全球半导体：ASML高NA EUV成本深度分析 上调目标价至2300欧元

ASML: High-NA EUV cost deep dive - lifting PT to €2300 on risinglitho intensity HNA EUV will likely be adopted first in DRAM than for logic due to lower cost ofexposure for DRAM. A common concern from investors is that HNA is too expensive (asTSMC said so) and won’t be adopted anytime soon. However, what’s rarely discussed isthe big difference between DRAM and logic: DRAM die size is much smaller than logic, andhence only uses 1 mask, while logic mostly needs 2 masks due to larger die size (stitching).The mask changing slows down throughput by 23% (135WpH vs. 175WpH), and the costof HNA exposure for DRAM is cheaper and easier to justify. So we believe DRAM will likelyadopt HNA in 2027 in 1d, when LNA EUV double patterning is needed, earlier than logic. David Dai, CFA+852 2918 5704david.dai@bernsteinsg.com Mark Li+852 2123 2645mark.li@bernsteinsg.com Stacy A. Rasgon, Ph.D.+1 213 559 5917stacy.rasgon@bernsteinsg.com Logic adoption of HNA is also in sight.The high cost of HNA exposure is largelyassociated to the low tool availability of 80-85%, while LNA is close to 95%. However,HNA tool availability is improving faster, and is targeted to reach ~95% by 2030. WpH isalso expected to increase from 135 to 175 even with 2 masks. While the litho equipmentcost of HNA exposure is currently ~3x LNA for logic, the gap will narrow to ~2x by 2030.Considering other costs and complexity of double patterning, it’s justifiable for logic tomigrate to HNA before that. We expect Intel to be the first to adopt HNA in 2028, whileTSMC will be slower, but still likely to adopt in 2030. Carmine Milano, CFA+44 20 7762 1857carmine.milano@bernsteinsg.com Juho Hwang+81 3 6777 6980juho.hwang@bernsteinsg.com Jack Lin+852 2123 2683jack.lin@bernsteinsg.com Rising litho intensity from 24% in 2025 to 26% in 2028. When HNA replaces LNAmultiple patterning, the other cost of double patterning (etch, depo, cleaning etc) canbe greatly reduced. Hence litho intensity is expected to rise with HNA, just like how lithointensity increased when LNA EUV replaced DUV (ArFi) multiple patterning. We forecastDRAM litho intensity to increase from ~20% in 1z to ~30% in 1d, due to increase in EUVlayer count and adoption of HNA in 1d. While logic litho intensity stabilizes at ~32% in N2and A14 due to limited adoption of HNA, we expect it to rise in A10. Mostly driven by DRAM,we forecast litho intensity to increase from 24% in 2025 to 26% in 2028, and ASML to gainshare. We materially increase our ASML topline forecastsfollowing the unprecedented AI-driven expansion in both advanced logic and DRAM capacity.We raise EUV (includingHNA) shipment forecasts in 2027 to 91 systems (vs. old 86) and 113 (old 87) in2028, and expect ASML to expand EUV capacity.This is on top of strong ASP expansionsupported by throughput improvements. As a result, we now expect EUV revenue to growat a 30% CAGR, reaching €42.7bn by 2030, more than 30% above street. We also revisedour DUV estimates upward, particularly toward the end of the decade, and now expect DUVrevenue to reach €20bn in 2030, from €13bn in 2026. Top Pick, lifting PT on capex cycle and litho intensity.Higher revenue and operatingleverage also support stronger profitability, and our 2028 EPS of €67 is 35% aboveconsensus. We also don’t see capex peaking, and expect ASML revenue to reach €80bnby 2030 (20% CAGR) and EPS €97 by 2030 (31% CAGR). ASML now trades at troughvaluation vs. peers. Given the clear acceleration in capex cycle and rising peer valuation,we raise our target P/E multiple from 35x to 40x (1SD above mean), increasing ourtarget price to €2,300. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We rate ASML Outperform, PT = €2,300, TSMC Outperform, PT = USD 430, Intel Outperform, PT = USD 100, SamsungOutperform, PT = KRW 440,000, SK Hynix Outperform, PT = KR 3,300,000, Micron Outperform, PT = USD 1,300. DETAILS Our ASML Model and Industry models can be downloaded here (ASML.NA / ASML Holding NV; EUV Model; DUVModel). One of the concerns from investors for our ASML long thesis is the delay of High-NA (HNA) adoption, notably by TSMC (ASML- High-NA EUV adoption postponed?). Investors often ask why is HNA not been adopted, and if TSMC doesn’t want to adopt it,does it mean the slowdown of litho vs. other processes such as etch and depo.This report analyzes the HNA cost vs. Low-NA (LNA) EUV, and conclude that HNA will likely be adopted by memory first, followed by logic, which drives increasein litho intensity. We reiterate ASML long and lift PT to €2300. HNA EUV greatly simplifies the process and is set to replace LNA EUV as the leading edge litho tool. •HNA EUV has demonstrated ~8 nm resolution versus ~13 nm for LNA EUV, driven by the move from 0.33 numericalaperture to 0.55 NA optics.In lithography, resolution scales roughly with wavelength divided by numerical aperture, sokeeping EUV wavelength constant at 13.5 nm while increasing NA by ~67% allows ASML to print significantly smallerfeatures. •That