台湾访问要点：半导体生产设备 | 日本

Suzune Tamura, CFAEquity Analyst Suzune.Tamura@morganstanleymufg.com+81 3 6836-8891Kazuo Yoshikawa, CFAEquity Analyst Taiwan Visit Takeaways Kazuo.Yoshikawa@mnorganstanleymufg.com+81 3 6836-8408 ComputexonJun2-5.Keytakeawayswereasfollows Interest in SPE appears more subdued compared with the strong momentum seen in late 2025 to early 2026.That said, compared with three months ago when concerns around memory cycle JapanIndustry ViewAttractive sustainability rose on the back of spot price declines, we heard limited pessimismon the memory cycle through 2027.Investor meetings showed strong interest in Kokusai, Lasertec, and Advantest; we recently made Advantest our Top Pick.Through Computex, etc., we reconfirmed memory demand expansion driven by agentic Al, as well as growing focus on co-packaged optics (CPO) as the nexttechnical bottleneck. spot prices turning upward again, our impression is that investor conviction in thesustainability of the memory cycle has increased. Memory shortages could persistfor the next 2-3 years, and with advanced logic investment remaining strong, front-end wafer process names should continue to see major benefits. In back-endprocesses, we believe CPO would become one of the major topics from 2027-2028onward and focus on potential upside for Advantest in testing Morgan Stanley does and seeks to do business withcompanies covered in Morgan Stanley Research. As a result,investors should be aware that thefirm may havea conflict ofinterest that could affect the objectivity of Morgan StanleyResearch. Investors should consider Morgan StanleyResearch as only a single factorin making their investmentdecision. For analyst certification and otherimportant disclosures,referto theDisclosure Section,located at the end ofthisreport. += Analysts employed by non-U.S. affliates are not registeredwith FINRA, may not be associated persons of the memberand may not be subject to FINRA restrictions oncommunications with a subject company.public appearancesand trading securities held by aresearch analyst account. Chan confirmed from supply chain checks that the company can produce 2.5-4.Omn VeraCPUs (link). Since a Vera CPU can accommodate 1.5TB of LPDDR5X, triple that of theapproximately 4-6mn GB. That means that Vera alone will need more than 10% of lastyear's total DRAM bit demand. Based on production at 1cnm, a rough estimate suggestsdemand equivalent to 100-200kwpm. CPO Marvell Technology (analyst: Joseph Moore) offered an explanation of the copper (Cu)wall in its keynote address at Computex. Optical signal transmission via fiber optics hastraditionally been used among and within data centers. However, electrical signaltransmission via copper is used for distances when feasible (e.g., racks, semiconductorpackages) due to copper's high reliability and cost advantages. That boundary is the Cuwall. In inter-rack copper electrical signal transmission, transmission speed and distance areinversely proportional. While a 100G/lane data rate allows transmission up to 5m, thatfalls to 2.5m at the 200G level increasingly in use at present, and will shrink further to1.25m at the 400G targeted for adoption around 2028. Since the height of the racks is 2mit is difficult to fully connect racks with copper for 400G transmission. It was explainedthat the Cu wall is currently moving from the left to right side of Exhibit 2 Teradyne (analyst: Shane Brett) has said that demand for CPO tester TAMs should reach$100mn in 2026, $300-700mn in 2028,and over $1bn thereafter as CP0 is adopted inscaled-up applications, indicating significant potentialfor market growth. Also, we believeDisco will benefit from hybrid bonding technology for COUPE.The placement of switchASICs and optimal engines onto substrate boards will lead to increased board size,bolstering demand for substrate boards and board equipment. That should in turn spurgreater demand for Ushio's stepper exposure equipment. Investorfeedback Kokusai Electric: There was much discussion regarding the firm's guidance. The company commented on its conservative guidance for 2H F3/27 atour Japan Summit on May 20-22, noting an upside risk to its currentsales guidance (-17% HoH) to flat growth HoH. It presently targetssales of ￥280bn for the whole of F3/27, but flat 2H growth would liftthat to ￥300bn. Further upside potential Lies in demand fromemerging Chinese companies, which are only cautiously incorporatedin current guidance, and additional demand from Chinese DRAMmanufacturers. Given that lead times have extended to eight months,we believe 2H shipments should be finalized by the time that 1Q (Apr-Jun)results areannouncedandstandbyour viewthat anupwardrevision is likely at that time. While Kokusai is generally seen as a NAND stock, it has been moreexposed in recent years to DRAM(Exhibit 3).NAND accounted forsome 40% of device sales at its peak in F3/23, but slid to around half that level in F3/26 with the contraction in the NAND WFE market. However, w