bene?ts should in?ltrate various aspects of decision-making.In April 2024,Knighthead Capital Management,owners of now English Football League One clubBirmingham City,announced its plans to create a Sports Quarter anchored around a new world-classstadium.1 Ch

297、airman Tom Wagner outlined his vision for the ambitious project to include the stadium,mensand womens training grounds,and academy teams all in one location within walking distance of the citycenter.2 Furthermore,hes alluded to communications with hotels and other commercial entities that areinteres

298、ted in being based in the site and part of the regeneration of east Birmingham.3 Through this,the“Blues”will be fully embedded as part of the city and act as a“beacon for excellence”that is recognizedworldwide,according to Wagner.4 The Sports Quarter project is anticipated to cost between 2 billion

299、and3 billion,with hopes to spur long-term socioeconomic impact for the West Midlands community.5 TheKnighthead ownership group is engaged with government and public sectors on a range of strategicpriorities.6In Major League Baseball,the Tampa Bay Rays signed a deal in July 2024 with the City of St.P

300、etersburg,Florida,to build a new ballpark.7 The development group for the new stadium pledged to create 1,250units of a?ordable housing,30,000 construction jobs,and 7,000 permanent jobs,with some reserved forlocal and historically disadvantaged residents.8 The project leaders have reiterated their c

301、ommitment tobridging the generational wealth gap in the surrounding communities with this new venue,considering theproject a“failure”if they do not reach that objective.9Enhancing fan engagement by meeting generationalpreferencesFans of di?erent generations are consuming sports in di?erent ways.Acco

302、rding to Deloitte UKs The Futureof Sport 2024,84%of global sports leaders surveyed said they expected di?erent consumption preferencesto be one of the most impactful next-generation trends over the next?ve years.Sports organizations shouldbalance embracing the core traditions of a game-day experienc

303、e with Generation Z and Generation Alphasheightened expectations of entertainment.10As a?rst step in developing the stadium experience,organizations should consider the basics:comfort andsafety,view,quality on-?eld product,and an exciting atmosphere.These attributes are important for manyfans and sh

304、ould be mastered before looking at any advanced plans.After establishing these foundations,some organizations may look to di?erentiate their experiences byproviding end-to-end entertainment options for fans before,during,and after the game.Not only can thishelp enable greater spending of both time a

305、nd money for fans at these stadiums,but it can alsofoster agreater sense of community around the sports organizations.By integrating community culture into thefabric of the stadium,the game-day experience can be uniquely localized.The Toronto Blue Jays embarked on new renovations on their stadium at

306、 the Rogers Centre,with twophases of work aimed at elevating the fan experience.Phase one,completed in 2023,unveiled?ve distinctout?eld“neighborhoods”within the ballpark stands,encompassing local cuisine and entertainment,representing di?erentiated fan experiences,and providing social spaces in each

307、.11 The renovations alsoincluded upgraded digital technology such as“Tap N Go,”a new automated market service for food andbeverage,as well as“Walk Thru Bru,”a self-serve beverage-concession stand to speed up service times.12Phase two of the Rogers Centre renovation includes fan-centric adjustments s

308、uch as angling seats towardhome plate for improved sightlines.13The smart stadiumThe next generation tends to consume sports in a digital-?rst manner,and their game-day experiences areoften no di?erent.14 Some sports organizations are designing“smart stadium districts”that integrateadvanced technolo

309、gy to personalize the fan experience.15 The global smart stadium market is growing,witha 2024 market size of over US$8 billion,which is expected to reach more than US$38 billion by 2033.16The requirements for fan engagement are changing,as Gen Z and Gen Alpha prefer shorter,more dynamiccontent.17 Ex

310、periences are becoming a di?erentiator for organizations,as fans increasingly want value andwill pay for greater experiences over material objects.18 New builds are incorporating elements of gaming,merchandising operations,and designs that consider“second-screen syndrome,”in which the majority offan

311、s tend to look at secondary screens while watching sports.According to Deloitte research,77%of sportsfans surveyed say theyve done at least one additional activity related to a game while watching a sportingevent,including looking up stats,using social media,or betting.19 New stadiums are using inte

312、gratedtechnology to broadcast these elements,keeping a greater portion of fans focus within the stadium.20The National Basketball Associations Los Angeles Clippers unveiled their new arena,the Intuit Dome,builtto di?erentiate the fan experience at the forefront.A key highlight of the new arena is th

313、e custom-built“Halo Board,”which optimizes sight lines from all seats and prioritizes the viewing experience of upper-bowl seating.The double-sided video board hanging above center court contains a game feed,“coachescorner”with in-depth statistics,instant replays,Steve Cam(keeping tabs on Clippers o

314、wner Steve Ballmer),player pro?le features including photos and other personal promotions such as player foundationinformation,and more.To enhance the fan experience,the Halo Board will also reinvent the coveted T-shirttoss with T-shirt cannons attached,enabling fans in the upper tiers to be able to

315、 receive merchandise aswell.The Intuit Dome plans to reach levels of engagement not seen before in sports,as fans will berewarded for their cheering and provided gaming consoles at each seat for use in game-day entertainmentactivities,further gamifying fandom.21The next generation of fans will likel

316、y expect personalized,seamless,and on-demand experiences.Stadiumdistricts can incentivize fans to stay longer and enjoy di?erent o?erings including food,music and culture,as well as social spaces for di?erent types of fans.Sports organizations are building these districts to givelocal and visiting f

317、ans a premium experience and generating new touchpoints for them while they are in thearea.22 New technologies are streamlining purchasing by making it quicker to purchase,such as click andcollect merchandise or food and beverage,as well as ticket software that can personalize messaging for eachfan.

318、23 Additionally,fans are entering the stadium in new ways,often contributing to a more convenient fanexperience.Mercedes-Benz Stadium partnered with Delta to create Fly-Through Lanes,which uses facialrecognition to gain quick entry for fans visiting the stadium.24Embracing infrastructure and tech to

319、 help diversifyrevenue streamsSports owners are also using enhanced infrastructure and digital technology to further diversify theirrevenue generation.The success of sports organizations is evolving as,historically,global sports organizations typically reliedheavily on broadcast revenue to supplemen

320、t spend on wages and other costs.25 Generally,an increase incentral broadcast revenue would lead to an increase in salary caps for North American sports leagues andspending allowed under certain cost-control regulations across European sports.26Organizations seem to recognize that an overreliance on

321、 one revenue stream can place a ceiling on earningpotential and expose them to market shocks,such as the COVID-19 pandemic.As elite sports clubs holdcultural and commercial capital,some organizations are leveraging their commercial assets to help bolsterrevenue.27 Whereas ticketing and broadcast rev

322、enue are generally capped at capacity or out of scope forindividual clubs to negotiate,commercial revenue can be an underutilized lever for organizations to growwithin their control.28 The evolution of stadiums into broader entertainment districts can help sportsorganizations increase their commerci

323、al footprint by expanding their o?erings.For example,the PremierLeagues Tottenham Hotspur has leveraged its new stadium to increase commercial revenue from 72million in the 2016 to 2017 season to 227 million in the 2022 to 2023 season,supported by non-soccerevents such as hosting National Football L

324、eague games and concerts.29Recent renovations of European soccer giant Real Madrids Santiago Bernabu stadium in 2023 saw theclub recognize record revenue in the 2022 to 2023 season,with every business line experiencing growth(outside of broadcasting rights,which are in the middle of a contract with

325、broadcasters).30 In July 2024,theclub announced it had generated over 1 billion in revenue,the highest?gure ever generated by a soccerclub.31 The club surpassed the billion-euro mark by initiating new business ventures during the latter part ofthe?nancial year,primarily through hosting major events

326、and introducing premium VIP experiences.32 Forexample,Real Madrid brought in 18 million from Colombian pop singer Karol Gs four concerts held atthe stadium.33 The club is expected to?nish renovations of the stadium in the upcoming season and islooking to add non-soccerrelated income in future years.

327、34Emerging trends in sports infrastructureFrom responding to the growth in womens sports to driving greater sustainability,sports organizations areworking to embrace new fan experiences and priorities.This evolution in sports infrastructure aims tocreate a more inclusive,innovative,and responsible f

328、uture for the industry.Infrastructure for womens sportsAs womens sports continue to garner attention and rising valuations,organizations are beginning to focusmore on womens sport-speci?c infrastructure.In the National Womens Soccer League,the Kansas CityCurrent christened their US$117 million river

329、front stadium,CPKC Stadium,recognized as the?rst stadiumbuilt speci?cally for a womens professional sports team.35 Further increasing the socioeconomic impact ofthe stadium,plans have been approved for a US$650 million mixed-use development,including apartments,hotels,restaurants,and retailers,to be

330、 built alongside the soccer stadium.36 The stadium is expected togenerate almost US$50 million in annual economic impact to Kansas City.37Across the Womens National Basketball Association,new training facilities are contributing to animprovement in the on-court product and,in turn,the overall valuat

331、ion of the franchises.The Las VegasAces opened their facility in 2023,with the Seattle Storm,Phoenix Mercury,and Chicago Sky all followingsuit.38In England,Womens Super League club Brighton&Hove Albion is in the planning phase to build astadium purpose-built for its womens team.39 Manchester City Wo

332、men has also had plans approved for apurpose-built training facility at the clubs City Football Academy site.40Premium and personalized hospitality offeringsHospitality may be moving away from traditional corporate o?erings and is now often being used as a toolto create more accessible,di?erentiated

333、 experiences for all demographics.Sports organizations are workingto partner with higher-end brands in these spaces to o?er more premium hospitality services,from celebritychefs to take-out goodie bags.41 Formula One provides an example of premium hospitality o?erings in itsPaddock clubs,engaging with social in?uencers,celebrities,and brand partners.42 As sports organizationsrenovate and build new