CXMT is entering the mainstream DRAM market, with Chinese AI demand, localization, and capacity expansion driving the Overweight thesis

Overview

The report examines whether CXMT can grow from a domestic Chinese DRAM supplier into a mainstream vendor and initiates coverage with an Overweight rating. The core view is that demand from Chinese AI infrastructure, memory localization, and continued capacity build-out will drive rapid share gains and earnings growth; meanwhile, restrictions on advanced lithography equipment, the difficulty of HBM mass production, and industry supply pressure after 2028 will determine the long-term outcome.

Core views

CXMT has already achieved meaningful scale and is transitioning from a domestic substitution supplier into the mainstream DRAM market. Morgan Stanley estimates that CXMT will have 300kwpm of capacity in 2026, equivalent to 13% of global DRAM wafer supply; bit shipments will be approximately 41bn Gb, representing 11% of global bit supply, with this share expected to rise to 15% by 2030. The report expects 2026 revenue of approximately US$57bn, corresponding to Rmb388.869bn in the financial model. Although its process technology remains about two generations behind Samsung and Micron, LPDDR5 can already meet the needs of Chinese smartphones, and CXMT is an important domestic supplier for Chinese networking equipment, AI GPU servers, and other infrastructure. Product competitiveness and customer adoption are improving. Supply chain research indicates that CXMT's DDR5 already supports 5600MHz and is planned to increase to 6400MHz, bringing it closer to mainstream server DRAM requirements; Gen4b has migrated to a 16nm-class process, primarily using domestic equipment apart from DUV lithography, and the report estimates mass-production yields above 80%. Alibaba, ByteDance, and Tencent may already have procured its server DRAM, while adoption of LPDDR5 by Chinese smartphone brands and certain low-end AI accelerator platforms is also expanding; Dell and HP reportedly use CXMT products in certain PC models in China and ASEAN markets. These signs indicate that product quality has reached commercially acceptable levels, although certification by high-end customers and for advanced products remains another hurdle. Apple reportedly tested CXMT memory for devices sold in China, but potential supply still depends on geopolitical factors and US approval conditions. Scale expansion is the main growth driver and also the stage where equipment constraints are most concentrated. The report expects capacity to rise from 180kwpm in 2025 to 300kwpm in 2026, then increase by approximately 100kwpm annually from the second half of 2026, reaching approximately 500kwpm in 2028, or 15% of that year's global capacity of 3374kwpm, and further reaching 800kwpm in 2031. The capacity expansion assumptions depend on DRAM industry margins remaining at approximately 80% and CXMT continuing to obtain overseas DUV tools. ASML's DUV lithography systems are viewed as the main bottleneck; China has been unable to obtain EUV since 2023, and ASML has never delivered EUV to China. If the proposed MATCH Act designates CXMT and others as restricted facilities, it could also restrict sales of immersion DUV equipment and maintenance services for the installed base, thereby affecting yields and capacity ramp-up. Meanwhile, domestic etching, deposition, cleaning, thermal processing, and inspection equipment is gradually replacing US suppliers. The central debate over the technology roadmap is whether CXMT can mass-produce HBM3E on schedule and use new architectures to circumvent EUV constraints. The report believes HBM3E could become the main bottleneck for Chinese AI GPU shipments in 2027; even with low yields, CXMT could still produce 3mn to 4mn HBM3E stacks in 2027, enough to support approximately 0.8mn to 1.0mn Chinese AI GPUs. The company already has 3D stacking capability for 12-layer HBM3E, but thermal management and manufacturing challenges remain, and Morgan Stanley expects mass production in the first half of 2027. Over the longer term, CXMT may adopt vertical channel transistor technology as a transitional architecture toward 3D DRAM, advancing to an approximately 15nm-class Gen5 process without relying on EUV and beginning limited production in the second half of 2028. If this roadmap succeeds, it could narrow the technology gap with international vendors; if development or yield ramp-up is delayed, both the growth and valuation theses would be impaired. Regarding global supply and demand, the report remains positive on DRAM demand and pricing in 2027 and believes CXMT will not be large enough to cause a material pricing impact before 2028. AI-related demand is expected to grow faster than DRAM supply, while HBM production consumes more wafers, EUV equipment supply is limited, and fab construction and yield ramp-up take time; market tightness could therefore continue through 2027. The report expects DRAM prices to rise by more than 20% to 30% in the third quarter of 2026, although year-over-year growth may peak in the fourth quarter, and near-term cyclical catalysts may weaken as pricing and earnings momentum slow. From 2028 onward, if the industry returns to stronger cyclicality, continuously expanding Chinese DRAM capacity will become a more important variable in global supply, demand, and pricing. Even amid moderate price declines, CXMT may continue expanding capacity because of strategic localization requirements rather than reducing investment entirely in line with the traditional cycle. The company's financial forecasts reflect the combined effects of capacity expansion, pricing, and product upgrades. The model expects net sales to rise from Rmb61.799bn in 2025 to Rmb388.869bn in 2026, Rmb643.077bn in 2027, and Rmb850.096bn in 2028, corresponding to a revenue CAGR of approximately 140% from 2025 to 2028; the operating margin is expected to increase from 14.2% in 2025 to 70.3% in 2026, 71.6% in 2027, and 71.8% in 2028. Net profit attributable to the parent is expected to rise from Rmb1.875bn in 2025 to Rmb189.693bn in 2026, Rmb319.243bn in 2027, and Rmb423.535bn in 2028; earnings per share are expected to be Rmb0.03, Rmb3.00, Rmb4.77, and Rmb6.33, respectively. Operating cash flow is expected to increase from Rmb36.520bn in 2025 to Rmb599.129bn in 2028. On valuation, Morgan Stanley derives a target price of Rmb88 using a residual income model, implying 56% upside from the reference share price of Rmb56.50. The target price implies 18.5x 2027E P/E, compared with 4.4x for global peers; the report believes CXMT merits a premium due to Chinese AI server demand, localization drivers, and faster capacity growth. The risk-reward framework also presents bull- and bear-case 2027E P/E multiples of 27.7x and 14.8x, respectively. The bull case assumes severe global shortages prompt US cloud service providers and consumer electronics brands to procure CXMT products, while the company achieves a breakthrough in 3D DRAM around 2028; the bear case assumes further tightening of semiconductor manufacturing equipment export restrictions, particularly those involving ASML DUV tools. The supply chain impact is not one-directional. Samsung Electronics and SK Hynix are still viewed as beneficiaries of stronger AI computing demand and memory pricing, but they will need to contend with incremental Chinese supply after 2028; Micron's long-term agreements, free cash flow, and lower valuation continue to support its investment thesis. CXMT's capacity expansion and technology migration will increase the intensity of etching, deposition, cleaning, inspection, and back-end equipment, prompting the report to reiterate its Overweight views on NAURA Technology, AMEC, and ACMR. GigaDevice can also benefit through specialty memory collaboration with CXMT, DDR4 capacity expansion, and wafer stacking technology cooperation. The extent to which Japanese equipment suppliers benefit depends on their exposure to CXMT, the progress of domestic substitution, and HBM process complexity; HBM grinding and testing companies, as well as server memory interface vendors, could also see incremental demand.

Analysis framework

The report first uses capacity, bit shipment share, process nodes, yields, and customer adoption to assess whether CXMT has entered the mainstream DRAM market, and then combines supply chain research to evaluate the feasibility of DUV equipment access, domestic equipment substitution, HBM3E, and the 3D DRAM roadmap. It subsequently assesses pricing impacts based on global and Chinese DRAM supply and demand, AI demand, and the timing of capacity ramp-up, and uses these findings to develop revenue, profit, and cash flow forecasts for 2025 to 2028. Finally, it derives the target price and supply chain mapping through a residual income model, peer P/E comparisons, and bull/bear scenarios.

Methodology notes

• Valuation MethodRIM Residual Income Model

  Residual Income Model

  The model estimates equity value based on the company's book capital and expected returns above the cost of capital; the report uses it to derive CXMT's Rmb88 target price.

• Valuation MethodPE/PEG valuation

  Comparison of 2027E P/E with global peers

  The report compares the target price's implied 18.5x 2027E P/E with 4.4x for global peers and explains the valuation premium through CXMT's faster growth.

• Industry/Sector Analysis FrameworkSupply-demand framework

  Global and Chinese DRAM supply-demand balance analysis

  The report compares AI-driven bit demand, HBM wafer consumption, equipment constraints, and the pace of new capacity ramp-up to determine that pricing will remain supported in 2027 and that CXMT's impact on global pricing may increase after 2028.

• Industry/Sector Analysis FrameworkUpstream, Midstream, and Downstream Supply Chain Transmission

  Transmission of capacity expansion and technology upgrades to equipment, materials, packaging and testing, and interface chips

  As CXMT expands wafer capacity and migrates to more complex processes, demand for etching, deposition, cleaning, inspection, grinding, and testing will increase; the report identifies domestic and overseas supply chain beneficiaries accordingly.

• (Method Outside Vocabulary)

  Bull, base, and bear case scenario analysis

  The report sets conditions including overseas customer adoption and a 3D DRAM breakthrough, the base-case capacity expansion path, and tighter DUV restrictions to illustrate the sensitivity of valuation and operating results to key assumptions.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• CXMT Corp (688825.SH)

  A key beneficiary of Chinese DRAM localization, AI server demand, and continued capacity expansion

  Strengths

  China's largest DRAM fab; possesses proprietary process nodes; expected to account for 11% of global bit shipments in 2026; expanding customer adoption of LPDDR5 and DDR5

  Weaknesses

  Process technology is approximately two generations behind Samsung and Micron, while advanced customer certification, HBM yields, thermal management, and manufacturing remain challenging

  Comparison

  The target price implies 18.5x 2027E P/E, above the 4.4x of global peers, with the report attributing the premium to higher growth

  Risks

  Earlier-than-expected DRAM price declines, delays in advanced products, tighter equipment and material restrictions, and weaker demand from major Chinese customers

• Samsung Electronics (005930.KS) and SK Hynix (000660.KS)

  Continue to benefit from AI computing demand and rising DRAM prices, but may face more pronounced competition from Chinese supply after 2028

  Strengths

  Major beneficiaries of AI computing and agentic AI trends, with strong DRAM pricing driving earnings upgrades

  Weaknesses

  Memory stocks are highly sensitive to changes in pricing and earnings growth

  Comparison

  The two companies trade at approximately 3.5x to 4.5x 2027E earnings

  Risks

  Year-over-year price growth may peak in the fourth quarter of 2026, while supply discipline after 2028 remains unclear

• Micron (MU.O)

  Benefits while AI demand continues to outpace DRAM supply, but CXMT will gain share through faster capacity expansion

  Strengths

  Long-term agreements improve earnings visibility, free cash flow and capital return potential are high, and the starting valuation is relatively low

  Weaknesses

  The report does not assume memory shortages will continue indefinitely

  Comparison

  Approximately 6x peak-run-rate earnings per share and 2.7x FY28E book value

  Risks

  Weaker AI demand or a faster-than-expected normalization of supply and demand

• ASML (ASML.AX)

  Global memory capacity expansion depends on its EUV supply, while DUV tools are the primary equipment bottleneck for CXMT's capacity expansion

  Strengths

  Low-NA EUV shipments are expected to rise from 44 units in the prior year to more than 60 units in the current year

  Weaknesses

  The effectiveness of capacity expansion also depends on customer capital commitments, fab readiness, process certification, yields, and HBM packaging and testing capabilities

  Comparison

  The constraints are not caused solely by delivery lead times but are jointly determined by the entire capacity expansion chain

  Risks

  Changes in export policy, order volatility, and customers delaying capital expenditures

• NAURA Technology (002371.SZ)

  CXMT's capacity expansion, process upgrades, and domestic equipment certification increase demand for etching, PVD, and wet cleaning equipment

  Strengths

  A broad equipment portfolio covering multiple front-end and back-end stages, with opportunities for cross-selling and increasing supply share

  Weaknesses

  Incremental opportunities depend on production validation and adoption in advanced process steps

  Comparison

  The report views it as one of the broadest beneficiaries of CXMT's capacity expansion

  Risks

  Equipment certification or CXMT's capacity expansion progressing below expectations

• AMEC (688012.SH)

  Higher-aspect-ratio capacitor hole structures in advanced DRAM increase demand for dielectric etching

  Strengths

  As a leading Chinese etching equipment supplier, it is positioned to benefit from localization and greater chamber intensity

  Weaknesses

  More complex capacitor etching imposes higher requirements for profile control, selectivity, and uniformity

  Comparison

  It has a clear process-driven benefit in critical capacitor hole etching

  Risks

  Failure to validate demanding processes or repeat orders falling below expectations

• ACMR (ACMR.O)

  Front-end and back-end cleaning intensity, interconnect complexity, and copper electroplating demand rise as CXMT upgrades its processes

  Strengths

  Its single-wafer cleaning and ECP platforms can address particle, residue, and metal contamination control, as well as certain copper electroplating applications

  Weaknesses

  Higher equipment value per wafer depends on validation in more critical process steps

  Comparison

  It covers opportunities in both front-end cleaning and back-end wet cleaning and electroplating

  Risks

  Critical-step certification or CXMT's technology migration proceeding more slowly than expected

• GigaDevice (603986.SH)

  As CXMT's specialty memory partner, it benefits from DDR4 capacity expansion, process migration, and wafer stacking collaboration

  Strengths

  The 2026 related-party transaction budget with CXMT is Rmb5.711bn, significantly above Rmb1.161bn in 2025; the two parties are also jointly developing wafer-to-wafer technology

  Weaknesses

  Growth is closely tied to CXMT's capacity build-out and execution of the partnership

  Comparison

  CXMT may add another 10kwpm of DDR4 capacity before 2027, above the current approximately 8kwpm

  Risks

  The partnership budget, DDR4 capacity expansion, or technology development falling below expectations

• Nanya Technology (2408.TW) and Winbond Electronics (2344.TW)

  Benefit from strong memory industry profitability and a rerating of global memory stocks

  Strengths

  Industry pricing and earnings conditions remain strong

  Weaknesses

  Sensitive to the memory cycle and pricing momentum

  Comparison

  The report maintains Overweight ratings and target prices of NT$580 and NT$288, respectively

  Risks

  Increased global supply or weakening DRAM pricing momentum

Key data

• Rating and Target PriceOverweight；Rmb88.00Initiation of coverage; 56% upside from the reference share price of Rmb56.50
• 2026 Capacity300kwpmApproximately 13% of global DRAM wafer supply, up from 180kwpm in 2025
• 2026 Bit Shipments41bn GbApproximately 11% of global DRAM bit supply
• 2030 Global Bit Shipment Share15%Up 4 percentage points from 11% in 2026
• Planned Capacity in 2031800kwpmExpected to add approximately 100kwpm annually from the second half of 2026
• DDR5 Performance5600MHz, roadmap to 6400MHzExpands its applicability in mainstream server DRAM
• Gen4b YieldAbove 80%16nm-class process, primarily using domestic equipment apart from DUV lithography
• 2027 HBM3E Output3-4mn stacksDespite low initial yields, expected to support approximately 0.8-1.0mn Chinese AI GPUs
• 2025-2028 Revenue CAGR140%Net sales are expected to rise from Rmb61,799mn to Rmb850,096mn
• 2026-2028 Operating Margin70-72%The model specifically forecasts 70.3%, 71.6%, and 71.8%
• 2026-2028 Earnings per ShareRmb3.00, Rmb4.77, Rmb6.33Corresponding to 2026E, 2027E, and 2028E
• Target Price Implied Valuation18.5x 2027e P/EGlobal peers are at 4.4x; the bull and bear cases show 27.7x and 14.8x, respectively
• Third-Quarter 2026 DRAM PricesIncrease of more than 20-30%Year-over-year growth is expected to continue accelerating but may plateau in the fourth quarter of 2026

Impact & implications

The report believes CXMT's capacity expansion is more likely to fill China's domestic DRAM supply gap in the near term than disrupt global pricing before 2028; Chinese AI server and localization demand therefore constitute its main growth pillars. As advanced DRAM structures increase the intensity of etching, deposition, cleaning, inspection, and back-end processes, domestic equipment suppliers and related memory partners will gain incremental opportunities. Over the medium to long term, if CXMT achieves breakthroughs in HBM3E, VCT, and 3D DRAM and secures certification from overseas customers, its global competitive impact will increase significantly; conversely, restrictions on DUV supply and maintenance would slow capacity and technology migration.

Risks

• Global DRAM prices may decline earlier than the report expects, weakening CXMT's earnings and valuation.
• Development and mass production of CXMT's new products, HBM, or 3D DRAM may be delayed.
• Restrictions on equipment, materials, or maintenance services for Chinese fabs may tighten further, particularly those involving ASML DUV tools.
• Demand from major Chinese customers may be weaker than expected.
• Low initial HBM3E yields and thermal management and manufacturing challenges may constrain supply capacity in 2027.

What to watch

• Watch whether CXMT can mass-produce 12-layer HBM3E as planned in the first half of 2027, as well as actual yields and shipment volumes.
• Track the availability of ASML DUV equipment and maintenance services for the installed base, as well as progress on the proposed MATCH Act.
• Monitor the DDR5 upgrade from 5600MHz to 6400MHz and customer certification across servers, smartphones, and AI accelerators.
• Watch whether Apple, US cloud service providers, and global consumer electronics brands begin procuring CXMT DRAM, as well as the related approval conditions.
• Track whether CXMT can increase capacity by approximately 100kwpm annually and reach approximately 500kwpm in 2028 and 800kwpm in 2031.
• Monitor whether year-over-year DRAM price growth peaks in the fourth quarter of 2026 and whether incremental Chinese supply affects global pricing from 2028 onward.
• Watch whether VCT and the approximately 15nm-class Gen5 process can begin limited production in the second half of 2028 and drive a breakthrough in 3D DRAM.