Morgan Stanley: China's advanced packaging to reach 100 billion by 2029, equipment makers outperform OSATs, top pick ACM Research (ACMR.US)

Morgan Stanley released a research report stating that China's advanced packaging remains an AI-driven long-term growth track, but there is a divergence between industry scale growth and OSAT profit growth. Morgan Stanley pointed out that OSAT profitability depends on whether capacity can be filled, while equipment makers' revenue depends on whether capacity is being built. Therefore, it prefers equipment makers over OSATs, and within the OSAT segment, favors share gainers. The industry view remains "Attractive." Among the four names under this round of capacity expansion, ACM Research is the only one primarily listed in the U.S. among Morgan Stanley's top picks. Market size to reach RMB 100 billion by 2029 The report estimates that China's advanced packaging market will reach approximately RMB 100 billion by 2029, with a compound growth rate of about 13% from 2025 to 2029. Among this, Chiplet/2.5D will grow the fastest, reaching approximately RMB 17.7 billion by 2029, with a compound growth rate of 40%; flip-chip and wafer-level packaging will grow at only 10% and 9% respectively over the same period. Morgan Stanley noted that AI accelerators' demand for computing density, memory bandwidth, and heterogeneous integration continues to rise, and against the backdrop of restricted access to advanced-node front-end equipment, packaging-driven performance improvement holds higher strategic value for China. Moderate overcapacity risk exists The report estimates that China's annual 2.5D capacity will increase from 120,000 wafers in 2025 to 436,000 wafers in 2027, roughly equivalent to 16% of overseas capacity (mainly Taiwan Semiconductor Manufacturing Co., Ltd. Sponsored ADR). Multiple players including JCET Group Co., Ltd., TongFu Microelectronics, Tianshui Huatian Technology, and Forehope Electronic are expanding capacity simultaneously. Combined with long equipment lead times and the need to secure positions in the domestic AI supply chain, companies generally tend to "build capacity first, wait for orders later." On the demand side, constraints come from advanced-node yields and HBM supply: the bank expects Chinese manufacturers' AI GPU shipments to reach 4.9 million units in 2027. Assuming approximately 21 good dies per wafer and an overall yield of 85%, this translates to wafer demand of approximately 277,000 wafers; meanwhile, domestic HBM3E at approximately 3 to 4 million units can only support 750,000 to 1 million GPUs, with the remainder still reliant on overseas supply. Based on comprehensive estimates, industry capacity utilization in 2027 will be only about 63%. In addition, domestic capacity mostly starts from CoWoS-S, while the new generation of products is shifting toward CoWoS-L, which may create a structural mismatch of "S-type capacity surplus, L-type capacity tightness." Equipment makers benefit more than OSATs Morgan Stanley stated that regardless of whether newly added OSAT capacity ultimately achieves ideal returns, capacity construction itself requires upfront equipment investment; at the same time, the migration to 2.5D/3D brings additional process steps including fine RDL, TSV, wafer-level processing, temporary bonding, TCB, and hybrid bonding, driving up equipment value per unit of capacity. Data shows that global OSAT capital expenditure grew 65% year-over-year in 2025 and is expected to grow 67% in 2026, with its share of revenue rising to a historical high. The bank's top picks are ACM Research (ACMR.US) and ASMPT: the former benefits from wet processes such as plating and cleaning, with advanced packaging revenue (excluding ECP) growing 153% year-over-year in Q2 2026, and has received the first mass-production order for 510515mm panel-level plating equipment from a mainland China customer; the latter is a leader in TCB and hybrid bonding, with advanced packaging business revenue reaching a record US$339 million in H1 2026, accounting for 30% of group revenue. 3DIC, HBM, and CPO short-term contributions may be overestimated The bank believes that these three technologies are strategically significant, but their profit contribution over the next two years may fall short of market expectations. Among them, 3DIC is constrained by cost and composite yield, making large-scale adoption difficult within the year, with China's market size expected to be only in the tens of millions of dollars range over the next 2 to 3 yearsalthough Huawei has already applied logic stacking technology to its flagship smartphone processors and has set a goal of achieving equivalent 1.4nm-class density by 2031; HBM, while having the largest potential demand, still retains most of its packaging value within the memory ecosystem, making it difficult for independent OSATs to benefit directly; CPO large-scale deployment is more likely to occur in 2027 to 2028. Main U.S.-listed beneficiary Specifically, Morgan Stanley assigned ACM Research (ACMR.US) an Overweight rating with a price target of US$130, implying approximately 94% upside from the September 15 closing price of US$66.9, with a risk-reward ratio of 6.3, the highest among all names. The company is positioned in several segments where equipment intensity is rising rapidlywet processes such as plating and cleaning, which happen to be the steepest part of equipment value per unit of capacity increase in 2.5D/3D and panel-level packaging. Fundamentals are already delivering: in Q2 2026, its advanced packaging revenue (excluding ECP) grew 153% year-over-year, and it shipped its 2,000th plating cell during the quarter; more critically, it secured the first mass-production order for 510515mm horizontal panel-level plating equipment from a mainland China customer, marking its panel-level packaging transition from "approaching mass production" to "true mass production." Moreover, ACM Research is listed at the group level on Nasdaq, but its main production capacity and customers are in China, giving it dual attributes of a U.S.-listed name and domestic substitution.