EU半导体与日本半导体混合键合：尽管HBM存在不确定性，逻辑采用仍具看涨逻辑；将Besi目标价上调至310欧元

EU / Japan semis: Hybrid Bonding — Bullish logic adoption despiteHBM uncertainties; Raising Besi PT to €310 Bonding technologies are essential to enable advanced packaging, and hybrid bonding(HB) is the ultimate bonding technology, as explained in our reports (The Bonder War: Part1 — HBM, Part 2 and Bonder Primer). This report provides our latest view of the currentbonding market as well as an update to our TAM forecast. David Dai, CFA+852 2918 5704david.dai@bernsteinsg.com Juho Hwang+81 3 6777 6980juho.hwang@bernsteinsg.com We see a clear trend for logic to adopt die-to-wafer (D2W) hybrid bondingto increaseinterconnection speed among logic chiplets and between logic and memory. For logic, weforecast 2028 HB TAM of $887mn, and 2028 shipment of 237 units (103 prior), to reflectour outlook of stronger adoption at TSMC for SoIC packaging. We expect TSMC’s SoICcapacity to grow 8x from 10kwpm in 2025 to 85kwpm in 2028, for AMD, Apple, Google,and Nvidia applications. Intel is also expected to resume HB capacity expansion from 2027,with the ramp up of Clearwater Forest and Diamond Rapids. Carmine Milano, CFA+44 20 7762 1857carmine.milano@bernsteinsg.com HBM adoption timing is less clear. Samsung may adopt HB in HBM4E (16-hi) in 2027,but if HBM4E turns out to be 12-hi only per Korean media, HB adoption is possiblypostponed to 2028 with HBM5, with few (maybe 1) initial layers only while the remainingstays on TCB. For HBM, we forecast 2028 HB TAM of $489mn, and 2028 shipment of130 units, based on our view that some portion of HBM4E will adopt hybrid, and HBM5 willmigrate to hybrid bonding. Raising HB TAM forecast.We forecast the hybrid bonder TAM to surge to $1.4bn in 2028,driven by the adoption in both logic and HBM. Compared to Besi’s market forecast, our2028 unit shipment estimates of 367 units is 24% above the mid case and 8% below thehigh case. Our HB TAM model is available for download here. Raising Besi TP to €310, as a technology leader and structural beneficiary of hybridbonding adoption.Assuming 2028 market share of 76% (80% logic, 70% HBM), Besi’srevenue from HB would be $1.1bn, which is based on unit sales of ~281 units and ASPassumption of $3.75mn per unit. We raise 2027 / 2028 EPS by 3.6% / 23% respectivelyand our 2028 EPS is now 16.6% above consensus. Based on 40x (45x prior, cut to reflectweaker visibility on hybrid bonding) 1-year forward P/E and rolling over a quarter, we takeBesi’s PT up to €310.00 (€280 prior), with 39% potential upside.Reiterate Outperformrating. For DISCO, we expect the real upside would come from hybrid bonding adoption, asdicing / grinding systems with high-cleanness specification would need to be adoptedto ensure a particle-free bonding surface, enabling a high yield. Overall setup for DISCOseems favorable into 2027. Current drivers of generative AI in which DISCO partakes ismostly HBM and CoWoS. Whereas from 2027, we expect further adoption of NAND CBAbonding from more players such as Samsung, increased adoption of BSPDN, and die-to-wafer hybrid bonding. In addition, power semis such as SiC is also likely to resumeinvestment.Reiterate Outperform rating. BERNSTEIN TICKER TABLE INVESTMENT IMPLICATIONS We rate Besi (TP=€310.00) and Disco (TP=¥85,000) Outperform. DETAILS Bonding technologies are essential to enable advanced packaging — we have written extensively on the different bondingtechnologies such as Thermocompression bonding (TCB) as well as hybrid bonding (HB) for both High Bandwidth Memory(HBM) and logic applications, in our previous reports (The Bonder War: Part 1 — HBM bonding technology shifts, The BonderWar: Part 2 — Path to advanced packaging for logic chips and The Bonder Primer: Essential enablers of advanced packaging). This report provides our latest view of the current bonding market as well as an update to our TAM forecast (Exhibit 1). In short, •We see a clear trend for logic to adopt die-to-wafer (D2W) hybrid bonding to increase interconnection speed between logicchiplets and between logic and memory. For logic, we forecast 2028 HB TAM of $887mn, and 2028 shipment of 237 units(103 prior), to reflect our outlook of stronger adoption at TSMC for SoIC packaging. •HBM adoption is less clear. While Samsung may adopt HB in HBM4E in 2027, it’s possible to be postponed to 2028 withHBM5, with few (maybe 1) initial layer only. We forecast 2028 HB TAM of $489mn, and 2028 shipment of 130 units, basedon our view that some portion of HBM4E will adopt hybrid, and HBM5 will migrate to hybrid bonding. •We forecast the hybrid bonder TAM to surge to c. $1.4bn in 2028, driven by the adoption in both logic and HBM. Comparedto Besi’s hybrid bonding market forecast, we are more optimistic in logic adoption despite being more conservative withHBM adoption of HB, and our 2028 unit shipment estimates of 367 units is 24% above the mid case and 8% below the highcase (Exhibit 2). •For Besi, assuming 2028 market share of 76% (80% logic, 70% HBM), Besi’s revenue from