《S&P Global：2024年全球人工智能趨勢報告（英文版）（46頁）.pdf》由會員分享，可在線閱讀，更多相關《S&P Global：2024年全球人工智能趨勢報告（英文版）（46頁）.pdf（46頁珍藏版）》請在三個皮匠報告上搜索。

1、2024 Global AI Trends ReportDiscover how AI is transforming industries and what it takes tostay ahead.Read the full 2024 Global AI Trends Report,basedon a survey of 1,500 AI practitioners,to unlock key insights andstrategies for navigating the rapidly evolving AI landscape.Explore our developer-frie

2、ndly HTML to PDF APIPrinted using PDFCrowdHTML to PDFTable of contentsIntroductionKey findingsDefining AI leadersAI maturity and adoptionAI project scaling challengesRise of generative AIGPU availabilityEnvironmental impactConclusionsMethodologyAbout the authorsExplore our developer-friendly HTML to

3、 PDF APIPrinted using PDFCrowdHTML to PDFIntroductionThe 2024 Global Trends in AI report delves intothe underlying trends surrounding AI adoption.In last years Global Trends in AI report,weexplored the divide between organizationsthat were successfully running AI in productionand those that were not

4、.In this years study,we revisit this AI leadership theme,drawing onsome key practices that leading organizationsare doing differently while deep-diving intothe value drivers,infrastructure decisions andenvironmental practices that are shaping AIstrategies.To develop this study,S&P GlobalMarket Intel

5、ligence surveyed more than 1,500global AI decision-makers and engaged in 1:1interviews with senior IT executives about theirAI projects and initiatives.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFKey findings1.AI applications are nowpervasive in the enterprise;inve

6、stments in product qualityand IT efficiency are top priorities.AI adoption continues at breakneck speed,withthe technology increasingly viewed as anembedded and strategic capability.2.Many AI projects fail to scale legacy data architectures are theAI initiatives are maturing rapidly:In thelast year,

7、reported levels of AI maturityhave undergone a radical shift.In 2023,survey respondents were still largelyexperimenting with AI or had isolateddeployments in small pockets of theirorganizations.This year,the majority ofrespondents report that AI is“currentlywidely implemented”and“driving criticalval

8、ue”in their organization.Product improvement and operationaleffectiveness are key investment drivers:Organizations are increasingly applying AIto enhance top-line revenue andcompetitive differentiation,with improvingproduct or service quality(42%)the mostpopular objective,and with manytargeting incr

9、eased revenue growth(39%).Simultaneously,organizations recognize the potential toboost their operational effectiveness byimproving workforce productivity(40%)and IT efficiencies(41%),along withaccelerating their overall pace ofinnovation(39%).Explore our developer-friendly HTML to PDF APIPrinted usi

10、ng PDFCrowdHTML to PDFculprit.AI projects are challenged by weak datafoundations.Legacy data architectures areimpeding broader deployment.Achieving scale remains a challenge:Organizations are facing significantchallenges in achieving the desired reachof their AI projects.The averageorganization has

11、10 projects in the pilotphase and 16 in limited deployment,butonly six deployed at scale.Availability of quality data is a majorobstacle:Data quality is the greatestchallenge to moving AI projects intoproduction.The challenge for projectteams is not so muchabout identifying relevant data,but itsavai

12、lability;organizations are struggling tobuild a consistent,integrated datafoundation for projects.Modernizing data architectures is criticalto success:Given this,it is unsurprisingthat the greatest proportion ofrespondents(35%)cite storage and datamanagement as the primaryinfrastructure issues hinde

13、ring AIdeployments significantly greater thancompute(26%),security(23%)andnetworking(15%).Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFGenerative AI trailblazersare the 24%of organizationsthat suggest thatgenerative AI is anintegrated capability acrossworkflows with

14、in theirbusiness.3.Generative AI has rapidlyeclipsed other AI applications.Generative AI has gained significant traction ina short time.AI trailblazers are realizingconcrete benefits and are poised tocompound their competitive advantage.Generative AI is the focus:An astonishing88%of organizations ar

15、e now activelyinvestigating generative AI,faroutstripping other AI applications such asprediction models(61%),classification(51%),expert systems(39%)and robotics(30%).Dedicated budgets for generative AI as aproportion of overall AI investments aregrowing as organizations begin torecognize the potent

16、ial benefits ofintegrated generative AI capabilities.Generative AI adoption is exploding:Despite only being in the market for arelatively short time,24%of organizationssay they already see generative AI as anintegrated capability deployed acrosstheir organization.Just 11%of respondentsare not invest

17、ing in generative AI at all,and the majority of organizations areactively in the process of turning thisinvestment into scaled-up,integratedcapabilities.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDF4.GPU availability continues to beconstrained,shapinginfrastructure

18、decision-making.Access to GPUs is a major concern fororganizations;GPU clouds may offer a scalablesolution.Generative AI trailblazers are expected tocompound their competitive advantage:Organizations that have alreadyintegrated generative AI across theirorganization plan to continue increasingtheir

19、investments:They expect generativeAI budgets to reach 47%of their total AIbudget in the next 12 months,faroutpacing less“AI-mature”organizations.The majority of these generative AItrailblazers are seeing a significantpositive impact from the technologyacross the full range of targeted benefits.Those

20、 benefits are likely to compoundtheir competitive advantage given thatthose still in the experimentation phasesof their generative AI projects are notseeing the same increases inorganizational innovation,new productdevelopment and time to market.Accessing GPUs continues to be achallenge:Four in 10 o

21、rganizationssurveyed suggest access to AIaccelerators is a leading consideration intheir infrastructure decision-making,and30%cite GPU availability among their topthree most serious challenges in movingAI models into production.Regional pressures persist:In somegeographies,particularly in Asia-Pacif

22、ic,lack of access to GPUs is restrictingorganizations from deploying AI;38%oforganizations in India see acceleratoraccess among their top three challengesto moving AI projects into production.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDF5.Concerns about AIsenvironme

23、ntal impact persist butare not slowing AI adoption;sustainable AI practices offeropportunity to mitigateemissions.AIs environmental and energy impact is still aprominent concern for many organizations,butit is not slowing the decision to invest in AIprojects.With many organizations seeingsustainabil

24、ity practices deliver meaningfulimpacts,there is a clear opportunity to addressthe emissions challenge.Hyperscaler and GPU clouds serve as keychannels for companies to access GPUs:The need for accelerators has driven 46%of surveyed organizations to leveragehyperscale public clouds for modeltraining,

25、as well as increasingly specialist GPU cloud providers(32%).Concerns about AIs energy and carbonimpact remain prominent:Nearly two-thirds(64%)of organizations say that theyare concerned about the impact ofAI/machine learning(ML)projects on theirenergy use and carbon footprint;25%oforganizations indi

26、cate they are veryconcerned.Adoption of sustainable datainfrastructure technologies is an area offocus:Clearly,sustainability credentialsfrom technology providers are becomingessential,with 42%of organizationsindicating that they have invested inenergy-efficient IT hardware/systems toaddress the pot

27、ential environmentalimpacts of their AI initiatives overthe past 12 months.Of those,56%believethis has had a“high”or“very high”impact.Others have found that making changesin data infrastructure vendors(59%)andExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFAI project s

28、cope(57%)have had a“high”or“very high”impact.Sustainability is an important,but notthe primary,factor in AI decision-making:More than a quarter(30%)oforganizations report that sustainabilityinitiatives are a driver of AI adoption asthey look to apply AI to improve energyefficiency and mitigate emiss

29、ions.Whilethis is notable,sustainability is,in fact,theleast-mentioned driver overall.Evenwhere energy-reduction initiatives are thegoal,meeting sustainability targets cantake a back seat to cost savings andimproving operational efficienciesas the principal objective.In the context ofall the issues

30、that most strongly inform AIinfrastructure decision-making,sustainability is mid-table:37%oforganizations are prioritizing it,but it isoutranked by more prominent issues suchas security(47%)and access to AIaccelerators(44%).Defining AIleadersExplore our developer-friendly HTML to PDF APIPrinted usin

31、g PDFCrowdHTML to PDFIn the past year,the sentiment toward AI hasrapidly changed,as has AIs strategic role.AImaturity has grown to the point that the vastmajority of organizations have some form of AIin production,so a comparison of the“haves”and the“have nots”offers little analytical value.Instead,

32、the emerging divide appears to bethose that are able to harness the latesttechnology breakthroughs and deliver AI atscale,and those that cannot.With the most significant AI breakthrough andarguably most important technologicalinnovation in the past decade,generative AI asa strategic imperative is in

33、escapable.Thosethat have quickly and efficiently tapped intothis technological breakthrough are distancingthemselves from the chasing pack.Based on these assumptions,we have definedAI leaders to be those that have reported thefollowing accomplishments:Explore our developer-friendly HTML to PDF APIPr

34、inted using PDFCrowdHTML to PDFFigure 2:Who are AI leaders in2024?10%of total(n=153)Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.There are several noticeable variances in themakeup and characteristics of AI leaders in2024.AI/ML projects in production envir

35、onmentsdriving real-world impact in criticaloperations.Implemented AI/ML widely across theirorganization,achieving far greater scalethan limited and siloed AI projects.Capitalized on the most significanttechnological breakthrough of thisdecade(generative AI)and positioned itas an integrated capabili

36、ty across theirbusiness and workflows.The top 10%of the market engage inseveral distinguishing practices that setthem apart from the bulk of organizationsacross the reports five key themes.Industry:Healthcare respondents(18%)have a greater proportion of AI leaders thanother industries.Company size:E

37、nterprises(16%)withgreater access to capital,resources,AI skillsets and typically more mature digitaltransformation projects lead othercompany sizes proportionally as AI leaders.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFRegion:North America(16%)has asignificantly

38、 greater proportion of AI leadersthan Asia-Pacific(8%)and EMEA(6%).Contributing factors may be greater accessto AI talent from industry and educationalinstitutions,and regional availability ofventure funding and capital.Business models:AI providers(15%)aremore likely to be considered AI leaders than

39、other organizations.However,the differenceis less pronounced than one might assume;in some instances,AI providers,despitebuilding AI solutions for customers,have notnecessarily fully implemented capabilitieswithin their own company.AI applicationsare now pervasivein the enterprise;investments inprod

40、uct qualityand IT efficiencyare top prioritiesIn the last year,AIs role has changed withinmany organizations,shifting from a minorcomponent of an overall strategy to a criticalembedded capability.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFAI is becoming a fundamen

41、tal aspect of manyorganizations strategies,increasingly seen asboth widely implemented and critical.Theproportion of respondents who indicate that AIis a“minor component of a broader strategy”in their organization halved from last yearssurvey,while the proportion of respondentswho see AI as“widely i

42、mplemented,drivingcritical value”increased from 28%to 33%,becoming the most common answer.For NorthAmerica respondents,it is even higher at 48%,in comparison to Asia-Pacific(26%)and EMEA(25%).Key insights:The most common adoption status forAI has shifted from a minorcomponent of a broader strategy i

43、n2023 to currently widely implementedand driving critical value in 2024.Improving product or service qualityand generating cost savings from ITefficiencies are the leading drivers fordeveloping AI applications.Many strategic objectives such asimproving time to market and gainingproduct or service di

44、fferentiation areseeing greater focus.AI seeing wider deployment in2024AI organizational scale20242023Widely implemented33%28%Few use cases33%26%Single use case16%16%Minor component17%30%Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFAIs reach is not limited to the br

45、eadth ofimplementation but includes the technologysstrategic impact.Historically,AIs valueproposition has been closely associated withreducing costs.Previous AI advancements inrobotic process automation,for example,wereclosely aligned with objectives such asheadcount reduction or reducing outsourcin

46、gcosts.It is not so much that the cost-reductionopportunity presented by AI is being crowdedout indeed,generating cost savings from ITefficiencies is the second most popularobjective for AI rather,cost drivers are beingpaired with more strategic objectives.Forexample,more than a third(39%)ofresponde

47、nts in our survey see revenue growthas a key driver of their AI initiatives.As Figure 3illustrates,companies are not just trying toachieve more with AI than they were last year,but they also see a clearer alignment withrevenue drivers.They are significantly moreaware of the opportunity for AI to be

48、used togain product differentiation and driveimproved time to market than they were lastyear.Figure 3:The year-over-yearchange in drivers for AIapplication developmentExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFWhat are AI leaders doingdifferently?Leaders perceive

49、a wider range ofobjectives as driving their AIstrategies.This spread ofobjectives helps better informwhere AI could be most impactful.It also sets the basis for a strongerbusiness case for investing in AI,helping build a narrative that canappeal to a wider spread ofstakeholders.“I started with these

50、 experiments a yearand a half ago.And then it took us a yearto build we now have about 5-10 usecases in production.”CIO,transportation/logistics/warehousing,1,000-5,000 employees,USMany AI projectsfail to scale;legacy dataExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PD

51、Farchitectures arethe culpritAIs growing strategic importance is driving asignificant increase in initiatives acrossbusinesses.Broad experimentation andeducation are and organizations would beremiss not to encourage it.However,theopportunity is being throttled by projects thatlack a clear pathway to

52、 recognizing value,hampered by data challenges.AI projects riskstalling in a limited deployment purgatory,costing a company money,time andresources,while not seeing desired levels ofuse.Initiatives are becoming snagged on datasiloes,poor data quality and ineffective dataand model pipelines.As organi

53、zations invest to apply AI to an ever-growing set of objectives,a kink is emerging inorganizations project pipelines.While moreinitiatives are funneled toward AI projectteams,there is a buildup of initiatives that haveKey insights:In the average organization,51%of AIprojects are in production but no

54、tbeing delivered at scale.Data quality is the greatest inhibitorwhen moving AI projects intoproduction environments.Storage and data management arethe most common infrastructuralinhibitors to AI initiatives,identified by35%of organizations;however,thosethat have AI widely implemented feelthese chall

55、enges less keenly.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFonly been partially deployed.As Figure 4illustrates,respondent organizations,onaverage,have more projects classified asbeing in production with a limited deploymentthan scaled-up capabilities.In chasing

56、newinitiatives,many organizations may fail tomaximize the value of their existinginvestments.The crux of the problem appearsto be data quality and availability,with legacydata architectures causing this pipelinestoppage in many organizations.Data quality is the most frequently identifiedchallenge as

57、 organizations move their projectsfrom pilots to production.As Figure 5 illustrates,data quality concerns identified by 42%oforganizations as among their top three barriers are even more significant than skillsshortages(32%)and budget limitations(31%).Organizations in media and entertainment(59%),hi

58、gher education(53%),and aerospaceand defense(48%)feel the data qualitychallenge particularly keenly.Figure 4:Many projects fail tograduate from limiteddeployment to delivering at scaleExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFQ.How many AI/ML projects do you curr

59、ently have:Inpilot/proof of concept;in production,limited deployment;in production,at scale?Base:All respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.The data quality challenge is not a lack of datato build performant models but,rather,tha

60、t thedata is not set up in such as way that projectteams can take full advantage of it.Whenasked specifically to rank the primary datachallenges to move projects to productionenvironments,respondents indicated thatavailability of quality data is a more notableimpediment than identifying relevant dat

61、a.With 34%of organizations perceivingavailability of quality data as a top three datachallenge,outranked only by data privacyconcerns(35%),it is clear that manyorganizations are poorly set up for effectivedata management.Legacy data technologies seem to be aleading cause of these data managementshor

62、tcomings.Data management and storageare most commonly seen as the infrastructurecomponents that inhibit AI applicationdevelopment.More than a third(35%)ofrespondents see them as a more serious issuethan security(23%),compute(26%)andnetworking resources(15%).Explore our developer-friendly HTML to PDF

63、 APIPrinted using PDFCrowdHTML to PDFTellingly,organizations that are most effectivelyscaling AI initiatives are less constrained bythese data management and storagecomponents.Just 28%of respondents whoreported that AI is widely implemented withintheir organization perceive storage and datamanagemen

64、t challenges as their greatestinhibitor;instead,they feel greater pressurefrom networking or compute resources.Thiscompares to 42%of respondents who perceiveAI as being limited to a few use cases orprojects within their organization.Organizationsthat are delivering AI at scale appear to havefocused

65、on investing in upgrading the systemsand technologies used to store or managedata.Figure 5:Top three impedimentsto organizations moving an AI/MLapplication from pilot toproduction environments“We still have challenges with masterdata.Branches had different SKUs forinventory;if I take that siloed dat

66、a andput it into a model,well get the wrongresults.Cleaning up this data is ourfocus.”CIO,transportation/logistics/warehousing1,000-5,000 employees,USExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFQ.What are the primary challenges or impediments tomoving an AI/ML appl

67、ication from proof-ofconcept/piloting stages to production environments?Base:All respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.This investment appears to be key becausedata management and storage shortcomingsare filtering through into A

68、I project life cycles,with organizations struggling to effectivelyprepare data for model building anddeployment.Many organizations report thatthe most challenging aspects of AI initiativesare the data preprocessing stages(see Figure6).Despite the growing number oforganizations indicating that AI has

69、 beenwidely implemented within their organizationover the past 12 months,there has been nomeaningful improvement year over year interms of performance against these datapreprocessing steps.Bringing AI projects livebut limiting their value or extensibility with weakdata foundations sets a poor preced

70、ent for thenext wave of initiatives in the early stages ofexploration.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFFigure 6:Organizations find theearly data steps of the AI lifecycle as challenging as modelbuildingProportion of respondents thatidentify AI life cycle

71、 stage as“mostchallenging”Q.What stages of the AI/ML application life cycle arecurrently the most challenging(Rank 1)?Base:All respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.Immature data management toolsets are aworrying backdrop for th

72、e increasingly data-hungry AI strategies many organizations areembarking upon.More than three-quarters(80%)of respondents forecast an increaseover the next 12 months in the volume of data“The first thing Ive done is doubled downon data strategy,effectively building adata platform and governance andc

73、apabilities around that.This gives usmore control over our data.Youllprobably find in a lot of companies thatthey bolted on many of theseacquisitions and have a lot of disparatesystems,which means disparate data thats a challenge.”CIO,manufacturing/food and beverage1,000-5,000 employees,UKExplore ou

74、r developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFthey use to develop their AI models,and justless than half(49%)are forecasting growth indata volumes of more than 25%.Perhaps morefundamentally,though,is that the challenge fororganizations that underinvest in datamanagement may co

75、me with new data-related pressures in particular,the mix ofdata types organizations are employing formodel training.The proportion of organizationsusing unstructured rich media and text datafor AI initiatives has increased since 2023,andoutdated data management technologiesmay prevent organizations

76、from deliveringthese projects meaningfully.What are AI leaders doingdifferently?Leaders are significantly less likelyto see storage and datamanagement as their primaryinhibitors,presumably becausethese companies have alreadyprioritized modernizing their dataarchitectures.By building a soliddata foun

77、dation at the outset,AIleaders have ensured thatvaluable pilots have a clear path todeliver at scale.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFGenerative AI hasrapidly eclipsedother AIapplicationsOrganizations have rushed to invest ingenerative AI,with interest o

78、utstripping longer-standing forms of AI.As the dust settles on thisexplosion of investment,a small cohort ofgenerative AI trailblazers has emerged.Theseorganizations have more widely integratedcapabilities and are seeing significantcompetitive benefits from the technologyaround new product developme

79、nt,enhancedinnovation and faster time-to-market.Thesecompetitive advantages are likely to grow asgenerative AI trailblazers set out to establish asignificant gap between themselves andothers,shaped by their investment andinfrastructural advantages.Key insights:88%of organizations are activelyinvesti

80、gating generative AI.24%already see generative AI ashaving graduated to an integratedcapability across their organization.The majority of these generative AItrailblazers see a“high”or“very high”impact from generative AI initiatives onincreasing the rate of innovation(79%),supporting new productintro

81、duction(76%)and improvingExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFGenerative AI is the force driving enterprise AIstrategies in 2024.The vast majority oforganizations(88%)are actively investigatinggenerative AI models to create net-new dataor content.This intere

82、st outstrips much longer-standing forms of AI such as prediction models(61%),classification(51%),expert systems(39%)and robotics(30%).Considering thatawareness of generative AI only blossomed inlate 2022 and enterprise-grade solutions arestill undergoing development,this is atestament to the percept

83、ion of itstransformative potential.Most of the surveyrespondents also expressed an interest inhypothetical artificial general intelligence models that can outperform humans across allcognitive tasks suggesting that manyorganizations have their eye on AIs evolvinghorizons.This interest is being conve

84、rted into investment.Generative AI budgets on average are set togrow from 30%of total AI budgets to 34%overthe next 12 months.Many senior executives arekeenly aware of the implications for thetechnology and see a need for an acceleratedinvestment roadmap.time to market(76%),among otherareas of compe

85、titive differentiation.“People can now see it and interact withit.Our board members are fairlytraditional manufacturing folks.Whenthey see ChatGPT,it becomes very realto them.Were now looking at fieldservice maintenance,our technicianswho go out there to service machines.How can they use GenAI to qu

86、icklyExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFGenerative AI adoption is progressing rapidly.As Figure 7 illustrates,a set of trailblazers 24%of organizations have already graduatedgenerative AI investments into scaled-upproduction capabilities.In contrast,11%ofc

87、ompanies have not invested in generative AI,29%are still experimenting with the technologyand 37%have generative AI in production butnot yet scaled.This is a remarkable level ofuptake for a technology that only made it intothe public consciousness with the launch ofChatGPT in November 2022.access hi

88、storical maintenance recordsand ask support questions in real time?”SVP/chief digital officer,manufacturing/industrial products5,000-10,000 employees,USWhat are AI leaders doingdifferently?Leaders are investing in a broaderrange of AI types with a greaterlikelihood of engaging withrobotics,expert sy

89、stems andclassification models alongsidegenerative AI.This expandedportfolio equips them to identifysolutions that can meet diversebusiness needs and fosters a moreholistic approach where thesetechnologies can be broughttogether.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTM

90、L to PDFFigure 7:Generative AI maturityand levels of investmentProportion of respondents at eachstage:Q.Which of these statements most accurately reflects theuse of generative AI at your organization?Base:All respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI

91、custom survey,2024.Organizations that have integrated andbroadly deployed generative AI experience awide range of benefits.Importantly,thesebenefits are commonly seen in areas thatdeliver competitive advantage.More thanthree-quarters(79%)of these trailblazers seegenerative AI as having a“high”or“ver

92、y high”impact on their rates of innovation,76%on theirtime to market,76%in supporting new productintroduction,74%on improvements to theirproduct or service quality,and 67%on theirproduct and/or service differentiation.Theselevels outstrip organizations that are less“AI-mature”and suggests that relat

93、ive adoption ofgenerative AI may shape industry winners andlosers.Organizations that fail to rapidly institutemeaningful generative AI projects could endup losing out to those that can.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFFuture levels of investment appear t

94、o besetting the scene for these trailblazers toamplify their advantage.Generative AItrailblazers have invested heavily to lead thepack.The average organization thatsuccessfully graduated generative AI to anintegrated capability had invested 44%of its AIbudget into generative AI,a significantly highe

95、rlevel of investment than other organizations.Companies at earlier stages of generative AImaturity invested 26%on average.Trailblazercompanies are set to expand generative AIbudgets further,continuing to out-invest lessAI-mature organizations.Generative AI trailblazers are moresophisticated in their

96、 enabling infrastructuresand strategies.They use a wider array ofvenues for AI model training and inference.More fundamentally,however,they take farmore factors into account when it comes to AIinfrastructure planning.They are more likely toplan their infrastructure considering security,AIaccelerator

97、 access,data privacy,scalability,customer support,and access to AI tools andframeworks than organizations that have not“We created a chatbot for internal usage,such as employees searching for theideal health plan.We are trying to extendthat to publicly available information earnings and annual repor

98、ts to look atour last five years and understand ourstrategy.If these models are workingwell,then we can go to moreunstructured data.”CIO,transportation/logistics/warehousing1,000-5,000 employees,USExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFinvested to the same deg

99、ree.The only factorthese trailblazers are less likely to account forthan those experimenting with generative AI isup-front costs,which they see as lessimportant than longer-term operatingexpenditures.By looking at theseconsiderations at the outset of infrastructuredecision-making,these organizations

100、 areensuring these issues do not emerge asprojects progress.What are AI leaders doingdifferently?Compared to other organizationsinvesting in generative AI,AIleaders have prioritized generativeAI initiatives that boost innovationrates and enhance IT efficiencies.By prioritizing these areas,theycreate

101、 a virtuous cycle whereincreased innovation facilitatesfurther use of generative AI,andstreamlined IT processes ensuresustainable and effective delivery.GPU availabilitycontinues to beconstrained,Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFshapinginfrastructuredeci

102、sion-makingAI accelerators play an important role inoptimizing the performance of AI.Thesespecialized hardware devices GPUs beingthe most prominent example are designedto accelerate model training and inference;they are faster and more efficient than CPUsfor AI workloads.Organizations can facechalle

103、nges accessing GPUs,and that scarcity iselevating their position in infrastructureplanning and is encouraging uptake ofspecialist AI cloud computing platforms.The leading infrastructure decision factorsrelate to security,AI accelerator access,andKey insights:After security,the leading factor ininfra

104、structure decision-making isaccelerator availability,identified by44%of organizations.Hyperscaler public clouds are onepathway to GPUs,but many are alsoturning to specialist AI clouds.GPUclouds are emerging as a key venue forboth training employed by almost athird,32%of organizations andinference,31

105、%.In some geographies,particularly inAsia-Pacific,lack of access to AIaccelerators is already limitingorganizations from moving models intoproduction.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFreliability and availability.As Figure 8 illustrates,access to AI accel

106、erators ranks highly,outstripping even long-standing areas ofconcern such as operating costs and flexibility.Telecommunications companies(53%),highereducation(53%)and manufacturingorganizations(51%)prioritize this accessparticularly strongly.Figure 8:Security and access toAI accelerators leadinginfr

107、astructure decision-factorsQ.Which factors most influence the AI infrastructuredecisions made at your organization?Base:All respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.Hyperscaler public clouds offer an importantavenue for organizatio

108、ns seeking GPUs,butthey are not the only game in town.WhileExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFhyperscaler cloud computing incumbentsrepresent the most popular venue for AItraining and inference,identified by 46%and40%of organizations,respectively,specialis

109、t AIclouds have emerged as an auxiliary,or evenalternative,venue.GPU clouds have seen anexplosion in popularity that reflects the highdemand for GPUs.Almost a third,32%,oforganizations that have invested in AI areexecuting training workloads using GPU clouds,and 31%for inference.These specialist clo

110、udofferings are particularly popular withinformation technology and servicescompanies,with 51%citing GPU clouds as avenue for training.As the landscape of AI development anddeployment widens,GPU clouds are poised tosee further growth.Organizations areanticipating an increase in training andinferenci

111、ng venues that will be employed overthe next 12 months,and in this growthenvironment,GPU clouds are projected to growto a 34%adoption rate for both inference andtraining.Higher education institutions appearto be a particularly fast-growing customercohort.Our data highlights scalability as beingthe p

112、rincipal role organizations see GPU cloudsas stepping into;the ability for organizations toeasily and cost-effectively manage fluctuatingAI workloads is a clear driver of adoption.GPU availability challenges are felt keenly byorganizations in some countries,including anumber of major Asia-Pacific ec

113、onomies;India,Taiwan,New Zealand and Australia are morelikely to rank GPU availability among their topthree challenges to bringing a model intoproduction.Sweden,where 39%identify it as atop three challenge,and UAE(35%)also standout in this regard.Figure 9:National differences inthe level of impact G

114、PUExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFavailability is having on bringingmodels into productionQ.What are the primary challenges or impediments tomoving an AI/ML application from proof of concept/pilotstages to production environments?Ranks 1,2 and 3.Base:Al

115、l respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.What are AI leaders doingdifferently?Leaders are more likely to leverageGPU clouds for both training andinference and are particularlyinvested in how the technologycan be used to reduce th

116、e time ittakes to bring AI initiatives live.Leveraging specialist cloud-basedservices to expedite thedevelopment process by securingaccess to scarce GPU resourcesappears to be a clear avenue todrive through an AI advantage.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to P

117、DFConcerns aboutAIs environmentalimpact persist butare not slowing AIadoption;sustainable AIpractices offeropportunity tomitigateemissionsThe story this year is about expansion:expanding scope of AI,expanding workloadrequirements,expanding data infrastructure.Inthis context,sustainability practices

118、are key.With many sustainability practices deliveringmeasurable value and the potential ofapplying AI to address energy consumption,there is a clear opportunity to address theemissions challenge.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFSustainability remains imp

119、ortant toorganizations,with almost two-thirdsconcerned about the impact AI/ML has onenergy use and carbon footprints,including25%that are“very concerned.”The topic is aninfluential factor in AI strategies,with 37%seeing sustainability as among the factors thatmost influence the AI infrastructure dec

120、isionsmade at their organization.As Figure 8illustrates,this is comparable with operatingcosts,data privacy and scalability.This focuson sustainability is important in the context ofexpanding data demands,as well as venuesfor training and inference,forecast for AIworkloads over the next 12 months.Th

121、e most identified(by 42%of respondents)sustainability approach that organizationshave embarked on in the past 12 months isinvesting in more energy-efficient IT hardware.Notably,sustainability ranks as the secondmost common primary role for GPU clouds,justKey insights:Nearly two-thirds 64%of responde

122、ntssay that their organization is“concerned”or“very concerned”withthe sustainability of AI infrastructure.Popular sustainability measuresinclude investing in energy-efficient IThardware(42%of organizations),increasing investment in AIgovernance(40%),providing trainingand education on sustainability(

123、37%)and establishing sustainabilityguidelines(35%).Nearly one-third(30%)of respondentssay reducing energy consumption is adriver for AI/ML adoption at theirorganization.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFbehind scalability.More efficient resourceallocation

124、 and a commitment by manyleading GPU cloud providers to employ moreeco-friendly datacenters appears to alignclosely with this efficiency initiative.In addition to exhibiting a clear appetite toinvest in better optimized technology,manyorganizations are increasing investment in AIgovernance(40%),prov

125、iding training andeducation on sustainability(37%),andestablishing sustainability guidelines(35%).Less popular are initiatives that wouldrepresent rolling back AI investments.Just 5%say they have canceled AI initiatives in the past12 months,and 19%report they have changedproject scope.Organizations

126、largely find sustainabilitypractices to be effective.When identifying theprojects that are more likely to deliver highimpact than minor or no impact,we see thatpractices are more likely to deliver impact thannot,including those that are sometimesoverlooked(see Figure 10).While just 19%oforganization

127、s had engaged in changing AIproject scope to address environmentalconcerns in the past 12 months,57%of those“In terms of ESG,its really around how wemeasure,drive and achieve our targets,through leveraging digital and thenfinance.Its about end-to-end processefficiency and optimization.It is abouthow

128、 to automate things to drive back-office efficiencies it is top of mind.”CIO,manufacturing/food and beverage1,000-5,000 employees,UKExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFthat did see it as“highly”or“very highly”impactful.Another practice that may representsig

129、nificant opportunity is changinginfrastructure vendor a step taken by 27%but considered the most impactful practiceoverall.Figure 10:Sustainability stepsassessed by impactQ.What impact have you seen from the following actionson the environmental impacts of your AI projects?Base:Respondents from orga

130、nizations that had invested instep over past 12 months,all respondents(n=1,519).Source:S&P Global Market Intelligence 451 Research GlobalTrends in AI custom survey,2024.A number of organizations are alreadyapplying AI to energy consumption countering the resource-intensive nature of AIworkloads with

131、 AI-driven efficiency benefits.Nearly onethird(30%)of organizations treatreducing energy consumption and carbonfootprint as a key driver for developing AI/MLapplications.While currently the least popularobjective for AI/ML applications,it is notablethat almost a third of organizations mayalready be

132、applying AI to better assess andpredict emissions or to inform efficiencyimprovements that could contribute to energysavings.As workload pressures expand,weexpect this proportion to increase.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFThe drive to reducing energy c

133、onsumption isnot just about meeting sustainability goals.While 11%of respondents see meetingenterprise sustainability goals as their priorityin reducing energy consumption,that was lessthan improved operational efficiency(13%)andachieving cost savings(12%).Concerns aboutregulatory compliance(9%)and

134、customerexpectations(9%)are also apparent.In someinstances,this may give sustainabilityadvocates the ability to address sustainabilityimpacts while couching the value in thecontext of wider business goals.What are AI leaders doingdifferently?AI leaders are taking more steps toaddress the environment

135、al impactof their projects and are more likelyto see these steps as“highly”impactful.In particular,AI leadersare significantly more likely toconsider changing training orinferencing venues over the next 12months,establish sustainabilityguidelines and implementoffsetting methods.By engagingwith multi

136、ple steps in parallel,thevalue of each appears to bemagnified.While an emissionsimpact assessment may be useful,it delivers more value when thatassessment can be converted intodecisions about choice of vendorpartner or AI training or inferencingenvironments.Explore our developer-friendly HTML to PDF

137、 APIPrinted using PDFCrowdHTML to PDFConclusionsThe 2024 Global Trends in AI report representsa very different AI-adoption landscape thanthe 2023 edition.AI is gaining more widespreadimplementation,with greater focus ondelivering product and service qualityimprovements and revenue growth.Thematurati

138、on of generative AI is a key driver ofthis transition.Yet challenges remain.Manyorganizations are struggling to shiftinvestments into capabilities that they candeliver at scale,and they acknowledge there ispressure on the sustainability of businessoperations.Five key action points forbusinesses:Buil

139、d a robust data architecture forAI success:Organizations mustestablish a clear pathway forscaling AI projects into production,ensuring efficient datamanagement and storage.It iscrucial to invest in a strong datafoundation before committing tohigh volumes of pilot projects.Thiswill help enable seamle

140、ss AI valuedelivery.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFSmart investment is key togenerative AI success:Organizations benefiting fromgenerative AI have redistributedtheir budgets to focus on theseinitiatives.Success hinges onsophisticated decision-making an

141、drobust infrastructure.To emulatethis,organizations should ensurecomprehensive purchasingpractices and maximize GPUavailability and utilization,includinginvestigating specialized GPU and AIcloud services.Explore generative AI-driven ITefficiencies:Generative AI canautomate routine modeldevelopment t

142、asks and improve ITdecision-making to drive morestreamlined delivery.This self-reinforcing approach can underpina more sustainable AI roadmap.Expand sustainability practices:Changing infrastructure vendors orrevising AI project scopes can havea meaningful impact on overallemissions.The value ofsusta

143、inability measures iscompounded when they are usedtogether,so organizations shouldempower project teams to engagewith a variety of approaches.Build a holistic AI strategy:Generative AI offers significantopportunity,but organizationsshould create a well-rounded AIstrategy.A narrow approach to AI one

144、that that fails to investigate amix of technologies neglects theopportunity to bring togetherdifferent types of models andExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFcloses off many high-impact usecases.MethodologyThe findings presented in this report draw on asurv

145、ey fielded in North America,Europe,theMiddle East and Africa,and Asia-Pacific in Q22024.The survey targeted 1,519 AI/ML decision-makers/influencers,filtering for respondentswith AI/ML deployed in pilots and productionenvironments across the following industries:aerospace and defense,automotive,energ

146、y/oil and gas,finance,government,healthcare,higher education,IT and services,life sciences,manufacturing,media/entertainment,telecommunications,transportation and logistics,and utilities.Themost common respondent job roles were IToperations leadership,IT infrastructureleadership and executive manage

147、ment.Thisreport also draws on contextual knowledge ofExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFadditional research conducted by S&P GlobalMarket Intelligence 451 Research.Brought to you byWEKA helps enterprises and researchorganizations achieve discoveries,insigh

148、ts,andoutcomes faster by improving theperformance and efficiency of GPUs,AI,andother performance-intensive workloads.WEKA Turbocharges AI Workloads with an AI-Native Data PlatformWEKA Drives GPU AccelerationWEKA for Generative AIAbout the authorsExplore our developer-friendly HTML to PDF APIPrinted

149、using PDFCrowdHTML to PDFAlex JohnstonSenior ResearchAnalyst,Data,AI&AnalyticsAlex Johnston is asenior researchanalyst on the 451Research Data,AI&Analytics team at S&PGlobal MarketIntelligence.Hefocuses on emergingtechnologies and howthey can be applied inbusiness contexts.Alexs primarycoverage area

150、s areartificial intelligence,distributed ledgertechnology and eventstream processing.Alexs recent areas ofconcentration includemonitoring theemerging generativeAI market,tracking theevolution inblockchain use casesand investigating real-time architectures.DavidImmermanConsultingAnalystDavid Immerman

151、 is aconsulting analyst atS&P Global MarketIntelligence.He isresponsible forExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFAbout this reportA Discovery report is a study based on primary researchsurvey data that assesses the market dynamics of a keyenterprise technolo

152、gy segment through the lens of the“on the ground”experience and opinions of realpractitioners what they are doing,and why they aredoing it.About S&P Global MarketIntelligenceAt S&P Global Market Intelligence,we understand theimportance of accurate,deep and insightful information.Our team of experts

153、delivers unrivaled insights andleading data and technology solutions,partnering withcustomers to expand their perspective,operate withconfidence,and make decisions with conviction.S&P Global Market Intelligence is a division of S&P Global(NYSE:SPGI).S&P Global is the worlds foremost providerof credi

154、t ratings,benchmarks,analytics and workflowsolutions in the global capital,commodity andautomotive markets.With every one of our offerings,weexecuting on a rangeof custom researchinitiatives anddevelopment ofthought leadershipacross technologysectors includingindustrial IoT,digitaltransformation,edg

155、ecomputing,AI/machine learningand fintech,amongothers,and verticalssuch asmanufacturing andautomotive.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFhelp many of the worlds leading organizations navigatethe economic landscape so they can plan for tomorrow,today.For mo

156、re information, 800 447 2273Japan:+81 3 6262 1887Asia Pacific:+60 4 291 3600Europe,Middle East,Africa:+44(0)134 432 2024 by S&P Global Market Intelligence,adivision of S&P Global Inc.All rights reserved.These materials have been prepared solely forinformation purposes based upon information generall

157、yavailable to the public and from sources believed to bereliable.No content(including index data,ratings,credit-related analyses and data,research,model,software orother application or output therefrom)or any partthereof(Content)may be modified,reverse engineered,reproduced or distributed in any for

158、m by any means,orstored in a database or retrieval system,without the priorwritten permission of S&P Global Market Intelligence or itsaffiliates(collectively S&P Global).The Content shall notbe used for any unlawful or unauthorized purposes.S&PGlobal and any third-party providers(collectively S&PGlo

159、bal Parties)do not guarantee the accuracy,completeness,timeliness or availability of the Content.S&P Global Parties are not responsible for any errors oromissions,regardless of the cause,for the resultsobtained from the use of the Content.THE CONTENT ISPROVIDED ON“AS IS”BASIS.S&P GLOBAL PARTIES DISC

160、LAIMANY AND ALL EXPRESS OR IMPLIED WARRANTIES,INCLUDING,BUT NOT LIMITED TO,ANY WARRANTIES OF MERCHANTABILITYOR FITNESS FOR A PARTICULAR PURPOSE OR USE,FREEDOMFROM BUGS,SOFTWARE ERRORS OR DEFECTS,THAT THECONTENTS FUNCTIONING WILL BE UNINTERRUPTED OR THATExplore our developer-friendly HTML to PDF APIP

161、rinted using PDFCrowdHTML to PDFTHE CONTENT WILL OPERATE WITH ANY SOFTWARE ORHARDWARE CONFIGURATION.In no event shall S&P GlobalParties be liable to any party for any direct,indirect,incidental,exemplary,compensatory,punitive,special orconsequential damages,costs,expenses,legal fees,orlosses(includi

162、ng,without limitation,lost income or lostprofits and opportunity costs or losses caused bynegligence)in connection with any use of the Contenteven if advised of the possibility of such damages.S&P Global Market Intelligences opinions,quotes andcredit-related and other analyses are statements ofopini

163、on as of the date they are expressed and notstatements of fact or recommendations to purchase,hold,or sell any securities or to make any investmentdecisions,and do not address the suitability of anysecurity.S&P Global Market Intelligence may provideindex data.Direct investment in an index is not pos

164、sible.Exposure to an asset class represented by an index isavailable through investable instruments based on thatindex.S&P Global Market Intelligence assumes noobligation to update the Content following publication inany form or format.The Content should not be relied onand is not a substitute for t

165、he skill,judgment andexperience of the user,its management,employees,advisors and/or clients when making investment andother business decisions.S&P Global keeps certainactivities of its divisions separate from each other topreserve the independence and objectivity of theirrespective activities.As a

166、result,certain divisions of S&PGlobal may have information that is not available toother S&P Global divisions.S&P Global has establishedpolicies and procedures to maintain the confidentiality ofcertain nonpublic information received in connectionwith each analytical process.S&P Global may receive co

167、mpensation for its ratings andcertain analyses,normally from issuers or underwriters ofsecurities or from obligors.S&P Global reserves the rightto disseminate its opinions and analyses.S&P Globalspublic ratings and analyses are made available on itswebsites,(free of charge)and (subscription),and may

168、 bedistributed through other means,including via S&PGlobal publications and third-party redistributors.Explore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDFAdditional information about our ratings fees is availableat Analyst ReportExplore our developer-friendly HTML to PDF APIPrinted using PDFCrowdHTML to PDF