J.P. 摩根-全球-半导体行业-全球存储器市场：资本支出削减对供应的影响以及与前一个周期的不同之处-2019.2.7-29页.pdf

Asia Pacific Equity Research07 February 2019 Global Memory MarketFinding the bottom of the cycle:Impact on supply from capex cuts,and whats different vs the previous cycleTechnology-SemiconductorsJJ Park AC(82-2)758-Bloomberg JPMA PARK J.P.Morgan Securities(Far East)Limited,Seoul BranchVaibhav Arya(91-22)6157-J.P.Morgan India Private LimitedHisashi Moriyama(81-3)6736-Bloomberg JPMA MORIYAMA JPMorgan Securities Japan Co.,Ltd.Harlan Sur(1-415)315-Bloomberg JPMA SUR J.P.Morgan Securities LLCBill Peterson(1-415)315-J.P.Morgan Securities LLCSee page 25 for analyst certification and important disclosures,including non-US analyst disclosures.J.P.Morgan does and seeks to do business with companies covered in its research reports.As a result,investors should be aware that the firm may have a conflict of interest that could affect the objectivity of this report.Investors should consider this report as only a single factor in making their investment We are updating our Global Memory Model post 4Q18 results,revised capex and shipment growth guidance and our outlook for memory pricing based on upcoming supply and demand.We expect the memory market to reach US$111bnin 2019(27%annual decline after 100%+growth in the last two years)which includes a 30%reduction for DRAM and 23%reduction for NAND.Given rising inventory at memory makers and weak end-demand outlook,we expect 40%DRAM price correction and 45%NAND price correction in 2019.Capex cuts dont impact 2019 supply:Given the two-year lag between capex cuts and actual impact on supply growth,we look at the three-year moving average capex trend to assess upcoming supply growth.With a meaningful capex hike in 2017/2018 and still-high capex levels in 2019,we expect overall three-year moving average capex to remain high throughout our forecasting period.Hence,we do not expect any meaningful slowdown in production output despite recent capex cut announcements by all major DRAM makers.Prevailing oversupply likely to continue:Our updated model indicates that overall DRAM demand growth will be 12%Y/Y in 2019 vs.supply growth of 18%,given that the growth rate in all major applications is likely to slow along with a destocking cycle in 1H19 and an inventory sell-down on supply.Hence,we conclude that oversupply will persist throughout 2019 unless there is any meaningful upside to key demand drivers such as server and smartphone.Inventory destocking throughout 2019:Given the lack of any incremental demand from server following significant order cuts from multiple hyper-scalers toward 2018-end,we hear that memory makers are stacking up wafers without undertaking the back-end process to minimize reporting inventory level.We think DRAM makers are likely to clear out their stock at a discount price in return for a big-volume purchase,implying ongoing weak DRAM prices.Comparison with the 2016 cycle:Some people believe that capex cuts signal the bottom in the DRAM cycle and that the current cycle resembles the lastdownturn(15-16)when companies guided capex cuts in 1Q16 results(which happened during 2Q16),and the cycle bottomed in 2Q16.However,DRAM bit shipment growth actually accelerated in 2016 despite capex cuts.Also,2H16 recovery was driven by server DRAM demand(an initial stage of DCinvestment)and inventory restocking in China smartphone makers.However,the demand and inventory situation in 2019 is totally different from that of 2016and there have also been meaningful capex hikes in the last two years.Where is the bottom for DRAM and NAND?Our revised model suggests that the DRAM market will remain on a downturn cycle until 1Q20,with a moderate recovery in 2Q20 assuming that all DRAM inventory is digested in 2019.We expect the NAND market to begin to grow in 3Q19 on the back of price elasticity,since we expect NAND prices to fall by 60%from their peak by 2Q19,which should accelerate SSD adoption and content growth in major applications.2H19 recovery appears distant:Recent share price recovery for memory names is attributable to(1)positive reaction to WFE capex-cut announcements,(2)an expected V-shaped recovery in 2H19,and(3)undemanding valuationspost meaningful share price corrections.Given the extended inventory destocking cycle,lackluster end-demand outlook and lack of visibility for shipments/ASP,we think demand recovery is unlikely in the foreseeable future.DRAM/NAND market growth rates(%)Source:WSTS,J.P.Morgan estimates.每日免费获取报告1、每日微信群内分享7+最新重磅报告；2、每日分享当日华尔街日报、金融时报；3、每周分享经济学人4、行研报告均为公开版，权利归原作者所有，起点财经仅分发做内部学习。扫一扫二维码关注公号回复：研究报告加入“起点财经”微信群。2Asia Pacific Equity Research07 February 2019JJ Park(82-2)758- Global memory model:RevisionsAfter making revisions to our key assumptions,we forecast the DRAM market in 2019 to account for US$69bn,which implies a stronger than expected shrink of 30%driven by price declines amidst much weaker than expected demand in both smartphone and especially server.We are relatively more positive on NAND demandoutlook,however we think a supply glut will still result in steep NAND ASP decline(45%)causing a market shrink of 23%.For 2020,we expect NAND market to resume 15%Y/Y growth while DRAM market will decline by a further 7%.Our channel checks confirmed meaningful channel inventory for memorymanufacturers and OEMs which gives us confidence about an inventory destocking cycle in 1H19 at least and the destocking cycle could continue throughout 2019 given rising inventory at DRAM manufacturers and weak-end demand.Even though most memory makers guided for a demand recovery from 2H19onwards,we think that their visibility for end-demand is limited because they are mostly interacting with OEMs(direct business with enterprise forms a small portion).Instead we believe that the destocking will likely continue throughout 2019.We have noticed a series of earnings downgrades for memory makers since late-2018,but we still believe the market underestimates the extent of declines in memory prices and bit shipments in the downturn cycle since memory makers wont be able to sell production outputs as well as carry-over inventory due to weak demand.On a full year basis for 2019,we forecast DRAM and NAND prices to decline 40%Y/Y and 45%Y/Y followed by another decline of 21%Y/Y and 17%Y/Y for DRAM and NAND,respectively in 2020.Figure 1:Global memory market(DRAM+NAND)size and YoY change trend$(bn),%Source:WSTS,J.P.Morgan estimates.3Asia Pacific Equity Research07 February 2019JJ Park(82-2)758- Figure 2:DRAM and NAND market size(2014-20E)$(bn)Source:WSTS,J.P.Morgan estimates.Figure 3:DRAM and NAND market growth rates(2014-20E)%Source:WSTS,J.P.Morgan estimates.DRAMSupply:As memory makers toned down their capacity growth guidance in expectation of weak end-demand,we have revised down our forecasts for 2019/20shipments by 9%.We like to highlight that in an inventory destocking cycle the shipment growth would still be higher than the production growth.We still pencil in18%/18%bit growth in 2019/20,majority of which comes from tech-migration and sell-down in carry-over inventory from 4Q18.Demand:Weve revised down our consumption estimates by 10%/11%for 2019/20 to reflect both weaker volumes(for CE:smartphones/NBPC and Enterprise:Servers)and weaker content growth as BOM pressures remain high for set makers/OEMsaffected by trade-tariffs and increased costs of relocation.Negative adjustments to server density are made on account of a delayed upgrade cycle due to Intel Whitley(new platform)launch getting pushed into 2020,lower capex in hyperscale,falling UTRs at Chinese datacenters and macro-overhangs.Although dual-cam adoption,high-res screens,on-device AI/AR remains structural growth areas,overall smartphone contents get negatively impacted from a lower-mix of high-end segment.ASP:As inventory destocking happens throughout 2019 combined with weak incremental demand,we have revised down our ASP assumptions for 2019/20 by14%/9%respectively.We like to remind that although absolute correction of 40%looks high,were coming off a meaningfully high base after 48%/20%sequential ASP growth in 17/18.NANDSupply:We have made a 5%/4%downward revision to NAND supply estimates in 2019/20,as memory makers guided for adjusting production growth amidst prolonged NAND price decline.However,since it continues to be a positive operating margin business and 3D yields have stabilized even for higher layer counts(64L)as well as capex hike in the last two years,we model a healthy 37%/39%supply growth in 2019/20,which keeps the market in an oversupplied situation.Demand:We have adjusted NAND consumption forecasts down slightly by 5%/4%for 2019/20.Accounting for the price elasticity of NAND demand,we revised up SSD shipments to reflect steep HDD replacement for client/enterprise applications.However content growth in SSDs takes a hit amidst NBPC CPU shortage and lack of platform upgrade for enterprise servers.Given relaxed pricing,we also reflect 47 45 41 73 99 69 64 28 29 32 47 54 42 48 -20 40 60 80 100 12020142015201620172018E2019E2020EDRAMNAND4Asia Pacific Equity Research07 February 2019JJ Park(82-2)758- upside to content growth assumption in smartphones(both high/mid-end),however weak shipments pressurize the demand in this case.ASP:Given weak end-demand and inventory destocking cycle,we also made downward adjustments(19/20 ASPs 12%down)to reflect oversupply situation.This implies a 45%/17%price decline in 2019/20 following a 11%correction in 2018.Going forward,we are more constructive on NAND demand recovery and have higher visibility of price stabilization in 2020.Table 1:J.P.Morgan Global Memory Model:Summary of changesRevisedPriorChange(%)20182019E 2020E20182019E 2020E20182019E 2020EMemory revenue(US$,bn)153.4110.8112.2159.7140.6135.7-4%-21%-17%DRAM revenue99.168.964.0105.088.778.9-6%-22%-19%NAND revenue54.341.948.254.751.956.8-1%-19%-15%Memory price(US$)DRAM ASP(2Gb.)1.921.150.911.941.340.99-1%-14%-9%NAND ASP(64Gb.)1.91.10.91.901.201.02%-12%-12%Memory Consumption(mn)DRAM shipments(2Gb.)51,59659,95070,61854,24366,28979,440-5%-10%-11%NAND shipments(64Gb.)28,06039,48454,74929,54841,72557,294-5%-5%-4%Shipments(mn)Desktops99100991009998-1%1%1%Notebooks1621621621601611611%0%0%Servers121314131314-4%-1%-3%Smartphones1,5591,4681,4771,5851,6191,678-2%-9%-12%Tablets137139139146138131-6%0%6%SSDs2192703142102492974%8%6%Average DRAM memory(MB)Desktops5,2505,6696,3005,2155,6986,3501%-1%-1%Notebooks6,1916,6527,4776,1016,6757,4181%0%1%Servers255,183292,102374,899264,485334,580428,912-4%-13%-13%Smartphones2,6722,9593,4932,6292,9663,6602%0%-5%Tablets2,8583,1563,7692,9433,3993,791-3%-7%-1%Average NAND memory(MB)SSDs411,424493,080647,240439,255542,861706,007-6%-9%-8%Smartphones51,49970,35596,34251,72366,74488,2760%5%9%Tablets64,44085,195111,03266,38792,023103,290-3%-7%7%Supply assumptionsDRAM(2Gbe.,M)51,88461,09172,20754,87466,91379,779-5%-9%-9%NAND(64Gbe.,M)29,29640,27455,97030,67642,57458,371-4%-5%-4%Memory capex(US$,mn)54,35444,33746,05657,75460,709-6%-27%n/aDRAM25,90119,31922,35627,49328,118-6%-31%n/aNAND28,45325,01823,70030,26132,591-6%-23%n/aSource:WSTS,J.P.Morgan estimates.Figure 4:Quarterly market revenue comparison:DRAM vs.NAND$(bn)Source:WSTS,J.P.Morgan estimates.5111723291Q144Q143Q152Q161Q174Q173Q182Q19E1Q20E4Q20ENAND RevenueDRAM Revenue5Asia Pacific Equity Research07 February 2019JJ Park(82-2)758- Spending on WFE declines but overall capex healthyPost 4Q company guidance,we note a sequential decline in industry capex in 2019 on account of a)memory-makers adjusting capital spending given reduced bit shipment outlook and b)More cuts to WFE while aggregate capex remains relatively high due to infrastructure investment.The annual capex should come around US$44bn,which breaks down into US$19bn capex for DRAM(down 25%Y/Y)and US$25bn capex for NAND(down 12%Y/Y).Although DRAM is entering a downturn cycle,capex is a long-term commitment(at least 5 years to 10 years)and all DRAM companies are generating extremely high margins(mid-50%+OPM).We do recognize a change in the capital spending mix for infrastructure and WFE(equipment).For Samsung we estimate WFE as a%of total capex will decline to 50%in 2019(vs 70%+in 2018)and for SK hynix as well,WFE capex declines by 40%(much severe than a 15%decline to overall capex).At the company level,we expect Samsungs DRAM capex to decline by 23%Y/Y but still remain at W10trn level(as it spends for Pyungtaek line no.2 and 1Ynm transition)while NAND capex is likely to decline by 28%to W9trn,from a low-level in 2018.SK hynix capex for both the memories declines by high-teen%to W13.5trn(50:50 split for DRAM and NAND)on our estimates.Micron also cut its memory capex by 23%Y/Y to US$6bn while maintaining a 50:50 split for DRAM-NAND.Figure 5:DRAM capex trend$(bn),%Source:Company data,J.P.Morgan estimates.Figure 6:NAND capex trend$(bn),%Source:Company data,J.P.Morgan estimates.We also like to highlight that the capital intensity for DRAM(i.e.the ratio of annual capex to annual revenue)grows into 2019/20 to 28%/35%given the revenue decline(on back of meaningful ASP fall)is much steeper than the capex required for supporting ongoing tech-migrations.We noted a similar 30%capex intensity near the 2015-16 period,however it is reasonable to assume a higher intensity for upcoming years given an increasing difficulty in node-migration.6 7 14 13 11 14 26 19 22-27%16%105%-3%-17%30%79%-25%16%-75%-25%25%75%125%175%-5 10 15 20 25 3020122014201620182020EDRAM capex yoy(%)888111529282524-9%-6%5%29%44%93%-3%-12%-5%-40%-20%0%20%40%60%80%100%0510152025303520122014201620182020ENAND capex yoy(%)6Asia Pacific Equity Research07 February 2019JJ Park(82-2)758- Figure 7:DRAM Capital Intensity trends(2009-20E)Capex,Capex/Rev%-RHSSource:Company data,J.P.Morgan estimates Change in current capex only impacts future supplyAs highlighted above memory capex is a long-term commitment and current year capital spending only impacts the supply growth after 2 years(typical lag duration to build cleanroom,install equipment and begin mass production).Hence,we plot the following 3-year moving average trend for capex(grey-bars)to smooth-out the capex trend and estimate the impact of upcoming supply growth.While the annual capex goes down steeply by 25%in 2019,the moving average trend shows a steady growth of approx.16%,which in our opinion should imply ongoing supply growth for the year(even though the capacity expansion slows down).Of note,we assume 18%/18%shipment growth in 2019/20Figure 8:Capex trendActual versus Moving averagesUSDbnSource:J.P.Morgan memory model,Industry dataUpdate on fab expansion plansBased on the latest capex cut announcements and company guidance,we have adjusted our wafer growth assumptions and now expect average global DRAM/NAND capacity to register muted growth at 5%/6%respectively.21%28%27%22%19%29%30%27%20%26%28%35%0%10%20%30%40%-10,000 20,000 30,000200920112013201520172019EDRAM capex(US$,M)Capital intensity(%)05101520253020062008201020122014201620182020EDRAM capex(USD bn)Simple Moving avg 3yr-Capex25%down16%up7Asia Pa