欧洲科技硬件与亚洲半导体：2026年第二季度后点评——基本面改善但仍存高期待

2Q26 postview - betterfundamentals, but still a high barahead NEUTRALU.S.Semiconductors & Semiconductor With the SOxX still down >20% since its June/July peak, aresemis interesting again? 2028 valuations look attractive, but2027 whilst more palatable now, in our view, still appears full.We believe this cycle will last longer but expectations remainhigh and there are nearer-term debates to navigate Capital EquipmentNEUTRAL Simon Coles, CFA +44 (0)20 3555 4519simon.coles@barclays.comBarclays, UK Fundamentalsimprovedin2Q,butexpectationswerehigh(er).Semissawoneoftheirbestquarters in years in 2Q (S0xX up >90% vs market c14%). This set a high bar for earnings as wewrote in2Q26previews-Positivebutincreasingly reflected(29June2026).Mostcompaniesbeatconsensus butfailedto surpass investorexpectations.Technical factors andwider Aldynamics have added tothepressure.Looking at ourestimatespost 2Q,most nameshave derated5x+on 2027E P/E as share prices declined but EPSmoved up-we increased EPSfora numberof companies by10%post2Q reporting. +44 (0)20 7773 4104thomas.wu@barclays.comBarclays, UK Maurice Patrick+44 (0)20 3134 3622maurice.patrick@barclays.comBarclays, UK Seeking a catalyst.Ourfirstthoughts on3Q wouldsuggestthebartobeat expectations hasn'tcomedownafter2Q-theonlydifferenceisvaluationsnowlooksomewhatmorereasonable.This couldultimatelyprove to be enough,but a moretangible catalyst would increaseconfidence.Demand clearly remains strong, emphasised by Nvidia earnings. However, we donotseeclearreasonsfornarrativestoimprovemateriallyintheshortterm.Wewillalsoneedtonavigategeopolitics (presidential summitatthe endof September)aswell as wider Al newsflow(potential iPOs,newmodel releases,ARR trajectories, enterpriseAlmonetisation,etc.). Cautious nearerterm,still positivelongterm.We seea numberofdebates setto continuethat could hold back ourcoverage nearerterm (see bodyof report),but we think many nameslookattractive,evenhere,in particularTSMC,SKhynix,Samsung,ASMLand ASM (all Ow).Improved visibility on the sustainability of strength into 2028 is likely needed for our coverage torerate meaningfully, but these data points should start to come through in the coming 6+months. We think there is still significant latent demand and thatkeeps us positive longerterm. Interesting charts.5 Our overall takeaway is that2Q strengthened confidence in sectorfundamentals butthat investor expectations had already moved materially higher.Thus, the key question isdrive another round of estimate upgrades into 2027,in ourview,as 2027valuations currentlylookfull (andas 2028remainstoofaraway). estimate revisions. Therefore, the next leg higher likely requires either 1) Al revenue growth toaccelerate again, 2) visibility into 2028 earnings to improve, or 3) geopolitical risks to diminish.Conversely,if earnings continueto risebroadly inline with expectations,valuation expansionmay remain difficult. latest thoughts. .How can Al surpriseto the upside at this point? We believe the market is startingto focusprimarily on the risk of Al demand slowing rather than accelerating. Upside could come fromthree areas: (1) larger frontier model training requirements, (2) higher inference intensity asreasoning and agentic workloads increase, and (3) the emergence ofnew Al use cases thatstimulate additional compute demand.Iif Al revenues accelerate,then concerns onhyperscaler capex and Al ROl could abate,which would remove a major drag for semis.Hyperscaler cloud revenue growth accelerated meaningfully in 2Q (see here)and furtheracceleration is likely needed. We continue to think we are very early on the Al adoptiontimeline.Frontierlab ARR expectations are therefore potentially also a key driver of semis in4Q.Planned compute additions in 2027 appear set to exceed those in 2026, providing scopeforfurther ARR acceleration, but there are clearly a number of moving parts.Datacentres inspace remain a potential source of upside that does not appearto be reflected in investorexpectations,albeit forunderstandable reasons given the execution and timinguncertainties (see Bloomberg), Memorymultiples areattractive,but what isthe catalyst? We see no reason thatthesupply/demand tightness being seen in memory is going to improve anytime before 2028 atthe earliest, and even then it may only be a modest improvement. Thus, pricing shouldremain supportive through most of 2027, and the risk may now be skewed modestly to thecompounded the view that pricing growth had peaked. We don't dispute that and with DRAMone ofthefewplaces wecutnumberspost2Q, expectations heredon't seemas extreme.Typically, this would not be the time to engage, but post-cash P/Es (given significant cashreturns ahead) are extremely low, even vs history, so it may not take much. Nvidia disclosedthat its commitmentshave increased from $119bn to $279bn,primarily driven bymemorywhichshouldprovidesomecomfortthatmemorypricingissustainableinto2o28atleast.The risks tomemory include China expansion and/orfurt