全球半导体与半导体资本：全球半导体：如果内存不行，半导体资本还能行吗？

Global Semis - Can semi cap work if memory doesn't? that they are all correlated with memory share prices anyway. Today's note reviews theshare price history and debunks the myth. +852 2918 5704david.dai@bernsteinsg.com +852 2123 2645mark.li@bernsteinsg.com companiesandtop5WFEcompaniesasbenchmark.During2012-2018,memoryandSPEcompanies correlationwas only0.4.Since2019the correlation has increased,butonlyto0.6.Conversely,SPE andSOx havehad consistentlyhigh correlation of 0.8-0.9. +12135595917stacy.rasgon@bernsteinsg.com SPEand memory shareprices can go indifferent directions.Therehave beena fewnoticeable periods:FromJan2015-Dec 2016WFEwas up21.9%whilememorywasdown 16.2%,gaining38.2%outperformance in two years; In Jan 2021-Dec 2022,WFEwas up15.3% while memorywas down34%,gaining49%outperformance intwoyears.While we are positive on both WFE and memory from here, investors can still benefit fromdiversification between WFE and memory,rather than worrying that both must go up anddown together.This is especially the case considering that WFE growth is needed to addmemory capacity,which is the cause of memory pricevolatility. +81367776980juho.hwang@bernsteinsg.com Carmine Milano, CFA+44 20 7762 1857carmine.milano@bernsteinsg.com Edward Hou, CFA+85221232623edward.hou@bernsteinsg.com Mean reversion?Memoryand WFE performance mightbe the same in the long run...aswas the case during 2011-2019.While memory outperformed WFEbya whopping 661%since June 2025,it was just reversingthelengthyunderperformance during 2019-June2025,and onlymanaged tocatchup in Feb 2026when both were up36x in15 years.Nowthat memory has significantly outperformed WFE, if mean reversion happens it could favorWFE from here. Yipin Cai, CFA+85221232669yipin.cai@bernsteinsg.com Fundamentally,welikeWFElong here.Memorycapex is clearlyaccelerating;SK hynixhas recentlyannounced their KRw 10Otn (~$67bn)additional investment in their newCheongju fabs, and the Korean government seems to be considering measures to supportthe initiatives from Samsung and SK hynixin buildingfabs in the Southwestern Korea.Wesee consensus upward revision potentialfor WFEmarket andcompany EPS through 2028,and we expect further growth ahead. For memory, though we don't think LTA can help much (replay of our recent webinar),the shortage & the imminent HBM price hike will still bring significant upward earningsrevisions.We still like Samsung, SK hynix, & Micron esp.after the recent pullback (report)WerateKiOXIAUnderperformon valuation&long-term threat fromChina (report)。 INVESTMENTIMPLICATIONS WerateTEL(TP=￥59,200),Advantest(TP=￥39,200),Disco(TP=￥85,000),Lasertec (TP=50,000),Kokusai(TP=￥8,240.00),ASML(TP=2,300.00)and Besi (TP=280.00)Outperform. Werate Screen (TP=￥12,600) Market-Perform. SamsungElectronics:WerateSamsungElectronicsOutperformwithpricetargetofKRW440,o00. SKhynix:We rate SKhynixOutperform with price targetofKRW3,300,000. Micron: We rate Micron Outperformwithpricetarget of US$1,300.00. KIOXIA:WerateKIOXIA Underperform with price target of JPY40,000. AMAT(Outperform,$525.o0):Exposuretokey inflectionsisstrongandvaluationvspeersremains attractive. LRcX(Outperform,$34o.oo):The company is benefiting fromkey inflections (GAA,packaging, HBM, NANDupgrades)andCY26/27commentary seems supportive. KLAC(Outperform,$197.5o PT): Amid positive WFEtrends KLAC posseses structural growth drivers,a strong and durablecompetitiveposition,lower Chinareplacementrisk,anddisciplinedcapitalallocation,warranting premiumvaluation,inourview. can stillwork if memory stocks stop working.As such,we conducted a historical analysis and concludethat while the overallcorrelationbetweenmemoryandsemicapisstrong,therehasbeenmultiple instanceswheresemicapcontinuedtoworkeventhough memory stocks were muted. the correlation between semicap and the overall semis has been historically tight-averaging at O.82,whilememory hashistoricallyshown a weaker correlationwithothersectors,with bothcorrelations (vs.WFEand vs.SOX)averaging at O.54.Themonthly returns showa similar pattern as in Exhibit 2.WFEvs.SOX move moreorless intandem, whereas it is somewhatnoticeablethatmemorymoves idiosyncraticallytothebroadersemiconductorpeers,oftenforan. semicap correlates. Exhibit 3 adds the short-term (3-month)and long-term (24-month) perspective of howthe two subsectorsbehave in relation to each other.Althoughthe relationship itself is positive, but it is easily identifiable that the two sectors movebetweenperiods ofstrongerandweakercorrelations. If we defineperiods ofweaker fit as12-monthrolling correlation below0.4,three periods (June 2013-Nov2014/Nov 2015 -Jan 2016 / Jun 2017-Dec 2017)fit under this category.Conversely, two periods of strong fit -as defined as 12-month rollingcorrelation ofabove0.7-appearduringApr2019-Dec2019/Mar2020-Mar2021.Of note,July2026month-to-dateaverages at 0.6,slightlyabovehistoricalmean of 0.54. dispersion. Historically, low Memory-wFE correlation has not always resulted in large return