2024半导体产业内幕报告-关键趋势和专家见解（英文版）-GLG格理集团 - 副本

JULY2024

INSIDETHE

SEMICONDUCTORINDUSTRY

KeyTrendsandExpertInsights

Byasubject-matterexpertandGLGNetworkMember

TableofContents

EXECUTIVESUMMARY

1

ABOUTTHERESEARCH

2

SURVEYINSIGHTS

3

CONCLUSION

12

EXECUTIVESUMMARY

Theincreasingperformancerequirementsofhyperscaledatacenters,risingdemandforsupportofAIworkloadsonarangeofdevices,andacceleratinginterestinprocessingattheedgearedrivingnewopportunitiesandincreasingchallengesforthesemiconductorindustry.Thissurveyofsemiconductorleadersrevealstrendsthatwillpowerfuture

demandforsemiconductordesignandfabricationaswellasmakenewcapabilities

availabletothedeviceandequipmentmanufacturersthataretheprimaryconsumersofsemiconductors.

Findingsfromthesurveypointtoafutureofgreaterspecializationofthesemiconductorindustry,aspurpose-builtchipsareincreasinglydeployedtooptimizeperformanceforhighlyspecificworkloadsandindustryapplications.Softwareecosystemsareexpectedtoplayanevenmoreimportantroleinsimplifyingthedeveloperexperienceacrossan

increasinglyfragmentedhardwarelandscape.

Whilethesemiconductorsupplychainhasrecoveredsubstantiallysincethepandemic,changingglobalconditions,greateremphasisonsustainability,andincreasingdomesticinvestmentinmanufacturingcapacitycontinuetoimpactthestrategicchoicesavailabletosemiconductorcompanies.Facinglongproductcyclesandhighcostsofnew

productdevelopment,firmsinthisspacewillwanttocarefullyassessthesetrendsandconducttheirownfollow-onresearchtoconfidentlyidentifythemostpromisingmarketopportunities.

Thesurveyrevealedthat:

74%

ofrespondentsbelieve

thatacceleratingAIat

theedgeisoneofthetop

threegrowthmarketsfor

semiconductors

65%

ofrespondentsexpect

thatAI-specificchips

willovertakeGPUs

asacceleratorsofAI

workloadswithinfourto

sevenyears

90%

ofrespondentsviewan

open-sourcesoftware

ecosystemasimportant

tothemarketopportunity

forasemiconductor

productline

SemiconductorIndustry•1

ABOUTTHERESEARCH

FromMay23-June3,2024,GLGconductedanonlinesurveyof101seniorleadersintheUnitedStatesandCanadawithinfluenceoversales,go-to-marketplanning,orstrategyforasemiconductorfirmorequipmentmanufacturertocapturetheir

perspectivesonkeyareasofopportunityandconcern.

Thestudyrevealedthatmostrespondents:

•ViewAIasasignificantdriverofsemiconductoropportunity

•Expectincreasingspecializationandverticalizationofsemiconductorsasdemandforoptimizationaroundspecificworkloadsandusagescenariosincreases

•Believethatdomesticinvestmentsinsemiconductorcapacityhavehadapositiveimpactonthecompetitivenessandinnovationofsemiconductorfirms

Thefiguresbelowindicaterespondents’rolesandthedistributionoftheircompaniesbyannualrevenueandindustry.

PrimaryIndustry

Equipmentmanufacturer

Fablesssemiconductormanufacturer

Integrateddevicemanufacturer(IDM)

Semiconductorfabrication plant(foundry)Industrialautomation

company Outsourcedsemiconductorassemblyandtestprovider(OSAT)

43%

21%19%

7%5%I4%I2%

34%30%

17%12%

7%

1%

JobFunction

Executive/GeneralManagement

Sales

R&D/Engineering/ProductDevelopment SupplyChain/DemandManagement

Manufacturing

Design/Engineering

InformationTechnology

CompanySize

Fewerthan500

501-2,500

2,501-5,000

5,001-10,000

10,001-50,000

Morethan50,000

15%

30%

12%21%19% 4%

Revenue

$1B+

$101M-$1B

$50M-$100M

53%

36%

11%

SemiconductorIndustry•2

SURVEYINSIGHTS

90%

ofrespondentsrankAI

workloadsasatopgrowth

areafortheindustry

AIworkloadswilldrivedemand.

RespondentsexpectAIworkloadstobeanincreasinglyimportantsourceofdemandforsemiconductors,with90%ofrespondents

rankingAIworkloadsasoneoftheindustry’stopgrowthareas.Respondentsalsoindicatethat

thoseworkloadsarelikelytorunbothwithindata

centersandattheedge,sofirmsplanningtopursueAIopportunitiesshouldconsider

investinginproductlinesthataddressboththehigh-performanceandscaledemandsofdatacentersandthelowpowerconsumptionrequiredforedgedevices.Amajority

ofrespondentsalsoindicatedthatindustry-specificchips,suchasthoseusedincontrolsystemsintheindustrialorautomotiveindustriesorindevicesinthelifesciencesand

healthcaresectors,willbeanimportantdriverofsemiconductormarketgrowth.

TakingamoredetailedlookatthenatureofAIworkloadsthatneedtobesupported,72%ofrespondentsidentifieddataprocessingandmovementinsupportofAIworkloadsas

oneofthetopthreeindustryopportunities,and65%ofrespondentssimilarlyhighlighted

inferencingusedtosupportlargelanguagemodels(LLMs).Amajority(52%)alsoidentifiedvideoprocessingandmediaencodingasakeyopportunity.

ExpectedGrowthMarketofSemiconductorCompanies

%AmongTotalRespondents,SortedDescendingbyTotalRanked

Rank1–Highest-GrowthMarket

Rank2

Rank3

Rank4

Rank5

DedicatedchipsforacceleratingAIworkloadsindatacenters

DedicatedchipsforacceleratingAIworkloadsontheedge

Optimizedchipsforhigh-performancecomputing

Industry-specificchips(e.g.,automotiveorhealthcareindustries)

Optimizedchipsforhigh-speednetworking

SensorsandotherdedicatedchipsforsupportofInternetofThings(IoT)applications

Generalizedprocessorsorgeneral-purposechips(e.g.,CPUs)

Dedicatedchipsforsecuritytasks(e.g.,encryption)

Dedicatedchipsforperformingcalculationsrelatedtoblockchains

39%23%9%15%5%

25%32%17%10%7%

15%12%20%15%12%

8%8%12%14%12%

4%11%11%9%14%

4%5%10%13%16%

4%12%13%11%

2%

3%7%9%15%

1%

5%9%

1%3%3%

TotalRanked

90%

90%

73%

53%

49%

48%

42%

35%

21%

SemiconductorIndustry•3

OpportunityforChipsforAcceleratingAIWorkloads

%AmongTotalRespondents,SortedDescendingbyTotalRanked

Rank1-HighestOpportunity

Rank2

Rank3

Rank4

Rank5

Dataprocessingormovement

Inferencingforlanguageprocessing/LLMs

Imageclassification

Videoprocessing/Mediaencoding

Networktrafficmanagementorcoordination

Training

26%22%24%14%10%

40%16%9%16%14%

9%19%20%25%19%

8%19%25%23%16%

7%13%15%20%17%

11%12%8%24%

3%

TotalRanked

96%

95%

92%

91%

72%

58%

GPUgiveswaytoAI-specificchips.

Whenaskedtoprojecthowdemandfordifferenttypesofchipswouldevolveover

time,respondentsindicatedthatwhileGPUswillbetheprimarymeansofaccelerating

AIworkloadsinthenearterm,withinfourtosevenyearsthatrolewouldlikelyshiftto

AI-specificchipssuchasNPUs.Sixty-fivepercentofrespondentsindicatedthatthese

AI-specificchipswouldberunningthemajorityofAIworkloadsinthattimeframe,whileonly18%ofrespondentsbelievedthatGPUswouldretaintheircurrentsignificance.It

isprojectedthatavarietyofAI-specificchipswillbecomeavailabletomeetarangeofrequirementsfordifferentoperatingenvironments,suchaslow-powerAIacceleratorsformobiledevicesandhighlyperformantchipsoptimizedtorunlargelanguagemodelsasaserviceinthecloud.

AnticipatedChipUseinFuture

%AmongTotalRespondents

Nearfuture(1-3years)Mediumterm(4-7years)Longterm(morethan8years)

4%6%9%

CPUs

21%

7%8%

ASICs

63%

18%

6%

GPUs

2%1%2%

FPGAs

65%

61%

23%

AI-specificchips

1%2%1%

Other

SemiconductorIndustry•4

63%

ofrespondentsbelievethat

semiconductorsembedded

indevices,vehicles,and

industrialequipment

willdrivethemajorityof

semiconductoropportunity

withinfourtosevenyears

Fromdatacenterstodevices.

Respondentstothesurveyalsoexpectashiftovertimeinwhichoperatingenvironments

consumethemostsemiconductors.While

56%ofrespondentsbelievethatdatacenters,boththosethatsupportpubliccloudsand

thosemanagedon-premiseorinsupportof

privateclouds,aretheprimarysemiconductoropportunityoverthenextthreeyears,inthe

four-to-seven-yearhorizontheserespondentsexpectthatsituationtoshift,with63%of

respondentsanticipatingthatsemiconductorsembeddedinmobiledevices,vehicles,andindustrialequipmentwilldrivethemajorityofindustryopportunity.

DevicesorPlatformsAnticipatedtoDriveDemandforChipsinFuture

%AmongTotalRespondents

Nearfuture(1-3years)Mediumterm(4-7years)Longterm(morethan8years)

21%17%15%

Mobiledevices

34%

17%18%

Publicclouds/

Hyperscale

environments

22%

14%

6%

On-premisedata

centers/Privateclouds

26%

17%

11%

Industrialequipment

24%22%

11%

Vehicles

10%10%

2%

Medical/Scientificdevices

SemiconductorIndustry•5

Verticalizationiskeytogrowth.

JustasAIworkloadsareexpectedtobesupportedwithincreasinglyspecializedand

purpose-builtchips,respondentsindicatedthatthedemandsofspecificindustrieswillpushsemiconductormanufacturerstofindpurpose-built,verticalizedsolutionsfortheirneeds.Respondentssuggestthatthelargestbusinessopportunitiesforsemiconductormanufacturersarelikelytocomefromaddressingtheuniqueneedsoftheautomotive,telecommunications,andhealthcare/lifesciencessegments,with69%placingthe

automotiveindustryintheirtopthreeopportunities,51%indicatingtelecommunications,

and48%indicatinghealthcareandlifesciences.Allthreeindustriessharedemandfor

chipsthatcansupportlocalprocessingwithlimitedtoleranceforlatencyandconstraints

aroundpowerconsumption,suggestingthatrawperformancemaybecomelessimportantovertimethantheabilitytoprocesseventsinrealtimeacrossdistributeddevices.

IndustriesExpectedtoDriveOpportunity

%AmongTotalRespondents,SortedDescendingbyTotalRanked

Rank1-Mostlikely

Rank2

Rank3

Rank4

Rank5

Automotive

Healthcareandlifesciences

Telecommunications

Industrialmanufacturing

11%

Mediaandentertainment

Energyandutilities

Financialservices

Retail

Resourceindustries(agriculture,mining,andforestry)

28%24%17%10%12%

12%20%16%22%18%

23%12%16%13%9%

10%12%19%16%14%

19%11%10%9%9%

13%

11%

18%

2%

4%8%6%9%19%

2%3%2%

4%5%

1%1%

TotalRanked

91%

88%

73%

71%

58%

55%

46%

11%

11%

SemiconductorIndustry•6

Softwaremakeshardwaremorevaluable.

Increasingspecializationofhardwareplatformstosupportspecificworkloadprofiles

andindustryapplicationswillmakeitmoredifficultfordeveloperstooptimizetheir

softwareforeachparticularchip.Byinvestingintheopen-sourcesoftwareecosystem,

semiconductormanufacturerscanmakeiteasierfordeveloperstoshareoptimizations

andbuildmoregeneralizedframeworksthatenableapplicationstomakethebestuseofspecifichardwareconfigurations.Ninetypercentofrespondentstothesurveyindicatedthatthehealthoftheopen-sourcesoftwareecosystemaroundaparticularchipwas

importanttothemarketopportunityforthatchip.Inparticular,respondentsbelieved

thatopen-sourcesoftwaremadeiteasierfordeveloperstoobtainperformancebenefits,improveinteroperability,increaseenergyefficiency,andmaintainsecuritywhenworkingwithaparticularchip.

BenefitsofThrivingSoftwareEcosystem

Somewhatimportant

%AmongTotalRespondents,SortedDescending

Performanceoptimization

Interoperability/Compatibility

Energyefficiency/Heatreduction

Firmwareandsecurityupdates

Designandsimulation

Customization

67%

51%

50%

49%

39%26%

ImportanceofOpen-SourceSoftwareEcosystem

%AmongTotalRespondents

Veryimportant

Notveryimportant

45%

45%

11%

As-a-servicemodelsexpandopportunity.

Whilepurpose-builtchipswillenableoptimizationofworkloadsforthebest

combinationofperformance,cost,andpowerconsumption,theproliferationof

hardwareoptionswillincreasethecomplexityofmanagingthoseworkloadsand

maintainingenvironmentstosupportthem.Aswiththeshifttocloudforgeneralized

computeandstorageworkloads,companieswilllooktoas-a-servicemodelstomanageincreasinglymorespecializedworkloadsontheirbehalf,andmorespecificallytheywilllooktoprovidersthatofferdedicatedhardwareoptimizedfortheirparticularneeds,

ratherthandependingongeneral-purposecloudenvironments.Respondentstothesurveybelievethatinthenearterm(withinthreeyears),amajorityofcompanieswillshifttorunningkeyworkloadsindedicatedas-a-serviceenvironments;forexample,65%ofrespondentsbelievethatasignificantportionofAItrainingandinferencing

SemiconductorIndustry•7

workloadswillberunindedicatedGPU-as-a-serviceenvironmentswithinthree

years.Thesenewdedicatedmanagedserviceenvironmentswillbecomeincreasinglyimportantcustomersforsemiconductormanufacturers,drivingdemandandscaleforspecifictypesofworkload-orapplication-specificchips.

AnticipatedCustomersShiftingbyAlternativetoPublicCloudComputingApproach

%AmongTotalRespondents,SortedAscendingby‘Customerswillneveruse’

Mostofourcustomersarealreadyusingas-a-servicecomputationNearfuture(1-3years)Mediumterm(4-7years)

Longterm(morethan8years)Customerswillneveruseas-a-servicecomputation

GPU-as-a-serviceforAItraining/Inferencing

Managingedge/IoTcomputingendpoints

Imageprocessingorclassification

Industry-specificapplications

Networkmanagement/Optimization

CPU-as-a-serviceforhigh-performancecomputing

Videoprocessing/Encoding

Securitytasks(e.g.,encryption)

Blockchaintransactions

9%

25%

17%

48%

2%

18%45%22%12%4%

22%43%23%8%5%

19%31%33%12%6%

24%44%22%4%7%

18%34%28%13%8%

23%39%26%5%8%

34%33%15%7%12%

16%22%32%17%14%

Sustainabilityactionsspeaklouderthanwords.

Asmarketdemandforsustainablecomputingincreases,andmorepressureisputonthe

industrytomanagewasteandreducepowerandwaterconsumption,semiconductormanufacturersmusttakeactiontoimprovetheirsustainability;90%ofrespondentstothesurveyindicatedthatitwasimportantformanufacturerstoimprovetheir

ownsustainability,and86%suggestedthatthisresponsibilityextendstohelpingendcustomersbecomemoresustainableaswell.Theincreasedawarenessaroundsustainabilityalsomeansthatmanufacturersmustmakethemarketawareoftheir

investmentsinsustainability.Whenaskedtoidentifyonekeystrategyforengagingthemarketaroundsustainabilityissues,thelargestportionofrespondents(39%)suggestedthatsemiconductormanufacturerswerebestservedbydevelopingtheirownreputation

SemiconductorIndustry•8

forsustainablepracticesratherthanmarketingsustainabilitydirectlytocustomers.

Bytakingactionstomanagetheirownenvironmentalfootprintsandscrutinizingtheir

supplychains,firmscanmodelsustainablepracticesandbuildpositiveperceptioninthemarketplacethatcantranslateintogreaterdemandfortheirproducts.

ImportanceofCompaniesImprovingSustainability

%AmongTotalRespondents

Veryimportant

Somewhatimportant

Notvery

important

Notatallimportant

Importancethatcompaniesimprovetheirownsustainabilitybyreducingwaste/energyconsumption

Importancethatcompanieshelpcustomersbemoresustainable

8%

38%

52%

2%

41%45%11%4%

PreferredPrimaryStrategyofCompaniestoCommunicateValueofSustainabilitytoEndCustomers

%AmongTotalRespondents,SortedDescending

Buildthesemiconductormanufacturer’sownreputationforsustainabilityratherthantryingtomarketsustainabilitydirectlytoendcustomers

Emphasizecostsavingsfromusingmoreefficientproducts

Emphasizethepositiveimpactsusingsustainableproductscanhavetoacustomer’sreputation

Emphasizethereductionofharmandpotentialriskassociatedwithusinglesssustainableproducts

Semiconductorsshouldnotmarketorpromotesustainability

26%20%11%

5%

39%

Domesticinvestmentsinsupplychaincapacityareexpectedtohaveapositiveimpact.

Whenaskedtoidentifythemosturgentriskstothesemiconductorindustry,59%of

respondents(largelyU.S.-based)identifieddependenceonnon-U.S.-basedfabricationasakeyconcern,acknowledgingboththehighconcentrationofOSAT(outsourced

semiconductorassemblyandtest)providersinAsiaandthepotentialforgeopoliticaltensionsorseismicactivitytodisruptfoundryoperationsinTaiwan,while46%

SemiconductorIndustry•9

highlightedthelackofrelevantskillsinthelaborforce.Seventy-threepercentof

respondentssaidtheyexpectedtheUSCHIPSActtoimpactthehealthofthedomesticsemiconductorindustry,with53%ofthoserespondentsindicatingtheactwould

improvecompetitivenesswithnon-U.S.-basedfirms,51%suggestingthatitwould

helpaccelerateinnovationanddevelopmentofnewtechnologies,and35%expecting

thatitwouldincreasetheavailabilityofskilledlabor.Whenconsideringthestateof

thesemiconductorsupplychain,58%ofrespondentsindicatedthattheyweremore

confidentabouttheirabilitytoobtainrawmaterialssincethebeginningofthepandemic,but79%indicatedthattheystillhadconcernsaboutavailabilityofthesematerialsoverthenextthreeyears.

MostUrgentSemiconductorIndustryProblems

%AmongTotalRespondents

Sustainability

19%

Energyconsumption

7%

Pollution/Wastemanagement

35%

36%

59%

SupplyChainDependenceonrareearthelements

andotherdifficult-to-obtainmaterials

Industrydependenceonasmallnumberofkeyplayers

Industrydependenceonnon-U.S.-basedfabrication

Security

31%

Cybersecurity/Industrialespionage

34%

Supplychainsecurity

46%

Capacity

Lackofskillsinlaborforce

30%

LackofR&Dinvestment/Fundamentalresearch

Other(5%)

•Packing

•Actualmachinesthatmakethesemiconductor.Chinahasinvestedhardintheglobalsupply

•Laborcosts

•AllOSATsinAsia

•Politicaldecisions,suchastariffs,impactingsupplychainsandavailability

SemiconductorIndustry•10

ExpectedImpactofCHIPSActonSemiconductorIndustry

%AmongTotalRespondents

Agreatdeal

Some

AlittleNone

34%

39%

24%

4%

ConcernwithRawMaterial

AvailabilityinNextThreeYears

%AmongTotalRespondents

Veryconcerned

Somewhatconcerned

NotveryconcernedNotatallconcerned

16%

63%

19%

2%

ContributionofCHIPSActonHealthofSemiconductorIndustry

%AmongRespondentsConcernedwithAvailability,SortedDescending

53%

Improvecompetitiveness

51%

Accelerateinnovation/Developmentofnewtechnologies

42%

Improveefficiencyofsupplychain

40%

Createmarketopportunitiesforsmallerfirms

35%

Increaseava