MS - Latest thoughts on Deepseek and other concerns; why we remain positive on AI semispdf
MS - Latest thoughts on Deepseek and other concerns; why we remain positive on AI semispdf
Page 1 of 230
Tyler Durden shared this file. Want to do more with it?
  1. M UpdateMorgan Stanley & Co. LLCJoseph MooreEquity AnalystJoseph.Moore@morganstanley.com+1 212 761-7516Mason WayneResearch AssociateMason.Wayne@morganstanley.com+1 212 761-6012Ella TulchinskyResearch AssociateElla.Tulchinsky@morganstanley.com+1 212 761-2222Shane BrettEquity AnalystShane.Brett@morganstanley.com+1 212 761-1022SemiconductorsNorth AmericaIndustry ViewAttractiveSemiconductors | North AmericaLatest thoughts on Deepseek and other concerns; why we remain positive on AI semisThe DeepSeek release highlights evolutionary innovations in AI, some of which may be deflationary. That said, the stock market reaction is probably more important than the cause, and could bring further export controls or reduce spending enthusiasm; trimming PTs but remain positive.What’s Changed (NVDA.O)FromToPrice Target$166.00$152.00Astera Labs Inc (ALAB.O)Price Target$142.00$114.00Marvell Technology Group Ltd (MRVL.O)Price Target$120.00$113.00Broadcom Inc. (AVGO.O)Price Target$265.00$246.00Micron Technology Inc. (MU.O)Price Target$98.00$91.00The magnitude of the reaction to DeepSeek's release was surprising to us. While Deepseek R1 is an interesting product that by all accounts will be competitive, that's not all that surprising to us. We expected China to be competitive in AI, as they have been spending quite a bit of money developing AI and export controls cannot prevent those investments, and DeepSeek has been prominent in those discussions since at least mid last year.That said, there are several questions raised by this:Is China competing in AI without the access to higher volume GPUs? Broadly speaking, the answer to this is no. We believe that NVIDIA has sold well over $10 bn of H20s (a specialized product designed for China) in the last 12 months, and AMD has also done well with products that meet performance limitations. While export controls have prohibited sales of higher end GPUs into China, and we believe that our covered companies have scrupulously followed those rules, Chinese companies can train models outside of China, and have done so. But Deepseek was developed on only $6 million of hardware? Importantly, that $6 million per the Deepseek report does not include "costs associated with prior research and ablation experiments on architectures, algorithms, and data", which is an important caveat. It is still by all accounts an efficient design. In an NVDIA release, the company commented that "DeepSeek's work illustrates how new models can be created using that technique, leveraging widely-available models and Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.For analyst certification and other important disclosures, refer to the Disclosure Section, located at the end of this report.January 28, 2025 02:49 AM GMT
  2. M Update2compute that is fully export control compliant," which could imply some level of distillation.The innovations behind this efficiency were said to be such as training in FP8, multi token prediction, MLA, custom PTX are all important efficiency techniques - some of which we have learned about through presentations from NVIDIA in the last two years. Of course, the power of those tools is in the way they are built. By all accounts DeepSeek will be a viable competitor in the AI space, but we would point out that certain techniques touted as breakthroughs including the usage of MLA, FP8 training, and the Group Relative Policy Optimization (GRPO) reinforcement learning framework date back to the 6 month old DeepSeek-V2 and even older DeepSeekMath math model. Techniques like training in FP-8 and using multitoken prediction have been discussed by western AI players like MSFT (here) and Meta (here) in the research literature. We think it's unlikely that Deepseek has brought to market performance enhancing technology unknown to large hyperscalers, and if they have it likely falls into a long history of algorithmic breakthroughs that have only meant an expansion of compute demand. Nvidia has said they have improved single chip inference performance 1000x in the last decade, with model efficiency gains that have outpaced that. As long as scaling laws continue algorithmic performance gains shouldn't lower compute demand. We can see this idea in the following quote from the invention of the transformer architecture: "our base model surpasses all previously published models and ensembles, at a fraction of the training cost of any of the competitive models" (here). Isn't this going to change spending behavior for everyone in AI? We don't think so. We spent a good part of the weekend and Monday talking to industry contacts who have spent time analyzing the technology. Thus far, the feedback that we have heard is that there are no obvious shortcuts that are made clear by this situation. The techniques involved in creating a more selective parameter set, and other elements of the implementation are well worn ground. By all accounts the DeepSeek technology is impressive technology and will be a viable competitor, but does not seem to be impacting the planning process. While conceding that it might take more than a few days to get a fully informed response, we generally heard the same comments over and over: "This will not change our spending plans".... "There's nothing magical here"..."We still have plans for large Blackwell clusters". We are not technology experts and cannot opine on specific breakthroughs but our industry contacts are quite familiar with the DeepSeek technology, so this feedback is our best guidepost, for now.That said, there are ramifications to consider:1. China is going to be competitive in AI software and eventually hardware. We have always believed this. We understand the attempt to use the US strength in AI hardware to steer AI for the next decade, but resourceful Chinese companies have been working hard to overcome Commerce Department restrictions. And by all accounts, having to work through those restrictions will drive certain efficiencies of necessity.From a hardware standpoint, the H20 performance limitations are likely to jumpstart domestic hardware efforts, as they do not have to compete with the best
  3. M UpdateMorgan Stanley Research3of what the US has to offer. We still see H20 as the share leader in China, but local solutions definitely have some traction. That said, we continue to see significant revenue contribution from China for NVIDIA, AMD, and Broadcom over the next several years, except for ramification #2.2. Export controls will continue to be a cautionary factor. Clearly, technology restricting China to H20 levels of performance has not kept China from developing LLMs. What this means for government policy is hard to answer at this point. Setting even lower performance thresholds would be a huge shot in the arm for Chinese domestic silicon development. The Biden proposal for AI restrictions - currently suspended - would have limited cluster sizes, but this would have the same impact. That proposal would also have forced other countries that could support Chinese AI development to get a license. 3. The stock market reaction may have an influence on financial decision makers. What we are hearing is that the major developers involved in the "arms race" to build the biggest clusters are maintaining full enthusiasm. But venture capitalists, cloud service partners, or other financial decision makers have already raised concerns about the costs of these clusters. When companies last week discussed higher cloud capex plans, those stocks went up; this week, they may not. All of that could have some pressure on spending levels. To be clear, we don't see that as a reaction to any of the specifics here, but the stock market's enthusiasm for AI capex is one of the inputs. The technology evangelists will spend as much money as the financiers will allow them to spend, in our view, but the finance vote still matters.4. Demand drivers for NVIDIA this quarter and next continue to be somewhat lackluster. In mid December, we wrote about several near term concerns for NVIDIA (Addressing recent concerns; reiterate Top Pick). Demand for Hopper is starting to stall as customers wait for Blackwell, and while Blackwell supply is higher than our initial expectations pre quarter, there continue to be yield concerns on GB200. All the Blackwells produced will likely be sold, but the yield loss on GB200 could limit that somewhat. We still think there is likely upside to January, but at this point we would expect the company to guide about in line for the April quarter.But importantly, the underlying demand for NVIDIA Blackwell - and other AI products - remains very strong. While the unprecedented complexity (mgmt words) makes Blackwell hard to ramp, our contacts still indicate very strong demand for Blackwell in all of its forms, and confidence that Blackwell Ultra - now GB300 - is much more manufacturable are increasing confidence in 2h outcomes, and we expect to return to meaningful upside in 2h. And while we are less enthusiastic than the market on ASICs vs. merchant, we see strong growth opportunities for both.That said, given some of the anxieties now and lack of short term momentum, we are modestly reducing PTs: NVDA (OW) PT from $166 to $152, on reduction of multiple from 35x to 32xAVGO (OW) PT from $265 to $246, on reduction of multiple from 42x CY26 EPS to 39xALAB (EW) PT from $142 to $114, on reduction of multiple from 25x CY26 sales to
  4. M Update420xMU (EW) PT from $98 to $91, on reduction of multiple from 14x to 13x through cycle EPS of $7MRVL (EW) PT from $120 to $113, on reduction of multiple from 55x CY26 EPS to 52x
  5. M UpdateMorgan Stanley Research5Risk Reward - NVIDIA Corp. (NVDA.O)$152.00~32xourMWCY25EPSestimateof$4.75,apremiumtopeers,reflectingtheexpansioninaAInamesaswellasourhigherconvictioninestimatesgivenNVIDIA’shigherAIexposure.WebelievethatNVIDIAshouldtradeatapremiumgivenitshigherprobabilityofupwardrevisionsinthenearterm.MeanMorgan Stanley EstimatesConsensus Price Target Distribution$75.40$220.00Source:Refinitiv,MorganStanleyResearMS Rating▪Blackwellstillinitsearlystages,withallsignspointingtoastrongrampstillahead▪WeexpectNVDA'sDataCenterbusinesstodrivemuchofthegrowthoverthenext5years,asenthusiasmforgenerativeAIhascreatedastrongenvironmentforAI/MLhardwaresolutions-NVDA'sbeingoneofthemostimportant.IncrementalopportunitiesinAI/MLsoftware&services,networking,andADAScandrivegrowthevenhigher.▪BroaderBlackwellavailabilitylaterin2025thisyearshouldfurthercementNVDA'scompetitiveposition,withhigherdollarcontentthroughagreatermixofsystemsalesConsensus Rating Distribution90%Overweight10%Equal-weight0%UnderweightSource:Refinitiv,MorganStanleyResearRisk Reward ThemesNewDataEra:PositivePricingPower:PositiveSecularGrowth:PositiveView descriptionsofRiskRewardsThemeshere$203.00~42xbullcaseMWCY26EPSof~$4.84BullcasehasDC revenuescontinuingtogrowthrough2025.Upsidefromnetworking,GB200basedsystems,andsoftwarecreatepotentialforafullstackAIcomputingcompanyworthyofanevengreatervaluationpremium-HighermargindatacenterandAI-focusedsoftwareandservicesgrowthaccelerates-GPUbasedAIPC gainstraction,widelyincreasingtheclientTAM-Automotiveopportunitytakesoff,allowingthecompanytoearnrecurring,per-carlicensingrevenue$152.00~32xourMWCY26EPSof$4.7435xvaluationisapremiumtothesemisgroup,butclosertoinlinewithoverlargecapAIpeerAVGO.reflectingtheexpansiinallAInamesaswellasourhigherconvictioninestimatesgivenNVIDIA’shigherAIexposure.WebelievethatNVIDIAshouldtradeatapremiumgivenitshigherprobabilityofupwardrevisionsinthenearterm.-Revenuegrowsby44.1%in2025and14.6%in2026-Datacentercontinuestogrowsignificantin2025andinto2026assupplyremainsconstrained$94.50~30xbearcaseMWCY26EPSof~$3.15Twokeydebatesbothgothewrongdirection,causinginvestorstoquestionfutureprospectsforgrowth-GrowthinDC slowssubstantiallyassupplycatchesuptodemandfasterthananticipated-AIdevelopmentcostscomedownmaterially,astrongcompetitorentersthemarkettotakemarketshare,orcustomersbegininsourcingcustomhardwaresolutions-GreaterthanexpectedimpactfromexportcontrolsRiskReward–NVIDIACorp.(NVDA.O)Top PickOWaslarge language modelenthusiasmistransformingcloudcapexPRICETARGETMS PT$171.69RISKREWARDCHARTANDOPTIONSIMPLIEDPROBABILITIES(12M)Key: Historical Stock Performance Current Stock Price Price TargetSource:Refinitiv,MorganStanleyResearch,MorganStanleyInstitutionalEquitiesDivision.TheprobabilitiesofourBulBase,andBearcasescenariosplayingoutwereestimatedwithimpliedvolatilitydatafromtheoptionsmarketasof27Jan2025.Allfiguresareapproximaterisk-neutralprobabilitiesofthestockreachingbeyondthescenariopriceineiththree-months’orone-years’time.View explanationofOptionsProbabilitiesmethodologyhere$118.42$203.00(+71.42%)Prob(>203.00)~3.3%$94.50(-20.20%)Prob(<94.50)~36.5%$152.00(+28.36%)Prob(>152.00)~18.3%JAN'24JUL'24JAN'25JAN'26050100150200USDOVERWEIGHTTHESISBULLCASEBASECASEBEARCASE
  6. M Update6Drivers20242025e2026e2027eGAAP Revenue ($, mm)60,922128,866185,737212,810MW Gross Margin (%)73.675.472.773.5MW EPS ($)1.212.843.994.74Inventory ($, mm)5,2828,71311,66813,306DOI114.497.482.084.2DateEventSource:Refinitiv,MorganStanl26 Jun 2025 -30 Jun 2025NVIDIA Corp Annual Shareholders Meeting5/5BEST24 MonthHorizon4/5MOST3 MonthHorizonGrowthinAIcapexfromcustomersNextgenGPUscontinuetooutpacethecompetitionSystemsapproachallowsforhighermonetizationovertimeNewdriversemergeforNvidiasuchasAIPCs,autonomousvehicles,robotics,andsoftware0-10%Europe ex UK0-10%Japan0-10%Latin America0-10%UK10-20%APAC, ex Japan, MainlandChina and India10-20%North America40-50%Mainland ChinaSource:MorganStanleyResearchEstimateView explanationofregionalhierarchieshereSource:Refinitiv,FactSet,MorganStanleyResearch;1thehighestfavoredQuintileand5istheleastfavoredQuintileInst. Owners, % Active51.9%HF Sector Long/Short Ratio2.1xHF Sector Net Exposure23.5%Refinitiv;MSPBContent.IncludescertainhedgefuexposuresheldwithMSPB.Informationmaybeinconsistentwithormaynotreflectbroadermarktrends.Long/ShortRatio=LongExposure/Shortexposure.Sector%ofTotalNetExposure=(Foraparticularsector:LongExposure-ShortExposure)/(Acrossallsectors:LongExposure–ShortExposure).GrowthintrainingandinferencepropeldatacenterrevenueGamingsalesaccelerateasGPUbasedAIPCsgaintractionDGXplatformsgaintraction,especiallyamongenterprisecustomersAIendmarketsdon'tmaterializeasexpected,customerssharplyreduceGPUpurchasesAMDreemergesasaviableGPUcompetitorCloudcustomersareabletodevelopcompetitivecustomhardwareMeanMorgan Stanley EstimatesSource:Refinitiv,MorganStanleyResearFY Jan 2025eSales /Revenue($,mm)123,659133,331EBITDA($,mm)82,37890,643Net income($,mm)67,89577,902EPS($)2.913.14RiskReward–NVIDIACorp.(NVDA.O)KEYEARNINGSINPUTSCATALYSTCALENDARINVESTMENTDRIVERSGLOBALREVENUEEXPOSUREMSALPHAMODELSRISKSTOPT/RATINGRISKS TO UPSIDERISKS TO DOWNSIDEOWNERSHIPPOSITIONINGMSESTIMATESVS.CONSENSUS128,866129,31287,73087,08772,81972,9532.942.95
  7. M UpdateMorgan Stanley Research7Risk Reward - Micron Technology Inc. (MU.O)$91.00~13xthrough-cycleearningsofUS$7.00,toreflectcurrentmacroeconomicexpectatioincludinginflationarypressures,consumerspending,andriskappetite,aswellastcompany'sguidanceforsustainedhighercapitalspendingwithawidedivergencebetweenoperatingmarginandFCFmargin.MeanMorgan Stanley EstimatesConsensus Price Target Distribution$70.00$250.00Source:Refinitiv,MorganStanleyResearMS Rating▪DRAM fundamentalsbeginningtoworsenasdespiteenddemandincoremarkets(PCs,servers,smartphones)remainingweak.▪WeseethestockasalreadydiscountingpeakcashflowsthatwerebelowUS$5psharein2021;wearenowlookingatroughly5yearsofzerofreecashflowinthetrailiperiod.▪ExecutiononAIcontinuestoimpress,andweexpectMicron'sshareinthoseproductstocatchuptothecompetitionovertime.SupportingmarginsanddrivingahighermultiplethanpriorcyclesConsensus Rating Distribution85%Overweight13%Equal-weight3%UnderweightSource:Refinitiv,MorganStanleyResearRisk Reward ThemesNewDataEra:PositiveSecularGrowth:PositiveView descriptionsofRiskRewardsThemeshere$153.0017xthrough-cycleearningsofUS$9.00Grossmarginimprovementcontinues,drivenbyscale,mix,andcostimprovementsinnewproducts.Pricingpressurealleviatesasdemandmovesabovesupply.Changesbymanagementconvinceinvestorsthatsustainableprofitsaretobefoundintstock;fabutilizationremainslowandcapacityshrinksfurther.$91.00~13xthrough-cycleearningsofUS$7.00Ourthrough-cycleearningsestimateofUS$7.00isapremiumtoaverageearningsoverthelast7yearsduemostlytoHBM.Our16xmultiplereflectsthemarketenthusiasmfortheHBMopportunity,butadiscounttothebroadersemigroup.$55.0010xthrough-cycleearningsofUS$5.50Memorydownturnseesafalsebottoming.Thehighlevelsofinventorythroughouttheindustrywithpotentiallylacklusterenddemandrecoverykeepusonadecliningtrajectory,wherealackofcashflseverelypressurestheearningsmultiple.RiskReward–MicronTechnologyInc.(MU.O)Memorystarting tosoften,butAIdriving ahighermultiple.EWonvaluationPRICETARGETMS PT$128.45RISKREWARDCHARTANDOPTIONSIMPLIEDPROBABILITIES(12M)Key: Historical Stock Performance Current Stock Price Price TargetSource:Refinitiv,MorganStanleyResearch,MorganStanleyInstitutionalEquitiesDivision.TheprobabilitiesofourBulBase,andBearcasescenariosplayingoutwereestimatedwithimpliedvolatilitydatafromtheoptionsmarketasof27Jan2025.Allfiguresareapproximaterisk-neutralprobabilitiesofthestockreachingbeyondthescenariopriceineiththree-months’orone-years’time.View explanationofOptionsProbabilitiesmethodologyhere$91.11$153.00(+67.93%)Prob(>153.00)~3.9%$55.00(-39.63%)Prob(<55.00)~17.7%$91.00(-0.12%)Prob(>91.00)~39.7%JAN'24JUL'24JAN'25JAN'2604080120160USDEQUAL-WEIGHTTHESISBULLCASEBASECASEBEARCASE
  8. M Update8Drivers20242025e2026e2027eGAAP Revenue ($, mm)25,11133,94340,71740,161Non Gaap Gross Margin (%)23.739.345.639.5Non-GAAP EPS ($)1.316.489.977.61Inventory ($, mm)8,8758,3747,7247,218DOI163.9143.7123.4105.35/5BEST24 MonthHorizon5/5MOST3 MonthHorizonImprovedpricinganddemandstrengthdriveearningsgrowth0-10%Europe ex UK0-10%Japan0-10%Mainland China0-10%UK30-40%APAC, ex Japan, MainlandChina and India40-50%North AmericaSource:MorganStanleyResearchEstimateView explanationofregionalhierarchieshereSource:Refinitiv,FactSet,MorganStanleyResearch;1thehighestfavoredQuintileand5istheleastfavoredQuintileInst. Owners, % Active53.2%HF Sector Long/Short Ratio2.1xHF Sector Net Exposure23.5%Refinitiv;MSPBContent.IncludescertainhedgefuexposuresheldwithMSPB.Informationmaybeinconsistentwithormaynotreflectbroadermarktrends.Long/ShortRatio=LongExposure/Shortexposure.Sector%ofTotalNetExposure=(Foraparticularsector:LongExposure-ShortExposure)/(Acrossallsectors:LongExposure–ShortExposure).CustomerscontinuetodemonstrateanappetitetotakeoninventoryaroundmacroeconomicuncertaintyAdditionalwaferintensityofHBMfurtherimprovesoverallsupplyanddemandMIcron'sHBMsharesurpassesexpectationsPricingcanturnquickly;afalterinenddemandwithinventorieselevatedcouldleadtoaswiftpricereductionHBMdemandfaltersandcompetitionintensifies,pressuringpriciMeanMorgan Stanley EstimatesSource:Refinitiv,MorganStanleyResearFY Oct 2025eSales /Revenue($,mm)31,08536,509EBITDA($,mm)16,37719,560Net income($,mm)6,3969,003EPS($)5.617.91RiskReward–MicronTechnologyInc.(MU.O)KEYEARNINGSINPUTSINVESTMENTDRIVERSGLOBALREVENUEEXPOSUREMSALPHAMODELSRISKSTOPT/RATINGRISKS TO UPSIDERISKS TO DOWNSIDEOWNERSHIPPOSITIONINGMSESTIMATESVS.CONSENSUS33,94334,96916,72817,7357,3667,8926.486.92
  9. M UpdateMorgan Stanley Research9Risk Reward - Astera Labs Inc (ALAB.O)$114.00Weuse20xCY26saleswhichaddsapremiumforsmallcapscarcityaswellasreflectingthighestgrowthinourcoverage.MeanMorgan Stanley EstimatesConsensus Price Target Distribution$100.00$150.00Source:Refinitiv,MorganStanleyResearMS Rating▪Technologyleadershipincriticalareascentraltomoderncompute▪LargeAIexposure,agnostictoprocessors,withanearlyincumbencyadvantage▪AbilitytooutgrowtheAIrampthroughmultipleproductcycles▪WhilewearestillenthusiasticonthebroaderAIconnectivitytheme,valuationisfullasoneofthemostexpensiveAIsemistocksConsensus Rating Distribution85%Overweight15%Equal-weight0%UnderweightSource:Refinitiv,MorganStanleyResearRisk Reward ThemesSecularGrowth:PositiveTechnologyDiffusion:PositiveView descriptionsofRiskRewardsThemeshere$160.0025xEV/SalesonCY26sales$1,043mnWereflectabetterthanexpectedrafromnewproductsawellasAsteraretainingtheirmonopolisticmarketshare.-RevenuegrowthacceleratesfromsuccessfullaunchofnewproductsandmaintainedleadershipinAries-AIspendcontinuestoaccelerate$114.0020xCY26salesof$918mnWeexpectAsteratotradeatapremiumtotheothermajornetworkingplayersduetheiruniqueAIexposure(estimated80%+in2024).Inaddition,theirrevenuerampisjustbeginningandweforecasthighergrowththanpeers.-20xsalesisapremiumtolargecaps,butthereisascarcityofsmallcapAInamesandweexpectthemtooutgrowpeers$62.0012xEV/SalesonCY26sales$794mnWereflectweakerthanexpectcontributionnewproductsandsignificaAriesretimershareloss.Investorsquestionfutureprospectsforgrowthaskeydebatesgointhewrongdirection.-ToplinegrowthsignificantlyslowsasAstelosesdominantretimerpositioning-NewLeoandTaurusproductsfailtosuccessfullyrampRiskReward–AsteraLabsInc(ALAB.O)AIconnectivityleaderpoisedforoutsizedgrowth,butenthusiasmispricedinPRICETARGETMS PT$120.42RISKREWARDCHARTANDOPTIONSIMPLIEDPROBABILITIES(12M)Key: Historical Stock Performance Current Stock Price Price TargetSource:Refinitiv,MorganStanleyResearch,MorganStanleyInstitutionalEquitiesDivision.TheprobabilitiesofourBulBase,andBearcasescenariosplayingoutwereestimatedwithimpliedvolatilitydatafromtheoptionsmarketasof27Jan2025.Allfiguresareapproximaterisk-neutralprobabilitiesofthestockreachingbeyondthescenariopriceineiththree-months’orone-years’time.View explanationofOptionsProbabilitiesmethodologyhere$83.16$160.00(+92.40%)Prob(>160.00)~0.0%$62.00(-25.44%)Prob(<62.00)~46.0%$114.00(+37.09%)Prob(>114.00)~11.6%JAN'24JUL'24JAN'25JAN'2604080120160USDEQUAL-WEIGHTTHESISBULLCASEBASECASEBEARCASE
  10. M Update10Drivers20232024e2025e2026eRevenue ($M) ($, mm)116383623918Non-GAAP Gross Margin (%) (%)69.077.072.570.5Non-GAAP EBITDA Margin (%) (%)(14.7)30.835.338.7Non-GAAP Net Income ($M) ($, mm)(16)124211310Non-GAAP EPS ($) ($)(0.12)0.721.151.614/5BEST24 MonthHorizon1/5MOST3 MonthHorizonAIdatacenterspendPCIe,CXL,andEthernettechnologyupgradecyclesCompetitivepositioning0-10%Mainland China90-100%North AmericaSource:MorganStanleyResearchEstimateView explanationofregionalhierarchieshereSource:Refinitiv,FactSet,MorganStanleyResearch;1thehighestfavoredQuintileand5istheleastfavoredQuintileInst. Owners, % Active91.3%HF Sector Long/Short Ratio2.1xHF Sector Net Exposure23.5%Refinitiv;MSPBContent.IncludescertainhedgefuexposuresheldwithMSPB.Informationmaybeinconsistentwithormaynotreflectbroadermarktrends.Long/ShortRatio=LongExposure/Shortexposure.Sector%ofTotalNetExposure=(Foraparticularsector:LongExposure-ShortExposure)/(Acrossallsectors:LongExposure–ShortExposure).IncreaseAIspendtopropeldatacenterrevenueQuickerupgradeofunderlyingtechnologydrivingcontentgrowthNewproductlaunchesincreasingTAMIncreasedcompetitionreducesAstera'smarketsharemeaningfullyPauseinAIspendanddatacenterinvestmenthaltsfuturerevenuegrowthCXLserversand1.6Tportspeedsaresignificantlydelayed,preventingnewproducfromgainingtractionMeanMorgan Stanley EstimatesSource:Refinitiv,MorganStanleyResearFY Dec 2024eSales /Revenue($,mm)350385EPS($)0.700.74RiskReward–AsteraLabsInc(ALAB.O)KEYEARNINGSINPUTSINVESTMENTDRIVERSGLOBALREVENUEEXPOSUREMSALPHAMODELSRISKSTOPT/RATINGRISKS TO UPSIDERISKS TO DOWNSIDEOWNERSHIPPOSITIONINGMSESTIMATESVS.CONSENSUS3833810.720.72
  11. M UpdateMorgan Stanley Research11Risk Reward - Marvell Technology Group Ltd (MRVL.O)$113.0052XCY26eBaseCaseMWEPS$2.18,in-linewithhighgrowthAIsemisMeanMorgan Stanley EstimatesConsensus Price Target Distribution$74.20$160.00Source:Refinitiv,MorganStanleyResearMS Rating▪ThecompanyhasAIrelatedopportunity,primarilythehigherspeedelementsoftheInphiopticalbusinesses(PAM4and400ZR),andlongertermcloudcustomsilicon.Wedoexpectthisgrowthtoplayout,inparticularontheopticalside,andweexpectMRVLtooutgrowdatacenterspendingoverall.WeremainEWgivenvaluationaswellashighstockcompensationexpensethatweighsonenthusiasm.Consensus Rating Distribution95%Overweight5%Equal-weight0%UnderweightSource:Refinitiv,MorganStanleyResearRisk Reward ThemesNewDataEra:PositiveView descriptionsofRiskRewardsThemeshere$161.0062XBullCaseCY26eMWEPSof$2.60-CY26Revenuecomesinat$9.6bn-Marvell'sAIopportunitycomesinlargerthanexpected,helpingdriveupsidetorevenueestimatesaswellasdrivemultipleexpansion-Thecompanyhitsonanumberofcloudgrowthopportunitiesthatdrivethemultiplehigher$113.0052XBaseCaseCY26eMWEPSof$2.18-WeexpectCY26revenuetocomeinat$9bn-MWGMof59.0%in2026(includesSBC)$77.0042XBearCaseCY26eMWEPSof$1.83-Storageandenterprisedatacenterremainweak,withlittlelineofsightforarecovery-Networkingcontinuestobeaheadwind,asinventorybothatcustomersandinthechannelcontinuetooffsettailwindsinthecustomASIC business-StockovershootstothedownsideclosertoitstangiblebookvalueRiskReward–MarvellTechnologyGroupLtd(MRVL.O)AttractivegrowthcapabilitybutvaluationleavesusEWPRICETARGETMS PT$126.09RISKREWARDCHARTANDOPTIONSIMPLIEDPROBABILITIES(12M)Key: Historical Stock Performance Current Stock Price Price TargetSource:Refinitiv,MorganStanleyResearch,MorganStanleyInstitutionalEquitiesDivision.TheprobabilitiesofourBulBase,andBearcasescenariosplayingoutwereestimatedwithimpliedvolatilitydatafromtheoptionsmarketasof27Jan2025.Allfiguresareapproximaterisk-neutralprobabilitiesofthestockreachingbeyondthescenariopriceineiththree-months’orone-years’time.View explanationofOptionsProbabilitiesmethodologyhere$100.33$161.00(+60.47%)Prob(>161.00)~6.5%$77.00(-23.25%)Prob(<77.00)~35.3%$113.00(+12.63%)Prob(>113.00)~28.6%JAN'24JUL'24JAN'25JAN'2604080120160USDEQUAL-WEIGHTTHESISBULLCASEBASECASEBEARCASE
  12. M Update12Drivers20242025e2026e2027eGAAP Revenue ($, mm)5,5085,7558,2039,017MW Gross Margin (%)60.360.159.059.0MW EPS ($)0.660.431.912.18Inventory ($, mm)864680792837DOI98.272.870.268.65/5BEST24 MonthHorizon5/5MOST3 MonthHorizonCompanyseesrevenuegrowthinStorageandNetworkingresumingdrivenbytheenterpriseandstoragemarketsAccretionfromtheAquantiaandAveraacquisitionscomeinaboveexpectations0-10%North America20-30%APAC, ex Japan, MainlandChina and India20-30%Europe ex UK40-50%Mainland ChinaSource:MorganStanleyResearchEstimateView explanationofregionalhierarchieshereSource:Refinitiv,FactSet,MorganStanleyResearch;1thehighestfavoredQuintileand5istheleastfavoredQuintileInst. Owners, % Active64.7%HF Sector Long/Short Ratio2.1xHF Sector Net Exposure23.5%Refinitiv;MSPBContent.IncludescertainhedgefuexposuresheldwithMSPB.Informationmaybeinconsistentwithormaynotreflectbroadermarktrends.Long/ShortRatio=LongExposure/Shortexposure.Sector%ofTotalNetExposure=(Foraparticularsector:LongExposure-ShortExposure)/(Acrossallsectors:LongExposure–ShortExposure).AIopportunities(Inphiopticalbusinesses)realizeearlierthanexpectedCloudcustomsiliconprojectsarelarger thanexpectedThestoragemarketrecoversAIopportunitiesaresmallerthanexpectedThestrengthweestimateforStorageandNetworkingsurprisetothedownsideEnterpriseDCandNetworkingcontinuetoweighonresultsMeanMorgan Stanley EstimatesSource:Refinitiv,MorganStanleyResearFY Jan 2025eSales /Revenue($,mm)5,4855,772EBITDA($,mm)1,1791,969Net income($,mm)1,2841,392EPS($)1.471.61RiskReward–MarvellTechnologyGroupLtd(MRVL.O)KEYEARNINGSINPUTSINVESTMENTDRIVERSGLOBALREVENUEEXPOSUREMSALPHAMODELSRISKSTOPT/RATINGRISKS TO UPSIDERISKS TO DOWNSIDEOWNERSHIPPOSITIONINGMSESTIMATESVS.CONSENSUS5,7555,7391,9531,7241,3581,3601.551.56
We use cookies to provide, improve, protect and promote our services. Visit our Privacy Policy and Privacy Policy FAQs to learn more. You can manage your personal preferences, including your ‘Do not sell or share my personal data to third parties’ setting using the “Customize cookies” button below.