報告_MS_CXMT_20260826

PDF 原檔：報告_MS_CXMT_20260826_original.pdf

圖片清單（已驗證 2026-08-30）

未讀（<40KB 或同組抽樣）：002、004–009、011、012、014–018、020、021、023–027、030、032、033、039–045、048、049（<40KB）

原始內容

aluvv

Stock Rating

Industry View

Price target

Overweight

Attractive

Rmb88.00

M August 26, 2026 08:30 AM GMT

Asia-Pacific Technology

China DRAM Going Mainstream; Initiate CXMT at OW

1

1

4

CXMT benefits from China's strong local AI/server demand for HBM/DDR5. Ongoing global DRAM supply tightness supports pricing but also spurs global brands to adopt CXMT's DRAM.

P/E

P/BV

ROE (%)

EV/EBITDA

framework

-

11.1

6.3

4.2

China's memory industry leader: CXMT is China's largest DRAM fab, with proprietary process nodes and an 11% global bit-shipment share in 2026, per our estimate. It is backed by the China Integrated Circuit Industry Investment Fund ('Big Fund') and counts Alibaba, Tencent, and Xiaomi among its strategic investors. CXMT's DRAM process trails Samsung and Micron by about two generations, but its LPDDR5 products already meet Chinese smartphone requirements. More importantly, CXMT is a key domestic supplier for China's AI infrastructure and national security, including in networking equipment and AI GPU servers.

Ongoing capacity expansion despite restricted access to US semi equipment: Our supply-chain checks suggest ASML DUV lithography tools may be the main equipment bottleneck under export controls. Meanwhile, domestic Chinese equipment vendors continue improving and are replacing US suppliers across several process steps. Assuming DRAM industry margins stay near 80%, 2026-28, we expect CXMT to add ~100kwpm annually, reaching 800kwpm capacity by 2031.

Key Debate #1: Can CXMT produce HBM3E soon? Our recent checks suggest HBM3E supply will be the major bottleneck for China'a AI GPU shipment in 2027. We believe CXMT can produce 3-4mn units of HBM3E stacks in 2027 despite low yields, sufficient to support ~0.8-1.0mn units of China AI GPU shipment.

Key Debate #2: Will CXMT affect global DRAM pricing before 2028? We remain constructive on 2027 global DRAM pricing and demand, given AI-driven demand, and we see no pricing impact before 2028 when CXMT should have enough capacity for large-scale shipments. However, Apple reportedly tested CXMT memory products for devices sold in China, and we have seen US PC brands (incl. HP) adopt CXMT DRAM, suggesting commercially acceptable quality, though any Apple supply relationship remains subject to geopolitical and US approval considerations.

Key Debate #3: How to evaluate CXMT? Worth a premium to global peers? Our PT Rmb88, based on our residual income (RI) model, implies an 18.5x 2027e P/E (vs global peers' 4.4x), reflecting CXMT's stronger growth profile. We expect CXMT to benefit from: 1) strong China domestic AI server market demand; and 2) continued capacity expansion even if prices moderate in a memory downcycle, given DRAM's strategic importance in China. Downside risks include: 1) earlier-than-expected global DRAM price declines; 2) CXMT delays in new products, HBM or 3D DRAM development; 3) tighter restrictions on China's fabs for semi equipment, materials, or servicing; and 4) weaker demand from major China customers.

Morgan Stanley Taiwan Limited+

Charlie Chan Equity Analyst Charlie.Chan@morganstanley.com

+886 2 2730-1725

Greater China Technology Semiconductors

Asia Pacific

Industry View

Attractive

Exhibit 1 : We initiate on CXMT (688825.SS) at OW

Morgan Stanley does and seeks to do business with companies covered in Morgan Stanley Research. As a result, investors should be aware that the firm may have a conflict of interest that could affect the objectivity of Morgan Stanley Research. Investors should consider Morgan Stanley Research as only a single factor in making their investment decision.

For analyst certification and other important disclosures, refer to the Disclosure Section, located at the end of this report.

+= Analysts employed by non-U.S. affiliates are not registered with FINRA, may not be associated persons of the member and may not be subject to FINRA restrictions on communications with a subject company, public appearances and trading securities held by a research analyst account.

MORGAN STANLEY TAIWAN LIMITED+

MORGAN STANLEY & CO. INTERNATIONAL PLC+

Shawn Kim

M

Lee Simpson

Contributors

MORGAN STANLEY TAIWAN LIMITED+

Daniel Yen, CFA +886 2 2730-2863 Daniel.Yen@morganstanley.com

MORGAN STANLEY TAIWAN LIMITED+

Tiffany Yeh

+886 2 7712-3032

Tiffany.Yeh@morganstanley.com

MORGAN STANLEY & CO. INTERNATIONAL PLC+

Cindy Huang Equity Analyst +44 20 7425-2915 Cindy.Huang@morganstanley.com

Kazuo Yoshikawa, CFA

Equity Analyst

+81 3 6836-8408

Kazuo.Yoshikawa@morganstanleymufg.com

MORGAN STANLEY ASIA LIMITED+

+852 2239-7357

Duan.Liu@morganstanley.com

MORGAN STANLEY ASIA LIMITED+

Research Associate +852 3963-2287 Ethan.Jia@morganstanley.com

MORGAN STANLEY & CO. LLC

Joseph Moore

Equity Analyst

+1 212 761-7516

Joseph.Moore@morganstanley.com

MORGAN STANLEY & CO. LLC

Shane Brett

Equity Analyst

+1 212 761-1022

Shane.Brett@morganstanley.com

MORGAN STANLEY ASIA LIMITED+

Daisy Dai, CFA +852 2848-7310 Daisy.Dai@morganstanley.com

MORGAN STANLEY MUFG SECURITIES CO., LTD.+

Suzune Tamura, CFA +81 3 6836-8891 Suzune.Tamura@morganstanleymufg.com

M

Executive Summary: China DRAM Supply - Can It Expand Into The Global Market?

How big and how advanced is China's CXMT in DRAM?

Some key facts and trends in capacity expansion

For 2026, we estimate CXMT will achieve:

• 300k wpm of capacity, representing 13% of global DRAM wafer supply
• 41bn Gb of bit shipments, representing 11% of global bit supply
• US$57bn in revenue

We note that YMTC and JHICC (two other Chinese memory integrated device manufacturers (IDM)) are also building DRAM capacity. However, we expect CXMT to remain the domestic market leader in terms of both technologies and capacity.

We estimate CXMT's share of global DRAM bit shipments will increase to 15% by 2030, from 11% in 2026. With DRAM industry margins currently at 80%, as disclosed by the leading DRAM vendors, we expect CXMT to accelerate capacity expansion from 2H26, adding ~100k wafers per month (wpm) annually.

The technology advancement continues

Recent discussions with DRAM customers and semiconductor equipment vendors point to improving product competitiveness. Our checks indicate CXMT's DDR5 supports 5600MHz, with a roadmap to 6400MHz, broadening its relevance to mainstream server DRAM.

CXMT has also developed 3D stacking capability for a 12H HBM3E package. Thermal management and manufacturing challenges remain, and we forecast mass production in 1H27.

Major Chinese CSPs, including Alibaba, ByteDance, and Tencent, may already be purchasing CXMT DRAM for server applications, according to our checks. Adoption of CXMT's LPDDR5 is expanding among Chinese smartphone brands and select lower-end AI accelerator platforms, while Dell and HP have reportedly used CXMT DRAM in select China and ASEAN PC models. These checks support the view that CXMT is increasingly viewed as a source in mainstream DRAM, although advanced qualification remains a separate hurdle.

M

Exhibit 2: DDR5 spec comparison: CXMT vs. global peers

Source: Company data, Techinsights, Morgan Stanley Research.

DUV is less efficient than EUV but works; 3D DRAM provides an opportunity for China to overcome the EUV limitation

We view ASML's deep ultraviolet (DUV) lithography tools as the primary bottleneck to this expansion. Our recent checks with DRAM customers and equipment vendors suggest that global investors may be underestimating CXMT's DRAM quality, particularly in server DRAM performance, high-bandwidth memory (HBM) generation and density, and adoption across China's AI compute ecosystem. Debate also continues on whether Apple could qualify CXMT's low-power double data rate (LPDDR) products, especially against a backdrop of tight global memory supply.

CXMT recently completed the migration to its Gen4b process, a 16nm-class node, using primarily domestic equipment, with the exception of DUV lithography tools, which we believe have achieved production yields above 80%. As a result, we expect CXMT to accelerate capacity expansion from 2H26. On our estimates, CXMT alone could expand capacity to 800k wafers per month by 2031, provided the company continues to secure DUV lithography tools from overseas suppliers. This would represent a slightly faster growth rate than the broader memory industry's expected capacity expansion, which we estimate will roughly double over the same period.

Our channel checks with equipment vendors suggest that CXMT may adopt vertical channel transistor (VCT) technology, a transitional architecture toward 3D DRAM, to advance to its Gen5 (~15nm-class) process node without relying on EUV lithography tools. CXMT will begin limited production of this next-generation DRAM technology in 2H28 (see Appendix 2: What is 3D DRAM Technology? Can China CXMT be the First Chip Maker to Adopt it? ).

M

Exhibit 3: Global DRAM capacity to reach 3,374kwpm in 2028e , while CXMT may account for 15% of that (around 500kwpm by 2028)

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Initiate China CXMT at OW - Both scale and technology are advancing

We initiate coverage of CXMT with an Overweight rating and a price target of Rmb88, implying an 18.5x 2027e P/E, compared with global peers at 4.4x 2027e P/E. We believe this valuation premium is justified by CXMT's stronger growth outlook, supported by sustained capacity expansion and increasing demand from China's AI infrastructure ecosystem. CXMT's higher growth relative to global peers reflects two key factors: (1) strong demand from China's domestic AI server market, including both GPU- and CPUbased systems; and (2) continued capacity expansion driven by localization requirements. We expect CXMT to continue expanding capacity even if DRAM pricing moderates.

Bull case: In this scenario, we assume US cloud service providers and consumer electronics brands begin sourcing CXMT DRAM amid severe global memory shortages, while CXMT achieves a breakthrough in 3D DRAM technology by 2028 (see Appendix 2: What is 3D DRAM Technology? Can China CXMT be the First Chip Maker to Adopt it? ). Our checks with equipment vendors suggest CXMT could adopt VCT technology, a transitional architecture toward 3D DRAM, to advance to its Gen5 (~15nm-class) process node without relying on EUV lithography tools.

Bear case: In this scenario, we assume tighter export restrictions on semiconductor manufacturing equipment, particularly ASML's DUV lithography tools.

M

Exhibit 4: CXMT's revenue to rise at a 140% revenue CAGR,

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 6:

CXMT to expand capacity from 180kwpm in 2025 to 300kwpm in 2026, followed by annual capacity additions of approximately 100k wpm, reaching 800k wpm by 2031

CXMT's Total DRAM Capacity

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 5: CXMT's OpM to be 70-72%, 2026-28e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Implications for Global Memory and Semiconductor Capital Equipment Stocks

We remain constructive on global DRAM pricing and demand in 2027. While CXMT's supply is expanding rapidly, we do not expect incremental Chinese DRAM supply to materially affect industry pricing before 2028, given strong AI-driven demand and the time required for new capacity to reach meaningful scale. We also do not expect CXMT's capacity expansion to stop even if industry profitability levels or cash-flow generation start to moderate. If the DRAM industry enters a more cyclical phase after 2028, Chinese DRAM supply could become an increasingly important factor in global supply-demand dynamics and pricing.

We think China's domestic DRAM supply remains insufficient to meet rapidly growing demand, providing CXMT with a substantial opportunity to expand market share (see Appendix 3: Assessing China DRAM Demand/Supply (Local Sufficiency) for a detailed supply-demand analysis).

So when should investors begin to focus on China DRAM as a competitive threat? As discussed in the previous section, the bull case for China DRAM rests on several developments:

• US CSPs and consumer electronics brands could begin sourcing DRAM from CXMT amid severe global memory shortages.
• Strong demand from China's domestic AI infrastructure buildout, including both GPU- and CPU-based systems, could drive sustained growth in local memory consumption.
• Continued strength in global AI infrastructure spending could support favorable DRAM pricing and industry profitability for longer than currently expected. Finally, successful technology migration-potentially through VCT and ultimately 3D DRAM architectures-could help CXMT narrow the technology gap with global peers and accelerate its capacity expansion plans.

Exhibit 7: DRAM supply sufficiency ratio: Supply growth continues to lag demand

Exhibit 8: China's domestic supply is even more constrained, providing significant room for CXMT to grow

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Source: TrendForce, company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Global memory and equipment implications

Our Korea Technology analyst Shawn Kim maintains his OW ratings on both Samsung Electronics (Top Pick, 005930.KS, PT W381,000) and SK hynix (000660.KS, PT W2,600,000). He continues to view both companies as key beneficiaries to AI compute and agentic AI trends (see Agentic AI - The Surge Begins).

We estimate DRAM prices will increase by more than 20-30% in 3Q26, which should keep the YoY rate of change accelerating. Continued pricing strength is driving earnings upgrades and supporting valuation multiples, with both stocks trading at ~3.5-4.5x 2027e earnings. That said, we remain mindful that memory stocks are highly sensitive to changes in the rate of earnings and pricing momentum. With YoY DRAM pricing growth likely to plateau in 4Q26, and visibility on supply discipline beyond 2028 still evolving, we believe the sector may lack near-term cyclical catalysts until industry supply-demand dynamics become clearer. Nevertheless, valuation multiples could continue to re-rate higher, supported by favorable long-term agreements (LTAs) and sustained demand from AI infrastructure deployment.

Our North America Semiconductors analyst Joseph Moore reiterates his OW rating on Micron (MU.O, PT US$1,200). He believes a meaningful supply-demand imbalance remains unlikely as long as AI-related demand continues on its current trajectory. Currently, all DRAM suppliers, including CXMT, are expanding capacity as aggressively as possible. However, we estimate AI-related demand will continue to grow at a significantly faster rate than DRAM supply. At the same time, industry supply growth is constrained by the higher wafer intensity of high bandwidth memory (HBM) production and limited availability of EUV lithography tools. As a result, DRAM is increasingly becoming a key bottleneck in the broader AI infrastructure buildout. We expect CXMT to gain market share through faster capacity growth. However, we believe the company will prioritize domestic customers and demand within China, while geopolitical restrictions continue to limit access to more attractive international markets. That said, our investment case does not assume an indefinite period of memory shortages. We believe the current valuation already reflects a more normalized industry outlook. Micron currently trades at approximately 6x our estimate of peak run-rate EPS and 2.7x FY28E book value, while also offering significant capital return potential through the cycle. Our base-case model assumes approximately US$80bn of capital returns by mid-2028, equivalent to ~7% of the current market cap. However, with total FCF >US$300bn over the same period, capital returns could ultimately be 2-3x than our current assumption. In our view, the combination of substantial FCF generation, increasing earnings visibility from long-term agreements (LTAs), and a relatively low starting valuation should continue to support Micron's investment case, particularly if AI-related demand remains strong. For more details, see Semiconductors: Selloff of US memory stocks creates a compelling entry point (20 Jul 2026).

Our European Technology Semiconductors analyst Lee Simpson reiterates his OW rating on ASML (Top Pick, ASML.AX, PT €1,930). In his view, the industry's supply response ultimately depends on demand visibility and commercial commitment. Our Europe Tech Semis team expects ASML to continue expanding capacity in line with sustained EUV demand. The team model low NA EUV shipments increasing from 44 tools last year to more than 60 this year, with further capacity expansion possible as demand

100

70

40

30

10

Low NA EUV Shipments (FY25-FY27e)

• Logic

Memory

86

M

24

20

FY25

FY26e

52

FY27e visibility improves. Importantly, the constraint is not simply ASML's lead times. Rather, the industry must navigate the entire capacity expansion chain: memory manufacturers need to commit capital, equipment suppliers need to deliver tools, fabs need to be prepared, processes need to be qualified, yields need to ramp, and HBM packaging and testing capacity needs to scale. See our Europe Semiconductors Team's latest ASML notes here and here.

The post-Covid memory cycle offers an important lesson. Both Intel and Samsung deferred EUV-related orders after demand weakened and execution challenges delayed capacity ramps, resulting in order volatility across the semi equipment supply chain. In the current cycle, memory makers and hyperscalers may need to provide stronger LTAs, prepayments, and capacity commitments before suppliers undertake meaningful capacity expansion. This dynamic supports our view that supply relief is likely to materialize gradually rather than immediately. As a result, we believe memory market tightness could persist into 2027, even as EUV capacity continues to expand.

Exhibit 9: ASML Low NA EUV Shipments

Exhibit 10: ASML Order Book and Backlog

Source: Company data, Morgan Stanley Research estimates (e)

On the supply side, access to advanced lithography tools remains the key constraint.

EUV systems have effectively been unavailable to China since 2023, and ASML has never shipped an EUV tool to the country. We believe this is a major factor behind the apparent plateau in CXMT's leading-edge DRAM capacity expansion.

The proposed MATCH Act could further tighten these constraints by designating CXMT, YMTC, SMIC, Hua Hong, and Huawei as "covered facilities." The proposed legislation would restrict not only sales of DUV immersion lithography tools, but also servicing and maintenance of existing installed equipment.This is particularly important because DUV tool fleets require ongoing maintenance and support to sustain production yields. While the legislation remains under consideration and has not been enacted, it highlights the direction of policy risk. In our view, the regulatory trajectory continues to point toward tighter, rather than looser, constraints on China's ability to scale advanced memory production.

According to our Japan Semiconductor Production Equipment analyst Suzune Tamura, demand for Japanese SPE companies remains quite strong. Customers continue to accelerate equipment purchases and, in some cases, are willing to pay premiums for earlier delivery, particularly when their orders carry lower priority than those of other customers. Equipment suppliers and component vendors have been expanding production capacity since late last year. Even so, lead times continue to lengthen, underscoring the fact that demand remains ahead of available supply.

Source: Company data, Morgan Stanley Research

34

M

Among Japanese semiconductor equipment companies in Tamura-san's coverage universe, Kokusai Electric (OW, 6525.T, PT JPY8,300) has the highest exposure to CXMT. Sales to CXMT accounted for 20% of total sales in F3/25 and an estimated 6% in F3/26. Tokyo Electron (OW, 8035.T, PT JPY65,000) also has meaningful exposure to CXMT, with its share of CXMT's semiconductor equipment spending broadly in line with its overall market share. SCREEN Holdings (OW, 7735.T, PT JPY17,000), on the other hand, has lost share at CXMT over the past several years, as Chinese semiconductor manufacturers increased adoption of domestically produced equipment in line with government localization initiatives. More recently, however, SCREEN has begun to regain share at CXMT. As CXMT advances its HBM development efforts, process complexity continues to increase and wafer cleaning steps have become more critical.

HBM capacity expansion in China should also benefit semiconductor back-end equipment companies. Disco (OW, 6146.T, PT JPY78,000) maintains a leading position in grinding equipment used in through-silicon via (TSV) manufacturing. The TSV process includes edge trimming and back grinding, both of which require high-cleanliness equipment. These tools typically command prices of several hundred million yen per unit, compared with tens of millions of yen for conventional back-end equipment.

In memory testing, Advantest (OW, Top Pick, 6857.T, PT JPY36,000) has a leading position in the HBM testers market. While Teradyne (EW, TER.O, PT US$397, covered by Shane Brett) has expanded into high-speed HBM testing in recent years, Avantest maintains a strong presence in memory testing and is working to regain market share through a new product platform scheduled for launch next year.

Our Japan Semiconductor analyst Kazuo Yoshikawa has an OW rating on Renesas

Electronics (6723.T, PT JPY6,000). He believes the company is well positioned to benefit from AI-driven server demand through both CPU unit growth and rising memory interface content per server system. He forecasts a 29% CAGR in Renesas's memory interface revenue over CY25-28, driven by continued growth in server CPU shipments and increasing memory interface content as data-transfer speeds and bandwidth requirements between CPUs and memory continue to rise. He also estimates that memory interface products generate gross margins well above the company average, making this segment an important contributor to Renesas's overall gross margin expansion and earnings growth over the medium term.

The company is also well positioned for multiplexed rank dual in-line memory module (MRDIMM) adoption, with each module incorporating one multiplexed rank registering clock driver (MRCD) and ten multiplexed rank data buffers (MDB), representing ~3x the content of a conventional DDR5 RDIMM. Renesas has already introduced a full MRDIMM chipset solution and commenced mass production in 2025.

M

Exhibit 11: DRAM inventory levels

Source: TrendForce, Morgan Stanley Research.

Exhibit 12: Consolidated DRAM inventory level across different end-markets

Source: TrendForce, Morgan Stanley Research.

Greater China semiconductors - Any winners or losers?

Taiwan DRAM vendors Nanya Tech and Winbond should remain among the

beneficiaries: We maintain our OW ratings on Nanya Tech (2408.TW, NT$580) and Winbond (2344.TW, NT$288), given the industry's strong profitability and the continued re-rating of global memory stocks.

We reiterate our OW ratings on CXMT's equipment vendors NAURA, AMEC, and, ACMR: As CXMT expands capacity and migrates toward more advanced DRAM technologies, we expect its equipment intensity to rise. More complex, higher-aspect-ratio capacitor structures require greater etch and deposition intensity, while tighter process tolerances increase demand for cleaning, thermal processing, and inspection tools. On the back end, continued interconnect scaling should also support demand for deposition, wetcleaning, and electrochemical-plating equipment. At the same time, export restrictions and growing manufacturing experience are likely to accelerate qualification of domestic equipment. In our view, the combination of capacity expansion, technology migration, and rising localization should provide continued support for domestic wafer fabrication equipment (WFE) demand over the coming years.

• NAURA (OW, 002371.SZ, PT Rmb818): We see NAURA as one of the broadest beneficiaries of CXMT's expansion. Its ICP etchers can address selected conductor and dielectric-related etch applications, while its PVD tools should benefit from increasing requirements for additional metal deposition and barrier-layer formation. NAURA also has exposure to back-end wet-cleaning processes, where rising interconnect complexity is increasing contamination-control requirements. Capacity expansion should support tool volume growth, while technology migration toward more advanced process nodes could expand the number of addressable process steps. In addition, NAURA's broad portfolio creates opportunities for cross-selling and may support further wallet-share gains following successful initial production qualification.
• AMEC (OW, 688012.SS, PT Rmb488): We see AMEC as a key beneficiary of rising capacitor-hole etch requirements in advanced DRAM manufacturing. CCP etch plays a critical role in DRAM production, as increasingly complex capacitor structures require high-aspect-ratio dielectric etching with stringent profile, selectivity, and uniformity requirements. As China's leading domestic etch

M

equipment supplier, we believe AMEC is well positioned to benefit from ongoing localization efforts as CXMT advances its process technology. Successful qualification in more demanding capacitor etch applications would represent an important strategic milestone, supporting repeat orders, higher chamber intensity, and expansion of the company's addressable etch opportunity per wafer.

• ACMR (OW, ACMR.O, PT US$130): ACMR should benefit from rising cleaning intensity across both front-end and back-end DRAM manufacturing processes. As process technologies advance, tighter control of particles, residues, and metallic contamination becomes increasingly important, supporting demand for singlewafer cleaning solutions. On the back end, growing interconnect complexity should create additional opportunities for bot wet-cleaning and electrochemical-plating (ECP) equipment. We believe ACMR's ECP platform is well positioned to address selected copper-plating applications as CXMT expands capacity and advances its technology roadmap. Successful qualification in more critical process steps could support repeat orders and increase ACMR's content per wafer.

We also expect Chinese Design houses in our coverage to benefit

• GigaDevice (OW, 603986.SS, PT Rmb750, covered by Daniel Yen): We view GigaDevice as a key beneficiary of CXMT's continued expansion, given its position as one of CXMT's major partners in the specialty memory market. GigaDevice disclosed a Rmb5.711bn budget for related-party transactions with CXMT in 2026, significantly above the Rmb1.161bn budget for 2025 and the previously disclosed Rmb1.547bn budget for 1H26 (link). We also believe CXMT could increase DDR4 capacity by an additional 10k wpm by 2027, on top of its current ~8k wpm, to address strong domestic demand in China (link). Beyond wafer-capacity growth, Gigadevice also benefits from CXMT's process migrations, which supports improvements in performance, density, and cost competitiveness. We believe CXMT's continued capacity expansion and process migration will provide ongoing support for GigaDevice's DRAM business. In addition, we understand that GigaDevice is collaborating with CXMT on wafer-on-wafer (WoW) technology development, where CXMT could provide meaningful technology support (link).
• Montage (OW, 6809.HK, PT HK$432; 688008.SS, PT Rmb377, covered by Daniel Yen): As a global leader in memory interface chips, including RCD/DB and MRCD/MDB solutions, Montage currently derives a significant portion of its memory-interface business from the three leading global memory suppliers. However, as CXMT continues to expand capacity and makes further inroads into the server DRAM market, we believe it could create a meaningful opportunity for Montage to increase its penetration within China's domestic memory supply chain. A growing domestic server DRAM ecosystem could enable Montage to further strengthen its leadership position in memory interfaces while reducing its reliance on the three global memory vendors, i.e., Samsung, SK hynix, and Micron.

45,000

40,000

35,000

30,000|

25,000|

20,000

15,000

10,000

5,000

(USS mn)|

M

What could the bull case be for China DRAM (or a bear case for global DRAM)?

2030e|

(1) US cloud service providers (CSP) and consumer electronics brands start to source CXMT DRAM amid severe global memory shortages.

We expect CXMT to benefit from a broadening customer base as tight global memory supply continues to support demand for DRAM. In particular, select US CSPs and consumer electronics brands have begun evaluating or sourcing CXMT DRAM products, reflecting both ongoing supply constraints and a growing willingness among customers to diversify their supplier base. While qualification cycles and volume adoption may take time, incremental demand from international customers could improve CXMT's capacity utilization, strengthen its competitive position, and increase its strategic relevance within the global DRAM supply chain.

(2) Stronger demand from China's domestic AI server market, including both GPU and CPU

Demand from China's domestic AI server market also represents a key growth driver. Deployment of both GPU- and CPU-based AI servers is accelerating, supported by increasing investment in large language models (LLM), data centers, and sovereign AI infrastructure. This trend should drive sustained demand for high-capacity server DRAM, particularly as memory content per server increases. We believe CXMT is well positioned to capture a growing share of this demand given its status as a domestic supplier and improving product availability. Key variables remain product mix, performance requirements, and customer qualification timelines.

Besides DRAM-related development, we expect CXMT to explore additional growth opportunities, including 3D DRAM, NAND and advanced packaging technologies. While we have not seen any detailed capacity expansion plans for these initiatives, we believe they could provide additional revenue streams over the longer term and further diversify the company's business beyond conventional DRAM products.

Exhibit 13: China domestic CoWoS capacity to produce local AI GPU chips, which would need support from local HBM and LPDDR memory

Exhibit 14: We estimate China's CPU TAM to rise to US$42bn in 2030, which should drive substantial demand for CXMT DRAM

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

2025

202бe|

2027e|

2028e|

• Server

Agentic = DT = NB

M

(3) Global AI infrastructure demand continues to support DRAM pricing

At the industry level, robust global AI infrastructure spending should continue to support DRAM pricing. AI servers require substantially more memory than traditional enterprise servers, while strong demand for HBM and high-density DDR5 is absorbing leading-edge wafer capacity. This shift in capacity allocation is tightening supply across the broader DRAM market and supporting a favorable pricing environment for both advanced and mainstream DRAM products. In our view, sustained AI-driven demand should extend the current DRAM upcycle, although the pace of price increases may moderate as suppliers gradually expand capacity.

(4) Technology migration: Progress toward next-generation DRAM architectures

While restrictions on extreme ultraviolet (EUV) lithography tools are likely to slow CXMT's migration to a 15nm-class node, comparable to the 1y node of leading global DRAM suppliers, the company continues to explore alternative technology pathways. Our industry checks, together with commentary from DRAMeXchange, suggest that CXMT may pursue 3D DRAM architectures to advance toward its Gen5 technology node without relying on EUV tools.

Technology migration remains the primary medium-term constraint for CXMT. Restrictions on access to EUV lithography equipment are likely to delay its transition toward a 15nmclass process node, broadly comparable with the 1 γ node of leading global DRAM suppliers. However, our industry checks, together with indications from DRAMeXchange, suggest that CXMT may explore a three-dimensional DRAM architecture as an alternative pathway to improve density and extend process scaling, although we note that CXMT has not made such public announcements. If confirmed and successful, we think this approach could support its migration toward a fifth-generation technology platform and partially offset the limitations of conventional planar scaling. That said, execution risk remains elevated, as 3D DRAM development will require advances across process integration, yield optimization, cost management, and high-volume manufacturing.

M

Core Debate: Can CXMT Produce HBM3E Soon?

Why this debate? HBM is the high-entry barrier DRAM product, and if China can enter this market, it may disrupt global DRAM value.

Our recent checks suggest HBM3E supply will be the major bottleneck for China'a AI GPU shipment in 2027. Global memory vendors' HBM4 conversion may further reduce their supply of HBM3E to China vendors. It would be critical for China to own a local HBM3E supply for its new-generation AI GPUs.

Market's view: It would take time. China CXMT needs to build HBM technology from scratch, such as HBM2, and 8-Hi stacking, before it can achieve the mainstream HBM3E 12-Hi stacking.

Our view: Our checks suggest CXMT can produce 12-Hi stacking with wafer-onwafer technology. We believe CXMT can produce 3-4mn units of HBM3E stacks in 2027 despite low yields, sufficient to support ~0.8-1.0mn units of China AI GPU shipment. That said, we believe the global HBM supply will remain tight given strong global AI demand.

HBM: A high-end global DRAM segment that CXMT could potentially enter over time

CXMT has evolved from a portfolio focused primarily on legacy DRAM products toward a broader range of advanced memory solutions, including:

• DDR5 chips and PC memory modules.
• LPDDR5/LPDDR5X products for Chinese smartphones and mobile computing applications.
• Server memory offerings, including products supporting RDIMM and MRDIMM platforms.
• Longer-term ambitions in HBM4 and 3D/vertical-channel DRAM technologies.

The core earnings driver remains conventional DDR and LPDDR products. While HBM is strategically important and represents a potential longer-term growth opportunity, we do not yet view it as a proven earnings pillar. We forecast HBM3E mass production in 2027 and HBM4 in 2028. However, successful HBM commercialization depends on far more than producing a functional memory stack. Key factors include known-good-die yield, TSV yield, thermal behavior, packaging, test capacity, and accelerator qualification.

Exhibit 15: HBM roadmap

Source: Company data, Techinsights, Morgan Stanley Research. E = Morgan Stanley Research estimates

M

Exhibit 16: CXMT's HBM TSV capacity to reach 20kwpm in 2026e, 40kwpm in 2027e and 70kwpm in 2028e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

What's CXMT's edge in HBM?

HBM capacity is ramping rapidly, but the market remains highly concentrated among Samsung, SK hynix, and Micron. We forecast combined year-end HBM capacity among the three vendors will rise from ~500kwpm in 2026e to ~600kwpm in 2027e. Current export restrictions prohibit shipments of standalone raw HBM stacks into China. However, processors and accelerators that incorporate HBM can still be shipped provided they comply with applicable performance thresholds and export-control requirements.

We believe CXMT is accelerating its HBM roadmap to support growing demand from domestic AI accelerator developers. Our forecasts assume HBM2E production in 2026, followed by HBM3E 12-Hi in 2027e and HBM4 in 2028e. Our checks suggest CXMT can produce 12-Hi stacking with wafer-on-wafer technology, so 24GB per HBM chip is possible.

Exhibit 18: CXMT's HBM to account for 3.8%/4.4% if the total HBM market in 2026e and 2027e in terms of bit shipment

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 17: Global HBM capacity build

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

HBM consumes a disproportionate share of DRAM capacity

HBM is more than just a source of incremental DRAM demand. It also imposes a meaningful throughput penalty on the memory manufacturing ecosystem. Compared with conventional DRAM, HBM requires significantly larger die sizes and introduces additional complexity through 3D stacking, TSV integration, advanced packaging, extensive testing, and customer qualification. These additional manufacturing steps reduce effective bit output per wafer and increase demands on packaging and testing infrastructure.

Our model suggests the HBM bit-output penalty will increase from ~2.5x conventional DRAM wafers in 2021-23 to ~3.0x by 2028e. In other words, each unit of HBM bit output consumes materially more wafer capacity than conventional DRAM output.

At the same time, HBM is absorbing a growing share of the industry's most constrained wafer capacity. We estimate HBM's share of leading-edge DRAM wafer consumption will increase from ~6% of leading-edge memory wafers in 2023 to ~34% by 2028e. As AI adoption accelerates, AI customers are not only creating incremental memory demand, but also consuming an increasing share of the leading-edge capacity that would otherwise support DDR, LPDDR, and conventional server DRAM products.

Exhibit 19: HBM die penalty - bit-output is reduced due to much larger DRAM dies

Exhibit 20: HBM　cannibalization: Share of advanced DRAM wafer is rising sharply

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Even with this capacity expansion, supply is likely to remain constrained because HBM availability depends on far more than wafer starts. Effective HBM supply is also a function of yield, utilization, TSV capability, advanced packaging capacity, testing infrastructure, and customer qualification. Qualification remains particularly important, as not all HBM output can be deployed across every AI accelerator platform or customer environment.

After incorporating CXMT's projected HBM capacity expansion into our supply estimates and factoring in expected growth in China's AI GPU shipments, we still forecast a tight HBM market. Our model indicates an HBM supply sufficiency ratio of approximately 17% in 2026e and 15% in 2027e.

M

Exhibit 21: HBM TAM: Base case scenario

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 22: Even with CXMT's capacity adding in, the overall HBM market remains undersupply

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

CXMT (688825.SS): Key Beneficiary Of China's Memory Localization Trend; Initiate At OW

CXMT has evolved from China's first large-scale domestic DRAM producer into the world's fourth-largest DRAM manufacturer and a key beneficiary of the country's memory localization strategy. The company is expanding across DDR5, LPDDR5/5X, server DRAM, and HBM, supported by rapid capacity growth, improving yields and sustained R&D investment.

We forecast total DRAM capacity to rise from 180k wpm in 2025 to 500k wpm by 2028. At the same time, the transition toward G4B production and the planned ramp of HBM3e and HBM4 should enhance CXMT's product mix and strengthen its competitive position across higher-value memory segments. In addition, growing adoption by Chinese CSPs, smartphone manufacturers, and PC OEMs, together with structurally strong demand for AI-server memory, should support utilization rates and market-share gains. However, CXMT's earnings are subject to several risks, including DRAM pricing, elevated capex requirements, rising depreciation expenses, export-controls, and execution risk associated with next-generation process technology migration.

Earnings: Our strong 2026-28 earnings outlook is supported by strong capacity additions, elevated DRAM pricing, and substantial operating leverage. We project revenue to rise at a 140% CAGR during 2025-2028e, while maintaining gross margin above 75% and operating margin of ~70-72%. We expect the company's increasing exposure to higher-value products, including DDR5, server DRAM, and HBM, to further support earnings growth, while scale should help offset higher depreciation expenses and continued R&D investment associated with capacity expansion and technology migration.

Valuation: We derive our price target of Rmb88 using a residual income model, which we believe best captures CXMT's long-term value creation potential and ability to expand market share. Our price target implies approximately 18.5x 2027E P/E, representing a premium to global memory peers. We believe this premium is justified by CXMT's stronger growth profile, supported by faster capacity expansion, advantages from China's memory localization trend, and exposure to growing demand from the country's AI infrastructure ecosystem.

Key downside risks include (1) a sharper-than-expected decline in DRAM prices, (2) delays in G5, HBM or 3D DRAM technology development, (3) tighter restrictions on semiconductor equipment, materials, or servicing, (4) weaker-than-expected demand from major customers, and (5) higher-than-expected depreciation, inventory write-downs, or capital intensity.

M

CXMT: Financial Summary

Exhibit 23: CXMT: Financial summary

E = Morgan Stanley Research Estimates

Source: Morgan Stanley Research, Company Data

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Risk Reward - CXMT Corp (688825.SS) Risk Reward - CXMT Corp (688825.SS)

China DRAM Going Mainstream; OW

Rmb88.00 PRICE TARGET

Base case, residual income model. Key assumptions:

• 9.9% CoE (1.2 beta, 2.0% risk-free rate, and 6.6% risk premium)
• 51% payout ratio
• 11.0% medium-term growth rate
• 3% terminal growth rate

We believe these assumptions are justified, given CXMT's fast growth, capacity expansion, localization tailwinds and AI-memory exposure. Base case PT implies an 18.5x 2027e P/E vs global peers' 4.4x 2027e P/E; the premium reflects CXMT's stronger growth profile.

RISK REWARD CHART

Key:

Source: Refinitiv, Morgan Stanley Research

BULL CASE

27.7x 2027e P/E

We assume stronger sales/earnings growth, with faster global market share expansion:

(1) 200% revenue CAGR, 2025-28e, as we assume a stronger-than-expected DRAM commodity cycle, supported by AI inference demand moving into 2027 and beyond.

• (2) faster-than-expected HBM ramp-up.

(3) Gross margin improves to over 90%, driven by economies of scale and CXMT's technology competitiveness.

• (4) >800% earnings CAGR, 2025-28e.

Rmb132.00

BASE CASE

18.5x 2027e P/E

We forecast strong revenue/earnings growth, with global market share expansion.

• (1) 140% revenue CAGR, 2025-28e, as we assume DRAM pricing remains strong, supported by AI demand
• (2) moderate pace of HBM ramp-up
• (3) 2026-28e gross margin of over 75%, driven by economies of scale and CXMT's technology competitiveness
• (4) 480% earnings CAGR, 2025-28e

Rmb88.00

OVERWEIGHT THESIS

• CXMT is now the world's #4 DRAM maker and a key beneficiary of China's memory localization push.
• We forecast that DRAM capacity will rise from 180k wpm in 2025 to 500k by 2028, with G4b, DDR5, LPDDR5/5X, server DRAM and HBM3e/HBM4 improving mix.
• We expect 2025-28E revenue CAGR of 140%, 2026-28E GM >75% and OPM of ~70-72%, supported by pricing, scale and higher-value products.
• Our Rmb88 price target is based on a residual income model and implies 18.5x 2027E P/E. We see the premium to global memory peers as justified by CXMT's faster growth, capacity expansion, localization tailwinds and AI-memory exposure.
• Risks: DRAM pricing, capex/depreciation, export controls and migration execution.

Risk Reward Themes

Pricing Power:

Positive Positive

Secular Growth:

View descriptions of Risk Rewards Themes here

BEAR CASE

Rmb70.00

14.8x 2027e P/E

We assume revenue/earnings growth is slower, with slower global market share expansion.

(1) 80% revenue CAGR, 2025-28e, as we assume a period of AI computing digestion in 2028, leading to more aggressive margin erosion for the HBM segment owing to competitors' aggressive capacity expansion and pricing competition, as well as faster DRAM pricing drops with muted demand recovery.

(2) customer bargaining power pushes gross margin below 40%.

(3) 200% earnings CAGR, 2025-28e.

M

Risk Reward - CXMT Corp (688825.SS)

KEY EARNINGS INPUTS

INVESTMENT DRIVERS

• Commodity DRAM pricing
• HBM capacity expansion and tech migration
• Demand from major customers

GLOBAL REVENUE EXPOSURE

RISKS TO PT/RATING

RISKS TO UPSIDE

• resilient DRAM pricing, supported by AI demand
• faster-than-expected tech migration
• export control relief
• stronger demand from major customers

RISKS TO DOWNSIDE

• earlier-than-expected global DRAM price declines
• CXMT delays in new products, HBM or 3D DRAM development
• tighter restrictions on China's fabs for semi equipment, materials, or servicing
• weaker demand from major China customers

MS ESTIMATES VS. CONSENSUS

M

CXMT: Investment Positives and Concerns

Investment positives

China memory localization and AI demand create a structural growth runway: CXMT

has evolved into the world's fourth-largest DRAM supplier, supported by an expanding product portfolio that includes DDR5, LPDDR5/5X, server DRAM, and emerging HBM products. Adoption by leading Chinese cloud service providers, smartphone manufacturers, and PC OEMs, combined with improving product performance and production yields, suggests that CXMT is becoming increasingly competitive across mainstream DRAM applications rather than serving primarily lower-end markets.

Rapid capacity expansion should drive significant revenue growth and operating leverage: We forecast total DRAM capacity to increase from 180k wafers per month (wpm) in 2025 to 500k wpm by 2028, while bit shipments grow at a 46% CAGR over 2025-28e. This scale expansion, combined with favorable DRAM pricing and a richer product mix, should support substantial earnings growth. Following its return to profitability in 2025, we project gross margins above 75% and operating margins of approximately 70-72% during 2026-28e.

A structurally tight global DRAM market provides a favorable pricing backdrop for CXMT's capacity ramp: AI servers are becoming increasingly memory intensive, with server applications projected to account for 59% of DRAM bit demand by 2028. At the same time, HBM production is absorbing a growing share of leading-edge DRAM wafer capacity, tightening supply across the broader memory market. As a result, we expect the global DRAM market to remain undersupplied through 2027, supporting both pricing and utilization even as CXMT expands capacity. CXMT's position within China's domestic memory supply chain, its planned HBM3E and HBM4 roadmap, and its potential adoption of vertical-channel transistor (VCT) or 3D DRAM technologies could provide further upside through market-share gains and technology advancement. These factors should strengthen CXMT's competitive position as China's memory localization strategy and AI infrastructure investment continue to drive demand growth.

Investment concerns

Accumulated losses and a capital-intensive business model could constrain financial

flexibility: As of the end of 2025, CXMT had accumulated losses of ~Rmb36.7bn, reflecting the substantial fixed-asset investment and sustained R&D spending required in the DRAM industry. Although the company returned to profitability in 4Q25, its assetheavy business model continues to require significant investment in capacity expansion and could limit balance-sheet flexibility if industry conditions weaken.

Rising depreciation and inventory exposure could amplify earnings volatility through the cycle: The carrying value of fixed assets increased from Rmb84.5bn in 2023 to Rmb183.0bn in 2025, while annual depreciation expense rose from Rmb10.6bn to Rmb24.7bn over the same period. Continued capacity expansion is likely to further increase depreciation charges. At the same time, DRAM's commodity characteristics expose the company to inventory impairment risk during periods of pricing weakness. As

M

demonstrated during the 2023 industry downturn, sharp declines in memory prices can lead to significant inventory write-downs and earnings volatility.

Technology, customer concentration, and external supply-chain constraints remain important execution risks: DRAM development involves high technical barriers, long qualification cycles, and rapid product iteration. As a result, delays or execution challenges in process migration could weaken CXMT's competitiveness. The company also derives a relatively high proportion of sales from its five largest customers. Any reduction in orders from its key customers could have a disproportionate impact on revenue growth and utilization. In addition, restrictions affecting semiconductor equipment, materials, technology transfer or servicing could disrupt production, slow capacity expansion, and complicate CXMT's longer-term technology roadmap.

M

CXMT: Financial Outlook

Capacity and production

CXMT is now expanding its DRAM manufacturing footprint, with near-term production concentrated in Hefei and Beijing, while Shanghai represents a potential longer-term growth location. We forecast CXMT's DRAM capacity to increase from 180k wpm in 2025 to 300k wpm in 2026, followed by annual capacity additions of ~100k wpm, reaching 800k wpm by 2031.

CXMT previously focused on Gen3 (~18nm-class) and Gen4 (~17nm-class) DRAM production, where we estimate yields have exceeded 90% for both nodes. Beginning in 2Q26, we believe the company completed its transition to 100% Gen4B (~16nm-class) production, with yields also reaching approximately 92% or higher. The transition to Gen5 (~15nm-class) is likely to be more challenging, as it requires help from DUV-based scaling techniques or successful adoption of alternative architectures such as VCT technology. Given both pathways face meaningful hurdles, including export-control constraints on advanced DUV tools and process-integration challenges associated with next-generation device architectures. As a result, we expect Gen5 mass production to begin in 2H28e.

On the other hand, we also expect CXMT to ramp its HBM capacity rapidly, from 20k wpm in 2026e to 70k wpm in 2028e. Our projections assume the company's HBM roadmap progresses from HBM2E in 2026, to HBM3E 12-Hi in 2027, and ultimately HBM4 in 2028.

Exhibit 24: CXMT to ramp its capacity from 180k wpm in 2025, 300k wpm in 2026 and expanding at an annual pace of 100k wpm to 800k wpm in 2031.

Exhibit 25: Total bit shipments to rise at a 40% CAGR during 2025-2028e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Exhibit 26: G4b capacity to be CXMT's main focus in the next few years …

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 28: CXMT to ramp up HBM capacity from 20k wpm in 2026e to 70k wpm in 2028e

HBM TSV capacity

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Revenue

From 2023 to 2025, CXMT delivered rapid revenue growth, with revenue of Rmb9.087bn, Rmb24.178bn, and Rmb61.799bn, respectively. Net profit attributable to shareholders improved from a loss of Rmb16.340bn in 2023 to a loss of Rmb7.145bnin 2024, before turning positive at Rmb1.875bn, marking the company's first profitable year. On a nonrecurring-items-adjusted basis, net profit attributable to shareholders reached Rmb5.3bn in 2025, suggesting a stronger underlying earnings profile than reported net income alone. Momentum appears to have continued into 2026. In 1Q26, CXMT reported Rmb50.8bn of revenue, up 719.13% YoY.

The company guides for 1H26 revenue of Rmb110-120bn. However, we project revenue of at least Rmb140bn, supported by continued capacity expansion and further increases in DRAM ASPs. According to the prospectus, the company's revenue outlook reflects the sharp rise in DRAM prices since 2H25, growing production and shipment volumes, and ongoing product-mix improvement. Looking beyond 2026, we expect CXMT to benefit from a combination of rapid capacity expansion, favorable DRAM pricing, and increasing contributions from DDR5, LPDDR5X, server DRAM, and HBM products. As a result, we forecast revenue growth at a 140% CAGR over 2025-28e.

Exhibit 27: G5 capacity to start contributing from 2028/29

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Exhibit 29: CXMT's revenue to rise at a 140% CAGR, 2025-28e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Gross profit and operating margin

CXMT's gross margin is highly sensitive to the DRAM industry cycle. During the reporting period, gross margin from its core business was -2.2% in 2023, 5.0% in 2024, and 41.0% in 2025. The sharp improvement in 2025 was driven by rising DRAM product prices, economies of scale, and product mix improvements.

However, DRAM remains a cyclical industry with commodity-like characteristics. If macroeconomic conditions weaken, end-market demand softens, or industry supply eventually exceeds demand, DRAM pricing and operating performance could fluctuate significantly. That said, we currently see no indication of meaningful DRAM price declines in the near term. Despite the higher depreciation burden associated with CXMT's planned capacity additions of approximately 100k wpm annually, we forecast the company will maintain gross margins above 70% during 2026-28e. Assuming disciplined operatingexpense management and continued operating leverage from higher utilization and production scale, we forecast operating margins to be 70-72% during 2026-28e.

M

Exhibit 30: CXMT's OpM to be 70-72% during 2026-28e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Net profit and EPS

From 2023 to 2025, CXMT's net profit attributable to shareholders improved from a loss of Rmb16.3bn to a profit of Rmb1.9bn, with the company achieving profitability in 2025 amid a favorable DRAM upcycle. Net profit attributable to shareholders was Rmb-16.3bn in 2023, Rmb-7.1bn in 2024, and Rmb1.9bn in 2025. We expect earnings to continue benefiting from operating leverage as revenue growth outpaces fixed-cost growth.

We expect CXMT to gradually consolidate ownership of externally financed capacity and facilities through potential buybacks or other ownership-integration arrangements. As a result, we use net profit and EPS including minority interests in our valuation framework.

Exhibit 31: CXMT's EPS to rise at a 480% CAGR, 2025-2028e

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

CXMT: Quarterly Financials

Exhibit 32: CXMT: Quarterly financials

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

CXMT: Valuation Methodology and Peer Comparison

Our price target is Rmb88: We derive our price target from a residual income (RI) model, as we believe this methodology is best suited to capturing CXMT's long-term value. We assume a 9.9% cost of equity (beta 1.2, risk-free rate 2.0% and risk premium 6.6%), a payout ratio of 51%, a medium-term growth rate of 11.0%, and a terminal growth rate of 3.0%, all of which are in line with other Greater China foundry and Asia DRAM names under our coverage universe.

Our price target implies 18.5x 2027e P/E, representing a premium to global memory players. We believe this premium is justified by CXMT's stronger long-term growth profile, supported by sustained market-share gains in the global DRAM industry and growing participation in China's domestic memory market.

Exhibit 33: CXMT: RI model

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 34: Global DRAM supply chain valuation comparison

Source: Company data, FactSet, Morgan Stanley Research. E = Morgan Stanley Research estimates. Note: As of the market close on Aug 21, 2026.

Bull/bear case scenarios

Our bull case scenario value is Rmb132, implying 28x 2027e P/E: In this scenario, we assume stronger sales/earnings growth, with faster global market share expansion:

• (1) 200% revenue CAGR, 2025-28e, as we assume a stronger-than-expected DRAM commodity cycle, supported by AI inference demand moving into 2027 and beyond.
• (2) faster-than-expected HBM ramp-up.
• (3) gross margin improves to over 90%, thanks to economies of scale and CXMT's technology competitiveness.
• (4) >800% earnings CAGR, 2025-28e.

M

Our bear case scenario value is Rmb70, implying 15x 2027e P/E: In this scenario, we assume revenue/earnings growth is slower, with slower global market share expansion.

(1) 80% revenue CAGR, 2025-28e, as we assume a period of AI computing digestion in 2028, leading to more aggressive margin erosion for the HBM segment owing to competitors' aggressive capacity expansion and pricing competition, as well as faster DRAM pricing drops with muted demand recovery.

(2) customer bargaining power pushes gross margin below 40%.

(3) 200% earnings CAGR, 2025-28e.

M

CXMT: Company Overview

CXMT is China's largest and most advanced integrated DRAM company, with capabilities spanning DRAM R&D, design, manufacturing, and sales under an IDM model. Since its establishment in 2016, the company has focused exclusively on DRAM and has successfully advanced through four process generations, achieving mass production at each stage. Its product portfolio has evolved from DDR4 and LPDDR4X to DDR5 and LPDDR5/5X. According to the prospectus, the company's core products and process technologies have reached internationally advanced levels..

CXMT operates under an IDM model and retains control over the most critical parts of the DRAM value chain, particularly design, process technology development, and wafer manufacturing. The company strongly emphasizes process control, manufacturing learning, and yield improvement. Packaging is primarily outsourced, while testing follows a hybrid model that combines internal capabilities with external partners.A portion of testing activities is conducted by the company's wholly owned subsidiary, CXMT Product Hefei. Module assembly and testing similarly combine in-house operations with outsourced providers, allowing the company to balance operational control with capacity flexibility.

CXMT adopts a 'leapfrog R&D' strategy and has successfully advanced through multiple process generations and product iterations within a relatively short period. Its R&D process follows a structured framework comprising four stages: concept, planning, development, and verification. This process spans the full product-development cycle, from initial product specification through commercial deployment. Key activities include product requirement definition, technical feasibility assessment, layout and circuit design, simulation verification, tape-out, engineering verification, and customer verification.

CXMT places strong emphasis on independent R&D. Its R&D expenditure increased from Rmb4.7bn in 2023 to Rmb6.3bn in 2024 and Rmb9.6bn in 2025, totaling approximately Rmb20.6bn over the three-year period. This represented 21.7% of cumulative revenue during the same period. As of the end of 2025, CXMT employed 6,259 R&D personnel, accounting for 32.4% of total headcount.

In terms of intellectual property, as of the end of 2025, CXMT held 3,929 domestic patents, including 3,165 invention patents, as well as 3,043 overseas patents. The company has stated that its core technologies cover DRAM product design, manufacturing processes, packaging and testing, module design, and applications.

Key shareholders

Immediately after CXMT's July 27 IPO (before exercise of the over-allotment option), CXMT's share capital is approximately 66.881 billion shares. Qinghui Jidian remains the largest shareholder with a 19.50% interest, followed by Changxin Integrated at 10.54%, National IC Industry Investment Fund II at 7.86%, Hefei Jixin at 7.53% and Anhui Investment Group at 7.12%. These five shareholders collectively hold 52.55% of the post-IPO share capital.

M

The five largest shareholders represent a range of institutional ownership types. Qinghui Jidian is an investment partnership established solely to invest in CXMT and is disclosed as having no actual controller. Its partnership interests are held by Xinrui Investment, Changxin Integrated and its general partner, with respective interests of 51.09%, 48.90% and 0.01%. Changxin Integrated is wholly owned by Hefei Industrial Investment Holding Group and is ultimately controlled by the Hefei municipal state-owned assets authority. National IC Industry Investment Fund II is a national integrated-circuit investment fund and is disclosed as having no actual controller. Hefei Jixin is CXMT's direct employee shareholding platform, while Anhui Investment Group is wholly owned and controlled by the Anhui provincial state-owned assets authority. The principal shareholder group therefore includes municipal and provincial state-owned capital, national semiconductor investment capital, employee ownership and dedicated investment vehicles.

Immediately before the IPO, all 60 of CXMT's direct shareholders were institutional investors, with no natural person holding shares directly. Beyond the five largest shareholders, the post-IPO register includes technology and semiconductor companies, local industrial investment vehicles, insurers, financial asset investment companies, private equity funds and foreign institutions. Among the larger named investors, Alibaba Cloud holds 3.47% after the initial offering, Hefei Chantou No. 1 holds 1.66% and GigaDevice holds 1.62%. The shareholder base also includes Alibaba Network, insurance companies, bank-affiliated investment entities, national and regional investment funds, and two foreign institutional shareholders.

The IPO further broadens the shareholder base through strategic allocations. The final strategic placement comprises approximately 1.667 billion shares, equivalent to 24.93% of the initial offering. The strategic investors include companies with existing or prospective business cooperation with CXMT, long-term institutional investors such as insurance companies and national investment funds, subsidiaries of the joint sponsors participating in the mandatory co-investment arrangements, and four asset management plans established for CXMT's senior management and core employees. The strategic placement therefore adds a further group of industry participants, long-duration financial investors and employee-linked investors to the post-IPO ownership structure.

CXMT is disclosed as having no controlling shareholder and no actual controller. Its board consists of 11 directors, including four independent directors. Of the seven nonindependent directors, two are nominated by Qinghui Jidian, two by National IC Industry Investment Fund II, one by Hefei Jixin, one by Anhui Investment Group and one is an employee representative. The two directors attributed to Qinghui Jidian are respectively recommended by its general partner and by Changxin Integrated. CXMT also has no special-voting shares, contractual control arrangement or similar governance structure.

Chairman Yiming Zhu beneficially owns 1,590,303,801 CXMT shares, comprising 858,879 shares held indirectly through Qinghui Jidian, 1,535,841,835 shares through Hefei Jixin No. 41 Enterprise Management Partnership, an upper-tier vehicle within the Hefei Jixin employee shareholding structure, and 53,603,087 shares through GigaDevice. This represents approximately 2.38% of CXMT's 66.881 billion shares post-IPO, before exercise of the over-allotment option.

M

Exhibit 35: Shareholding structure immediately after the IPO (before exercise of the over-allotment option)

Source: Company data, Morgan Stanley Research.

Key product portfolio

DDR series: This includes DDR4 and DDR5, mainly used in servers, desktops, laptops, and other consumer electronics. CXMT's DDR5 portfolio includes 16Gb, 24Gb, and 32Gb products, with data-transfer speeds of up to 8,000Mbps, supporting memory modules ranging from 8GB to 128GB. The company also notes that since the end of 2024, it discontinued production of its own DDR4 products and redirected resources toward newer-generation products such as DDR5.

LPDDR series: This includes LPDDR4X and LPDDR5/5X, mainly used in smartphones, tablets, laptops, wearable devices, and other consumer electronics. LPDDR5/5X features higher capacity, faster speed, lower power consumption, and enhanced data reliability. Based on our supply chain checks, relevant LPDDR5/5X products have already entered the supply chains of leading smartphone brands, including Xiaomi and Transsion.

M

In addition to DRAM chips, CXMT offers a range of product solutions such as DRAM wafers and DRAM modules. Its module products include RDIMM, MRDIMM, UDIMM/ CUDIMM, SODIMM/CSODIMM, and LPCAMM, serving applications such as servers, workstations, desktops, laptops, and high-performance mobile workstations.

Exhibit 36: Revenue breakdown, by product

Exhibit 37: DRAM-related products remain CXMT's focus

Source: Company data, Morgan Stanley Research.

Source: Company data, Morgan Stanley Research.

Key suppliers and customers

CXMT's products are used in servers, mobile devices, personal computers, and smart vehicles. According to the prospectus, core customers include Alibaba Cloud, ByteDance, Tencent, Lenovo, Xiaomi, Transsion, Honor, OPPO, and vivo.

CXMT uses a combination of direct sales and distribution channels, with the distribution model serving as one of its primary go-to-market strategies. According to the company, this approach helps accelerate channel development, expand market share, improve inventory and cash conversion, and enhance operational flexibility and responsiveness. During the reporting period, CXMT's largest customers consisted primarily of wellestablished distributors and leading downstream manufacturers. While no single customer accounted for more than 50% of revenue, customer concentration remained relatively high. Revenue contribution from the top five customers was approximately 74.1% in 2023, 67.3% in 2024, and 68.1% in 2025.

CXMT's main raw materials include chemicals, photoresists, silicon wafers, gases, targets, spare parts, and other semiconductor manufacturing materials. To manage supply-chain stability and quality control, the company has established a comprehensive suppliermanagement framework that includes supplier qualification, performance evaluation, regular audits, elimination procedures, and blacklist mechanisms. Suppliers are evaluated across multiple dimensions, including quality, delivery, cost, technology, service, and safety. We summarize CXMT's key suppliers and ecosystem partners across the semiconductor equipment, materials, packaging, testing, and memory-module segments in Exhibit 38 .

M

Exhibit 38: CXMT: Semi equipment supplier list (as of the date of this report)

Source: Company data, Morgan Stanley Research.

M

Exhibit 39: CXMT: Semi material supplier list (as of the date of this report)

Source: Company data, Morgan Stanley Research.

M

Exhibit 40: CXMT: Packaging, testing and memory modules supplier list (as of the date of this report)

Source: Company data, Morgan Stanley Research.

Key competitors

The DRAM industry is highly concentrated. Samsung Electronics, SK hynix, and Micron Technology have long held a combined global market share of over 90% in terms of revenue. According to Omdia data, in 2025, Samsung, SK hynix, and Micron had respective global DRAM revenue market shares of 33.96%, 34.48%, and 23.41%. Based on our estimates, CXMT's global DRAM revenue market share reached 7.67% in 4Q25, which we think indicates that it has gradually entered the ranks of major industry players.

We believe the company's competitive advantages mainly come from its position as China's domestic DRAM leader, capacity scale, comprehensive product portfolio, independent R&D capabilities, customer resources, and supply-chain ecosystem collaboration. However, we believe CXMT still lags behind the top three international manufacturers in terms of overall scale, technology accumulation, and product mix.

Management team

Yiming Zhu, the founder of CXMT, studied physics at Tsinghua University and later pursued graduate studies in electrical engineering at SUNY Stony Brook. After working in Silicon Valley, including at MoSys, he returned to China in 2005 and founded GigaDevice, which grew into a leading fabless supplier of NOR flash memory and microcontrollers. In 2016, he moved into the much larger but more demanding DRAM market through the Hefei-backed '506' project, which later became CXMT. Since DRAM manufacturing requires deep process expertise, advanced technology capabilities, and access to extensive intellectual property, CXMT initially benefited from technology and talent associated with the former European memory manufacturer Qimonda. By licensing a substantial portfolio of Qimonda patents and gaining access to extensive technical documentation, CXMT established an important foundation for its DRAM development efforts. This provided access to key DRAM technologies, including buried-wordline (BWL) and stacked-capacitor architectures. Building on this foundation, CXMT successfully scaled these technologies from approximately the 46nm-class to increasingly advanced process generations, ultimately progressing toward today's 16nm-class Gen4B production.

M

Still, CXMT's long-term progress depended not only on inherited intellectual property, but also on the engineering talent required to convert that knowledge into commercial production. The company recruited former Qimonda engineers, including veteran technologist Karl-Heinz Kuesters, whose manufacturing experience helped transform archived designs into technology concepts into commercially viable products with acceptable yields. CXMT also attracted professionals with backgrounds at Micron, SanDisk, Applied Materials, and semiconductor firms in South Korea and Taiwan. These hires strengthened the company's capabilities in process integration, materials, yield improvement, and next-generation memory. Over time, the combination of experienced international talent and returning Chinese engineers enabled CXMT to build an increasingly self-sufficient R&D and manufacturing organization. This internal capability supported the company's progression through multiple DRAM process generations, expansion from DDR4 and LPDDR4X into DDR5 and LPDDR5/5X, and more recently its development efforts in server DRAM and HBM. However, this transformation was neither quick nor inexpensive. It required nearly a decade of substantial capital investment, intensive R&D spending, process development, capacity expansion, and cumulative manufacturing learning before the company reached profitability.

M

Exhibit 41: Management team details (as of July 27, 2026)

Source: Company data, Morgan Stanley Research.

M

Appendix 1: Global DRAM Cycle: A Multi-year Bottleneck

The explosive demand for AI infrastructure has caused supply constraints in all types of memory (DRAM, HBM, NAND, HDD). As AI increasingly moves beyond data centers into edge devices and the physical world, it could trigger a further surge in demand for memory chips. In that scenario, these components would no longer be seen as commodity chips but as critical pillars of the global digital infrastructure.

AI and data centers are becoming the dominant buyers of memory - we expect the server share of DRAM demand to rise from 37% in 2023 to 59% in 2028e. The result is a structurally tighter market where non-AI buyers may face weaker access to supply for an extended period.

Memory is the working space a computer uses to hold data that it is actively processing the digital equivalent of the desk you are working on right now, as opposed to the filing cabinet behind you. Three types of memory matter for the AI economy:

• DRAM (Dynamic Random Access Memory) is the fast but expensive working memory used by every server, PC and smartphone. It is what the CPU uses to hold and fetch the data it is computing.
• HBM (High Bandwidth Memory) is a premium type of DRAM focused on providing maximum data bandwidth. HBM is DRAM stacked vertically and bolted directly onto AI chips - it is the memory layer that makes large AI models possible, feeding data fast enough to limit AI accelerator downtime. Think of DRAM as a single lane road and HBM as a 12-lane highway for data transmission.
• NAND (Not-And) is the storage memory used in solid-state drives (SSDs), which sit between working memory and slower HDD disk storage. NAND holds the trained model weights, the training datasets, and the cached results that AI systems retrieve from. It is also used for picture, video and data storage.

All three types of memory matter as Al workloads use all three at once - and in massive quantities. Next-generation rack-scale AI systems, such as NVIDIA's Vera Rubin NVL72, contain 20.7TB of HBM attached to GPUs and 54TB of LPDDR5X serving the CPUs, before considering local or external SSD storage.

AI demand is scaling across three layers at once - more memory per chip, more chips per system, and more systems per cluster. Agentic AI adds another layer of pressure through longer context windows, larger conversation memory and repeated inference across multiple agents. The result is a structural increase in memory intensity across all aspects of the AI stack.

Camnoll 4&.

Faster, more expensive

M

CPU orchestration

Agentic Al workflows

General computing

Model weights

Training datasets

Cached Al results aee. AllInIAAAenel

Exhibit 42: AI servers are becoming memory systems

and smartphone

Source: NVIDIA, Morgan Stanley Research. Note: Rubin NVL72 memory capacity based on platform disclosures. NAND/eSSD capacity varies by OEM and storage architecture.

Supplier map: Strategic memory remains led by the Big 3; China scales in commodity pools

The supplier landscape is highly segmented by product. HBM and advanced/server DRAM remain concentrated among Samsung, SK hynix and Micron, with SK hynix leading HBM today, Samsung positioned as the broadest DRAM/NAND supplier and potential HBM recovery candidate, and Micron emerging as the key US-headquartered strategic supplier. NAND supply is broader, with Samsung, Kioxia/SanDisk, SK hynix/Solidigm, Micron and YMTC all relevant. China is becoming more important through CXMT in commodity DRAM/LPDDR and YMTC in NAND, but neither is a near-term global HBM relief valve. The key implication is that supply relief is uneven: China can help in selected commodity memory markets, while frontier AI memory remains concentrated in the Big 3.

DRAM

54 TB per rack

NAND

100+ TB per rack (configurable)

Atata. Oohis tilt 70 moaeaet

salmon to. suppler positon, by mellory type

Supplier position by memory type

Market position by product pool: HBM remains Big 3-led; China emerges in commodity DRAM/NAND

• Market leader O Strong • Present / growing • Limited / emerging

SK hynix

South Korea

Samsung

South Korea

Micron

USA

Kioxia / SanDisk

Japan / USA

CXMT

China, state-backed

YMTC

China, state-backed

Not present

M

HBM tightens conventional supply.

Losing low-end to CXMT.

Enterprise

NAND | SSD

Present

Enterprise SSD via Solidigm.

Exhibit 43: Supplier position, by memory type

Recovery candidate

Leader

HBME yield issues cost share. HBM4 Largest conventional DRAM capacity.

is a critical recovery moment.

Growing fast

18M3E qualified for NVIDIA H20

nly US-headquartered HBM supplie

NAND only - no DRAM

Attempting

HBM3 development / sampling, but not a near-term global HBM relief valve.

NAND only - no DRAM

Leader

Largest LPDDR supplier. Inside most

Leader argest NAND capaaity globali

Consumer

NAND

Present

Lower priority - HBM and server absorb most capacity.

Leader eMMC, UFS, client SSDs across all

Source: Company commentary, Morgan Stanley Research estimates. Note: Market positions are illustrative and intended to frame supplier participation by memory type.

Memory cycle monitor

Where are we in the DRAM cycle?

Duration is getting shorter now. Historically, the memory segment tended to be 4-8 quarters in a downturn and 4-9 quarters in an upturn. The slowdown tends to be shorter than the upturn in the memory cycle because it is easy to cancel orders and rationalize inventory but harder to bring back capacity. The 2018 share-price correction lasted only four quarters, from 1Q18 to 1Q19. The most recent downcycle, in 2024, lasted for only 8 months before stocks found a bottom, and with tariffs and AI demand supporting an increase in pricing again, share prices also rebounded sharply. We are now 11 months into the current upcycle, which began in September 2025.

Exhibit 44: DRAM cycles in perspective over the past 30 years

Source: FactSet, Morgan Stanley Research.

M

How does the current supply-demand balance compare with previous cycles, and where are we in the cycle?

The current memory cycle is materially tighter than the 2023-24 cycle and is being driven less by a broad-based demand rebound than by AI-related capacity displacement.

In 2023, most conventional memory markets were in deep oversupply following the postCOVID inventory correction. Conditions began to normalize in 2024 as inventories cleared, pricing stabilized, and suppliers restored discipline. Today, however, HBM is effectively sold out, server DRAM is tight, and enterprise NAND is tightening as wafer capacity and investment are redirected toward higher-value AI products.

The key difference versus previous cycles is that the shortage is increasingly structural rather than purely cyclical. HBM demand is growing rapidly under multi-year commitments, hyperscalers are consuming more DRAM per system, and added HBM capacity can constrain supply for conventional DRAM and NAND. As a result, pricing pressure is spreading across the stack even though end-market demand remains uneven.

That said, the cycle is bifurcated. AI- and data-center-oriented products appear likely to remain tight, while PC, mobile, and consumer NAND are more exposed to demand destruction if higher prices cause customers to reduce units or spending. Overall, the market has moved from the broad oversupply of 2023, through recovery in 2024, into a tight and increasingly supply-constrained environment, with the strongest conditions concentrated in HBM and server-related memory.

The memory cycle is no longer uniform. AI-facing memory remains structurally tight, while much of the tightness in consumer-facing DRAM reflects capacity crowding-out rather than exceptionally strong end-demand. HBM is the most constrained segment because rapid AI accelerator growth, high memory content per system, complex packaging requirements, and multi-year supply commitments absorb both wafer capacity and capital expenditure. Server DRAM is also benefiting from AI and cloud infrastructure spending: AI servers require substantially more memory per system, while hyperscalers are increasing memory intensity across both accelerated and general-purpose compute. Together, HBM and high-capacity server DRAM are therefore setting the direction for industry allocation and pricing.

Enterprise demand provides an additional source of support as cloud providers and large data-center customers expand infrastructure, although the pace remains more dependent on deployment schedules and customer budgets than HBM demand. By contrast, PC and smartphone demand is recovering only gradually. Replacement cycles, AI-PC and premium handset features may raise DRAM content per device, but unit growth remains modest and is sensitive to higher component prices. Automotive DRAM is comparatively resilient because memory content continues to increase with ADAS, infotainment, connectivity, and centralized computing architectures. However, automotive is a smaller market with long qualification cycles, so it is more likely to provide stable incremental demand than to drive the overall DRAM cycle.

The implication for commodity DRAM is a tightening supply outlook despite uneven endmarket demand. As suppliers prioritize HBM, high-capacity server DRAM, and other higher-value products, fewer wafers and less incremental investment are available for conventional DDR and LPDDR. This can push commodity DRAM pricing higher even

Salmon 90. Merlory cycle conutons, by type

Memory cycle conditions by type

Supply/demand balance from oversupply to critically tight

Oversupply

HBM

Server DRAM

DDR5 / RDIMM

PC & mobile DRAM

LPDDR / DDR

Enterprise NAND

Data centre SSDs

Consumer NAND

Mobile / PC flash

Critically tight

M

Oversupply

Post-COVID glut. Hyperscaler digestion. DDR5 transition early. Prices

down 40-60%.

Deep oversupply

Prices fell 50%+. Worst PC/smartphone correction in years. Inventory

correction through H1 2023.

Manufacturers reporting heavy losses.

Severe oversupply

Worst correction in NAND's history.

Kioxia/WD in financial distress.

Forward

Structural shortage digit annually.

HBM4 ramping. TAM growing at double without a broad-based boom in PC or smartphone shipments. AI and hyperscale customers are likely to receive priority allocation, while PC, smartphone, automotive, and industrial customers may respond through lower memory configurations, delayed product refreshes, inventory discipline, or margin compression. The outlook is therefore increasingly bifurcated: AI-related products remain structurally undersupplied, while commodity DRAM tightens indirectly through constrained supply, higher product mix, and limited capacity additions rather than through uniformly strong downstream demand.

Recovering stable.

*Great Refresh of 2026" beginning.

VAR checks: 10-25% end-user price

Exhibit 45: Memory cycle conditions, by type

Deep oversupply

Synchronised with enterprise NAND - shared fab infrastructure. Smartphone

OEMs destocking. Consumer electronics demand collapsed.

meaningful in NAND by 2030.

Enterprise TAM expanding - but ClOs cutting storage budgets.

Source: Company commentary, TrendForce, Gartner, VAR checks, Morgan Stanley Research estimates. Note: Qualitative supply/demand conditions shown by product type; price increases refer to cited market checks where available.

Demand is shifting from consumer electronics to AI/server infrastructure

DRAM server share of bit demand rises from 37% in 2023 to 59% in 2028e , while the share of shipments for PCs and smartphones declines sharply. This matters because the dominant buyer increasingly has stronger purchasing power, longer-duration demand visibility and greater willingness to sign L TAs or prepay for supply. As a result, pricing and shipment allocation continue to gravitate toward AI/server infrastructure requirements at the expense of traditional consumer end-markets.

Today (2025-26)

M

Exhibit 46: DRAM - AI server demand becomes dominant…

Source: TrendForce, Morgan Stanley Research. E = TrendForce estimates.

Supply slow to respond

New memory capacity will come online from late 2027 but is still unable to meet demand for all markets. Strategically, most of the valuable capacity is likely to be dedicated to where pricing, margins and customer commitments are strongest - namely to HBM customers, server DRAM and enterprise SSDs. While AI semiconductors are forecast to grow at a ~50% CAGR by 2030 (as per TSMC forecasts at its annual technology symposium on May 14, 2026), we expect DRAM production to grow at a pace that will meet only about 60% of demand, given annual bit shipments are unlikely to rise much above the 30% level due to EUV constraints.

We model DRAM supply remaining tight even as capacity additions accelerate. DRAM undersupply widens from ~3-4% in 2023-24 to ~18% in 2026, before easing but still remaining tight, at ~14%, in 2027e. The key issue is timing.

• Annual DRAM wafer capacity additions rise from ~65kwpm in 2025e to ~498kwpm in 2027e and ~563kwpm in 2028e, led by Samsung, SK hynix, Micron and CXMT. This represents meaningful supply growth by historical standards, but arrives after the period of maximum tightness.
• Of these, CXMT's acceleration represents a meaningful shift in the supply-side landscape, and we expect CXMT's market shares to reach 15% in 2028e vs. 9% in 2025. From 2023 to 2028e, we expect Mainland China to account for ~24% of net global DRAM capacity additions, behind South Korea but ahead of the US. But even with the rapid increase in CXMT's capacity, we still see demand hard to be fulfilled by all the new supply added.

The timing issue in bringing on new capacity is compounded by execution risk. New memory capacity requires fab construction, tool delivery, process qualification, customer validation and yield ramp. Even when capacity is announced, usable output can lag by several quarters or years.

M

Exhibit 47: DRAM supply sufficiency ratio

Source: TrendForce, Morgan Stanley Research. E = Morgan Stanley estimates.

Exhibit 48: Global DRAM capacity is expected to reach 3,374kwpm in 2028e from 2,313kwpm in 2026e

Source: TrendForce, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 49: CXMT's global capacity market share to extend from 9% in 2025 to 15% in 2028e

Source: TrendForce, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Exhibit 50: DRAM annual wafer capacity additions are moving to all-time high

Source: TrendForce estimates, Morgan Stanley Research. E = Morgan Stanley Research estimates.

Exhibit 51: Global DRAM spending

Source: Company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

LTAs could create a two-tier memory market

As top-tier CSPs secure multi-year supply through LTAs, the residual memory market could become structurally tighter. Customers without LTAs may need to procure from a smaller pool of uncommitted supply, potentially at higher and more volatile prices. We view this as a potential structural cost borne by every buyer who is not at the front of the LTA queue.

Historically, weak PC or smartphone demand would help loosen DRAM markets and lower prices for everyone. In this cycle, AI demand keeps allocation tight even if consumer electronics demand softens, because HBM and server applications now set the priority order, and supplier incentives reinforce the allocation. After the last memory downcycle, suppliers have limited incentive to flood the market with commodity supply when AI customers offer stronger pricing, better margins and more durable contracts.

cxmbl oe. uul memuly sumulenluy Hamewur suyyeolo ru anlu omlaltphlunle boulu see 1e loro memory ohlortlall

PC memory sufficiency in 2027

Total DRAM + HBM supply (mn Gb)

573,503

M

Shortfall (mn Gb)

Memory content per device (Gb)

Shortfall (mn units)

vs. MS PC shipment forecast

Downside vs. MSe

Sensitivity analysis

Memory content per device (Gb)

Current assumption

Flat content YoY

De-specing

Smartphone memory sufficiency in 2027

Total DRAM + HBM supply (2027)

Server gets 70% of allocation

Non-Server gets 30% of allocation

Smartphone gets 58% of non-server allocation

Total smartphone demand (2027)

573,503

99,790

113,125

We run DRAM sufficiency analysis on the two largest non-server end-markets: PCs and smartphones Memory content per device (Gb) 99

Starting from total 2027 DRAM + HBM supply of 573,503mn Gb based on our projection, we allocate 70% to servers and 30% to non-server applications. Within non-server, we assign 17% to PCs and 58% to smartphones, consistent with historical patterns. On the PC side, bottom-up demand of 34,386mn Gb implies a shortfall of 5,138mn Gb (-15%). The picture is comparably challenging for smartphones: total smartphone DRAM demand of 113,125mn Gb against an addressable supply of 99,790mn Gb produces a 13,335mn Gb gap (-12%). De-specing 90 -2%

The shortfall proves highly sensitive to content assumptions. On the PC side, our base case of 89Gb per device implies a -15% shortfall; holding content flat year-on-year at 83Gb narrows this to -9%, and reducing spec upgrade assumptions brings it to 77Gb, reducing the shortfall to just -2%. Smartphones follow the same pattern: against a 99Gb base case (-12%), flat content at 95Gb narrows the gap to -7%, and limiting spec refresh brings it to 90Gb, a -2% shortfall.

The read-through is twofold. First, the unit-volume risk is a function of memory content assumptions, not an absolute supply wall - OEMs have a clear lever to protect shipment volumes. Second, that lever is not free: every gigabyte of limiting rising specs that protects unit volumes is a gigabyte of forgone content growth, so the adjustment ultimately resolves the unit shortfall by suppressing memory bit demand. The market clears either through fewer devices or lower memory intensity per device - and the more likely outcome is a blend of the two.

This reinforces supplier pricing power and changes the cycle. The more supply is committed to strategic AI customers, the less flexible supply remains for opportunistic or non-LTA buyers. This could support ASP resilience even if parts of consumer demand weaken.

Exhibit 52: Our memory sufficiency framework suggests PC and smartphone could see 12-15% memory shortfall

Source: Morgan Stanley Research estimates. Note: PC includes laptops, desktops and tablets.

89

83

77

M

Exhibit 53: AI prioritization turns 2027 supply growth into a consumer memory shortfall

Source: Morgan Stanley Research estimates.

allocation

CAnDIt O4.

Prospects for DRAM Cell Beyond 10nm

M

Capacito

Design Rule [nm]

Bit Line

D@©**

CAP

Appendix 2: What is 3D DRAM Technology? Can China CXMT be the First Chip Maker to Adopt it? Bit Line

'20

SAMSUNG

CAIIFAA CAMALIA

'25

'30

Our checks with equipment vendors suggest that CXMT may try VCT (vertical channel transistor), a transitional technology to 3D DRAM, to push its node to Gen5 (~15nm) without EUV tools. Our model assumes some production of this new DRAM technology in 2H28.

Exhibit 54: DRAM scaling through the transistor architecture change

Source: Samsung.

Planar DRAM: The 1c (1 γ ) inflection and the planar wall

DRAM scaling has historically followed a predictable cadence: reducing the physical dimensions of the transistor and capacitor to increase bit density and lower the cost per bit. The 10nm-class progression has been annotated sequentially as 1x, 1y, 1z, 1a, 1b, and now the latest 1c, or 1 γ , node. Samsung, SK hynix, and Micron are now actively massproducing or ramping up 1c DRAM.

The transition from the 14nm 1a node to the 12nm node delivered a significant geometric shrink. In contrast, the transition from 1b to 1c represents a shrink of less than 10%. This deceleration suggests that planar DRAM scaling is becoming less efficient as physical constraints around the transistor, capacitor, leakage, and process complexity intensify.

Planar DRAM: The EUV divergence - parametric yield vs theoretical density

EUV tools are essential for the 1c node. However, as the number of EUV-applied layers increases, the difficulty of subsequent processes, particularly etching, rises sharply. Major memory players have therefore made different strategic choices in balancing density, yield, cost, and process complexity.

• Micron: capital-efficient EUV usage . Micron has maintained a strategy of strict capital efficiency. It applies EUV to fewer critical layers while relying on advanced multi-patterning for others. This may allow Micron to achieve a roughly 30% density improvement over its 1b node without taking on the depreciation burden associated with a larger EUV tool base.

14/20

M

• Samsung: reduced EUV layer count . Samsung originally planned to use 8-9 EUV layers for its 1c node, but recently reduced its EUV layer-count target to 6-7 layers, a roughly 30% reduction. Samsung appears to believe that parametric yield losses associated with complex EUV etching can outweigh the theoretical density benefit. By substituting DUV lithography for certain layers, Samsung also avoids some of the capital intensity associated with additional EUV tools.

Exhibit 55: Comparison of 1B process specs across major vendors

Source: Company data, TechInsights, Morgan Stanley Research.

Next steps: 3D overcoming the planar wall

Following the 1c node, memory makers project 1d or 1e nodes. However, the physical limitations of the capacitor may become an insurmountable barrier in the traditional 6F² planar arrangement. As the lateral footprint shrinks, the capacitor requires a higher depthto-width ratio to maintain sufficient capacitance. This creates severe charge leakage, higher refresh rates, lower power efficiency, and worsening thermal bottlenecks.

The next structural breakthrough is the transition to a 4F² cell structure, also classified as the 0a node. This transition is expected to reach mass-production viability around 202728 and represents the next major step in extending DRAM density beyond the limits of conventional planar scaling.

While 4F² vertical-channel transistor structures can extend the life of single-layer DRAM, the longer-term architectural objective is monolithic 3D DRAM, conceptually analogous to the transition from planar NAND to 3D NAND. The design goal is to stack DRAM cells vertically within a single integrated manufacturing flow, thereby multiplying the number of bits per unit of die area and extending DRAM scaling beyond the limits of planar cell layouts.

1. Vertical-channel approaches . One candidate route is the vertical-channel approach, which extends the role of vertical transistor structures beyond single-layer DRAM. This route is intended to improve density while preserving a recognizable DRAM operating model.

2. Capacitor-inclusive stacked structures . A more difficult route is fully stacked, capacitor-inclusive 3D DRAM. The bottleneck is the integration of a sufficiently large capacitor into each stacked DRAM cell without making the stack too thick, costly, or yieldchallenged.

Cell

Threshold

Voltage

(VIN)

Increasing bits per cell

Higher density

ZAM (Z-Angle Memory)

Z-Angle Memory (ZAM) is co-developed by Intel and SAIMEMORY

(Softbank), ZAM reduces and reorganizes TSVs, routing connections diagonally around a central via-in-one TSV structure rather than

elying on dense arrays of straight vertical TSVs.

Conventional (Straight TSV)

TSV (Straight)

Memory

Layers

ZAM (Staggered Diagonal Interconnect)

Interconnect

Layers

Logic Die

Logic Die

/ 2-3k HBM capacity (barget) / -50% lower power (urget) / Up to 60% lower cont (arge) / Improved thermal detributio

Prototypes targeted 2027; commercialization by 2030; all figures are vendor targets

Panels 3 and 4 describe pre-commercial development programs. Performance figures are vendor targets.

MEMORY DESIGN INNOVATIONS

Advancing capacity, bandwidth, and efficiency through novel cell architectures and 3D integration.

M

→ Capacitor

Fully stacked, capacitor-inclusive 3D DRAM remains a longer-term technology. Early vertical-channel models and proof-of-concept silicon are appearing in R&D, but broad mass production is more likely a next-decade event. Manufacturability will depend on whether the industry can combine stacked channel formation, capacitor integration, highaspect-ratio etching, and atomic-layer deposition at acceptable cost and yield. Transistor interlayer base die, forming a 3D compute-memory stack.

Exhibit 56: Memory design innovation

Up to 4x power efficiency (early tests)

Water-to-wafer bonding, tens of thousands of I/O

No commercialization timeline disclosed

Hybrid copper

Source: Company data.

VS.

M

Appendix 3: Assessing China DRAM Demand/Supply (Local Sufficiency)

Based on our supply chain checks, Alibaba has begun purchasing CXMT server DRAM. ByteDance and Tencent also appear to be procuring from CXMT.

CXMT's LPDDR products are particularly well aligned with domestic mobile substitution because Chinese smartphone brands can qualify the product within local platform and module ecosystems. Smartphone and mobile OEMs include Xiaomi, Transsion, Honor, Oppo, Vivo.

PC and system customers include Lenovo, and other domestic PC and module manufacturers; Dell and HP use CXMT in selecte China and ASEAN models.

CXMT also sells through distributors and module partners. Its top-five customer metric shows concentration of approximately 68% in 2025, which creates account risk even though no individual customer represents more than half of revenue.

What is the realistic addressable market?

The TAM should be separated into four pools.

Pool 1: Chinese domestic consumer and enterprise demand

We view this as the highest-probability TAM, i.e., CXMT's most accessible market:

• Smartphones
• PCs and notebooks
• Other consumer electronics
• Networking and industrial equipment
• Servers
• Government and state-linked infrastructure
• AI accelerators and CPUs

CXMT can address most of this market over time, subject to technical qualification and product availability.

Pool 2: Chinese operations of multinational OEMs

We view China-specific PC models or other consumer devices as the next group of penetrable markets. Customers can compartmentalize regulatory exposure by limiting CXMT content to particular geographies and models.

Pool 3: Non-US or geopolitically neutral export markets

ASEAN, emerging-market and selected regional brands may be accessible, particularly where price and supply availability outweigh US policy risk.

Samor or. appilly ulmas supply aru vernarlu

2020

4,130

2,178

Total China Demand (mn Gb)

PCS

Servers

Other

2021

4,990

2,660

2022

5,581

2,858

2023

6,024

2,826

2024

7,108

3,001

2025

8,720

3,393

2026e

11,008

3,393

M

35%

2027e

14,506

3,563

30,856

13,671

9,934

5,121

35%

2028e

19,494

3,741

46,283

15,038

12,914

6,690

34%

Pool 4: Global premium OEM, hyperscaler and AI markets

This is the least certain pool; barriers include:

• Potential US procurement restrictions
• OEM concern about future supply interruption
• IP and reputational risk
• Qualification timelines
• Lack of leading HBM products

Global shortages could temporarily widen this TAM, but we believe that investors should not capitalize the full global DRAM market as freely addressable.

CXMT's risk-adjusted commercial TAM will remain concentrated in China and selected non-US markets unless geopolitical conditions become more favorable for CXMT.

China domestic demand versus local supply

China is one of the largest consumers and manufacturing centers for memory because it hosts a substantial portion of global smartphone, PC, server, networking and electronics production. Historically, almost all advanced DRAM was imported.

Even with CXMT's rapid expansion, domestic supply remains below China's total effective demand. We forecast CXMT reaching approximately 15% of global bit shipments by 2030. China's share of global electronics manufacturing and semiconductor consumption is larger than that, implying room for domestic substitution even at that scale. According to our calculation, we think the sufficiency ratio reaches only ~35% in 2027e-2028e, a level at which we see no risk of oversupply.

However, we also note that 'China demand' is not fully captive. Multinational factories in China may be required by global product specifications to use Samsung, SK hynix or Micron.

Exhibit 57: Mapping China's supply and demand

Source: TrendForce, company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

Exhibit 58: We believe China's domestic supply is insufficient to support the strong demand

Source: TrendForce, company data, Morgan Stanley Research. E = Morgan Stanley Research estimates.

How much China demand could ultimately be served domestically?

A useful segmentation is:

High substitution potential: 50-80% over time -

• Mainstream domestic PCs
• Domestic handset models
• Consumer electronics
• Modules and white-box systems
• Government and state-linked procurement
• Domestic cloud servers
• Industrial products with manageable qualification requirements

Moderate substitution potential: 25-50% -

• Premium smartphones
• High-density enterprise servers
• Automotive and telecom infrastructure
• China-based models of multinational OEMs
• Selected AI accelerators using domestic HBM or high-capacity DDR

Low near-term substitution potential: below 20-30% -

• Frontier HBM
• Globally standardized premium server platforms
• High-end automotive applications requiring long histories
• Products requiring leading global CPU/GPU vendor co-qualification
• Multinational products exposed to US government procurement or compliance restrictions

Exmol 0r. umas evolution or capacily, lechology anu talkel positioninly

Period

Estimated CXMT

capacity

Technology and products

Primary market position

M

2021-2022

2023

2024

2025 onward

~ 40-70kwpm

~ 120kwpm

~ 160-200kwpm

~ 250-300kwpm refinement

development

Aggregating these categories, CXMT and other domestic suppliers could plausibly serve a significant share of China's conventional DRAM demand. Further domestic market share upside, including leading HBM and premium international products, would require further progress in technology, qualification and packaging.

Mainstream domestic

China's DRAM industry evolution

China's DRAM industry has progressed from effectively no commercially significant domestic production a decade ago to a meaningful position in mainstream memory, led by ChangXin Memory Technologies (CXMT). Founded in 2016, CXMT moved into volume production of domestically manufactured 8Gb DDR4 in 2019 and subsequently expanded into LPDDR4X, DDR5 and LPDDR5/5X. Its current portfolio addresses PCs, smartphones, tablets and servers, with published specifications increasingly comparable to mainstream international products.

The most important development has been scale. We estimate CXMT's monthly wafer capacity rose from approximately 40k wafers in 2020 to 70k in 2022, 120k in 2023, 150k by the end of 2024, and to 180k by the end of 2025. With CXMT's capacity expansion, China has moved from negligible DRAM supply to a position capable of materially affecting global legacy-memory pricing. We believe China's strongest competitive position is currently in mainstream DDR5 and LPDDR5 products sold into the domestic PC, handset, consumer-electronics and industrial ecosystems.

HBM development

In the previous section, we mentioned CXMT's progress in the HBM3E 12-Hi product. Other new entrants are also pursuing HBM development. For example, XMC (Wuhan Xinxin) is reportedly building HBM-related advanced-packaging capacity using 3D multiwafer stacking, with planned capacity of at least 3,000 12-inch wafers per month. Current activity appears to remain at the R&D and pilot-production stage, but capacity could increase over the coming years. Our supply-chain checks suggest that JHICC is also targeting longer-term participation in the HBM market. The company could expand its HBM capacity over the long term, initially starting with JBM2e

Exhibit 59: China's evolution of capacity, technology and market positioning

Source: Company data, Morgan Stanley Research.

M

How has China's DRAM industry progressed over the past decade?

2015-2018: Policy-driven industry formation

China entered the previous decade as the world's largest consumer of semiconductors, but it remained heavily dependent on imported DRAM. Multiple domestic memory initiatives were launched as part of China's broader semiconductor self-sufficiency strategy, including CXMT, Fujian Jinhua and other smaller projects.

CXMT was established in Hefei in 2016. Unlike several Chinese semiconductor ventures that attempted to develop a technology platform from scratch, CXMT built its initial DRAM architecture partly around intellectual property originating from Germany's Qimonda. This provided a more defensible starting point and reduced some early intellectual-property risk relative to alternative technology-transfer routes.

The company also benefited from local-government financing, inexpensive land, infrastructure support and access to engineers recruited from established Asian memory manufacturers. Industry assessments have identified talent recruitment from Samsung and SK hynix operations in China as an important contributor to the rapid development of China's memory capabilities.

2019-2021: Commercial validation at 19nm-class DDR4

CXMT announced volume production of an 8Gb DDR4 device in 2019, generally associated with its 10G1, or approximately 19nm-class, manufacturing process. This represented China's first meaningful domestically manufactured DRAM product at commercial scale. By 2020, estimated capacity had reached approximately 40,000 300mm wafers per month, with the company producing DDR4 and subsequently adding LPDDR4 products.

At this stage, CXMT's positioning was primarily consumer and entry-level PC DRAM, lower-end to mid-range smartphone memory, domestic module manufacturers, and industrial and embedded applications with less demanding qualification requirements.

CXMT's technology remained several generations behind the leading global producers, but the products were sufficiently competitive for portions of the Chinese domestic market. The key achievement was therefore not technology leadership but the establishment of repeatable manufacturing, yield improvement and a domestic customer base.

2022-2024: Rapid capacity expansion and pressure on legacy DRAM pricing

From 2022, CXMT accelerated its capacity build-out. We estimate its monthly wafer capacity expanded from roughly 70k wafers in 2022 to around 120k in 2023.

During the earlier phase of this expansion, incremental output was concentrated largely in DDR4 and LPDDR4X. This capacity growth came amid weak global consumer-electronics demand and industry-wide inventory corrections, adding supply to legacy DRAM segments that were already experiencing significant pricing pressure.

Industry research, including TrendForce, linked CXMT's expansion in older-generation DRAM to weaker pricing in DDR4 and LPDDR4X. The effect remained visible into early 2025, when continued softness in DDR4 spot prices was partly attributed to abundant supply from CXMT. By late 2024, however, CXMT had begun shifting its product mix

M

toward DDR5, marking the start of a transition away from legacy DDR4 production.

2024-2025: Transition into mainstream DDR5 and LPDDR5X

CXMT began transitioning its mainstream DRAM portfolio toward newer-generation products in late 2024, when DDR5 entered mass production, followed by LPDDR5X in 2025. CXMT's 8,533Mbps and 9,600Mbps LPDDR5X products entered mass production in May 2025, while its 10,667Mbps variant was in customer sampling by October 2025.

In November 2025, CXMT publicly showcased a broader DDR5 portfolio with speeds of up to 8,000Mbps and die densities of up to 24Gb, together with multiple module formats including server-oriented RDIMM and MRDIMM products. Its LPDDR5X portfolio included 12Gb and 16Gb dies and extended to a published maximum data rate of 10,667Mbps.

Where CXMT fits within the broader China DRAM story

CXMT is China's most advanced DRAM producer and the clearest vehicle for China's push toward memory self-sufficiency. There are other Chinese players developing DRAM and HBM, but CXMT combines proven high-volume manufacturing, established relationships with domestic cloud, PC, and smartphone customers, and a comparatively mature process and R&D organization.

China's DRAM industry has moved through three stages: establishing volume production with CXMT's ~19nm-class 8Gb DDR4 in 2019, scaling DDR4 and LPDDR4X, and entering the mainstream market with DDR5 and LPDDR5X for PCs, mobile devices, and select server applications.

The product gap with global leaders has narrowed, but CXMT still trails in cost per bit, process maturity, qualification breadth, manufacturing scale, and advanced AI-memory integration.

Exhibit 60: CXMT's global capacity market share to extend from 9% in 2025 to 15% in 2028e

Source: TrendForce, Morgan Stanley Research. E = Morgan Stanley Research estimates.

M

This report references U.S. Executive Order 14032 and/or entities or securities that are designated thereunder. U.S. persons may be prohibited from buying certain securities of entities named in this report. Readers are solely responsible for ensuring that their investment activities are carried out in compliance with applicable laws.

This report references export controls and/or entities that may be subject to export control restrictions. Readers are solely responsible for ensuring that their investment or trade activities are carried out in compliance with applicable laws.

This report references U.S. Executive Order 14105 and/or entities that may be in scope of such order. U.S. persons may be prohibited from engaging in certain transactions or otherwise require certain other transactions be notified to the U.S. Department of Treasury. Readers are solely responsible for ensuring that their investment or trade activities are carried out in compliance with applicable laws.

M

Risk Reward Reference links

1. View explanation of Options Probabilities methodology -

Options_Probabilities_Exhibit_Link.pdf

1. View descriptions of Risk Rewards Themes - RR_Themes_Exhibit_Link.pdf

2. View explanation of regional hierarchies - GEG_Exhibit_Link.pdf

3. View explanation of Theme/Exposure methodology -

4. ESG_Sustainable_Solutions_External_Link.pdf

5. View explanation of HERS methodology - ESG_HERS_External_Link.pdf

M

Disclosure Section

The information and opinions in Morgan Stanley Research were prepared or are disseminated by Morgan Stanley Asia Limited (which accepts the responsibility for its contents) and/or Morgan Stanley Asia (Singapore) Pte. (Registration number 199206298Z) and/or Morgan Stanley Asia (Singapore) Securities Pte Ltd (Registration number 200008434H), regulated by the Monetary Authority of Singapore (which accepts legal responsibility for its contents and should be contacted with respect to any matters arising from, or in connection with, Morgan Stanley Research), and/or Morgan Stanley Taiwan Limited and/or Morgan Stanley & Co International plc, Seoul Branch, and/or Morgan Stanley Australia Limited (A.B.N. 67 003 734 576, holder of Australian financial services license No. 233742, which accepts responsibility for its contents), and/or Morgan Stanley Wealth Management Australia Pty Ltd (A.B.N. 19 009 145 555, holder of Australian financial services license No. 240813, which accepts responsibility for its contents), and/or Morgan Stanley India Company Private Limited having Corporate Identification No (CIN) U22990MH1998PTC115305, regulated by the Securities and Exchange Board of India ('SEBI') and holder of licenses as a Research Analyst (SEBI Registration No. INH000001105); Stock Broker (SEBI Stock Broker Registration No. INZ000244438), Merchant Banker (SEBI Registration No. INM000011203), and depository participant with National Securities Depository Limited (SEBI Registration No. IN-DP-NSDL-567-2021) having registered office at Altimus, Level 39 & 40, Pandurang Budhkar Marg, Worli, Mumbai 400018, India; Telephone no. +91-22-61181000; Compliance Officer Details: Mr. Tejarshi Hardas, Tel. No.: +91-22-61181000 or Email: tejarshi.hardas@morganstanley.com; Grievance officer details: Mr. Tejarshi Hardas, Tel. No.: +91-22-61181000 or Email: msic-compliance@morganstanley.com which accepts the responsibility for its contents and should be contacted with respect to any matters arising from, or in connection with, Morgan Stanley Research, and their affiliates (collectively, "Morgan Stanley"). Morgan Stanley India Company Private Limited (MSICPL) may use AI tools in providing research services. All recommendations contained herein are made by the duly qualified research analysts.

For important disclosures, stock price charts and equity rating histories regarding companies that are the subject of this report, please see the Morgan Stanley Research Disclosure Website at www.morganstanley.com/eqr/disclosures/webapp/generalresearch, or contact your investment representative or Morgan Stanley Research at 1585 Broadway, (Attention: Research Management), New York, NY, 10036 USA.

For valuation methodology and risks associated with any recommendation, rating or price target referenced in this research report, please contact the Client Support Team as follows: US/Canada +1 800 303-2495; Hong Kong +852 2848-5999; Latin America +1 718 754-5444 (U.S.); London +44 (0)20-7425-8169; Singapore +65 6834-6860; Sydney +61 (0)2-9770-1505; Tokyo +81 (0)3-6836-9000. Alternatively you may contact your investment representative or Morgan Stanley Research at 1585 Broadway, (Attention: Research Management), New York, NY 10036 USA.