Chinese memory manufacturers are entering global competition through scale rather than technological leadership, with 2028 potentially becoming an inflection point for industry pricing and returns

Overview

The report examines how Chinese memory manufacturers are reshaping global DRAM and NAND competition through policy support, domestic demand, manufacturing scale, and equipment localization. Its central conclusion is that China has not yet matched global leaders in technology, cost, customer qualification, or high-end product capabilities, but it does not need comprehensive technological leadership. As long as incremental supply is sufficient to affect marginal pricing, it can change industry economics. This impact will be absorbed by AI demand in the near term but could strengthen significantly beginning in 2028.

Core views

The report first considers China's rise within the historical shifts in memory industry leadership: DRAM was pioneered and dominated by the United States in the 1970s; Japan overtook it in the 1980s through high-quality, high-yield manufacturing; and South Korea took the lead in the 1990s through aggressive investment, advantages in cost per bit, and rapid process migration, maintaining its dominance since 1998. China is following a different path—CXMT and YMTC are first scaling up mainstream memory by relying on the vast domestic market, policy support, capital, and import-substitution demand, and then gradually advancing their processes and product tiers. The report emphasizes that China does not need to displace Samsung, SK hynix, and Micron, nor does it first need to achieve widespread adoption in the United States or Europe. As long as domestic output is sufficiently large, it can replace imports, force overseas supply into other markets, and influence the marginal pricing of mainstream memory. In DRAM, CXMT is transitioning from a technology follower into a scale supplier. The report expects its monthly capacity to increase from 180,000 wafers in 2025 to 300,000 in 2026, equivalent to 13% of global DRAM wafer capacity and 11% of bit shipments. Capacity is then expected to reach 500,000 wafers per month in 2028 and 800,000 in 2031, with approximately 15% of global bit shipment share by 2030. Global DRAM wafer capacity is expected to reach 3.374 million wafers per month in 2028, of which CXMT could account for approximately 15%, potentially surpassing Micron, currently the third-largest manufacturer. However, comparable wafer scale does not imply comparable bit output, cost, or product capabilities. CXMT has migrated to Gen 4B, an approximately 16nm-class process, and the report estimates that yields exceed 90%. However, its reliance on immersion DUV and multiple patterning requires more masks and process steps, resulting in longer production cycles and structural disadvantages in overlay, defects, and cost per bit. Even if visible DDR5 specifications are similar, leading manufacturers retain advantages in bit output per wafer, yields, the proportion of high-grade products, and the breadth of customer qualifications. The report argues that fully catching up technologically is not a prerequisite for influencing pricing. Strategic influence in commodity memory depends on whether incremental output is sufficient to change marginal prices. The traditional Big Three are shifting more wafers, R&D, and packaging resources toward HBM and advanced server memory, while CXMT is expanding conventional DRAM to meet domestic demand, improve yields, and build scale. Increased Chinese supply has already prompted overseas manufacturers to gradually exit lower-end markets such as DDR4 and traditional MLC NAND. China could therefore establish a price floor for older-generation products while pushing global manufacturers to transition more rapidly toward high-end AI memory. AI creates what the report calls a “paradox”: it both generates a high-margin HBM profit pool for traditional leaders and opens an entry point into mainstream memory for Chinese manufacturers. AI data centers require not only HBM paired with GPUs and ASICs but also large amounts of commodity DRAM. HBM uses larger dies and requires stacking, TSV, advanced packaging, testing, and customer qualification. Its bit-output penalty is expected to rise from approximately 2.5 times that of conventional DRAM wafers during 2021–2023 to approximately 3 times by 2028. HBM's share of advanced DRAM wafer consumption is expected to rise from approximately 6% in 2023 to 34% in 2028. This will draw traditional manufacturers' capital and advanced capacity toward high-end AI memory, creating room for CXMT to expand mainstream DDR5 and LPDDR5X. At the same time, AI demand will absorb incremental output in the near term, delaying the broad impact of Chinese supply on global prices. China's own HBM demand is also becoming urgent. The report expects HBM3E could become the primary bottleneck for Chinese AI GPUs in 2027 as global manufacturers shift more capacity toward HBM4. Research suggests that even at lower yields, CXMT could produce approximately 3 million to 4 million HBM3E stacks in 2027, enough to support approximately 800,000 to 1 million domestically produced Chinese AI GPUs. Its HBM bit shipment share is expected to reach 3.8% in 2026 and 4.4% in 2027. CXMT has also developed 3D stacking capabilities for 12-layer HBM3E, which the report expects to enter mass production in the first half of 2027, followed by HBM4 development in 2028. However, yields, packaging, testing, and accelerator qualification still prevent it from becoming a global substitute for the traditional Big Three. Even after incorporating CXMT's planned HBM ramp-up, Morgan Stanley's DRAM model still indicates market shortages of approximately 17% and 15% in 2026 and 2027, respectively. Accordingly, the report divides the timeline into two phases. Before 2028, AI demand, HBM's crowding out of advanced capacity, domestic import substitution, and customer qualification cycles are expected to absorb most of China's incremental supply. CXMT is expected to remain relatively moderate amid shortages, mainly serving mainstream DDR, LPDDR, and China-specific applications that traditional manufacturers no longer prioritize. Beginning in 2028, unprecedented capacity expansion by CXMT, YMTC, and traditional manufacturers could simultaneously reach sufficient scale to influence global prices and change how the industry responds during downturns. The key issue is not simply “cheap Chinese DRAM,” but rather the market's transformation from an oligopoly of three companies with broadly aligned profit objectives into a four-supplier structure with asymmetric strategic objectives. The traditional Big Three are profit-driven and cut capital expenditure, utilization, and bit growth when demand weakens. CXMT also pursues import substitution, technological learning, supply-chain resilience, and semiconductor self-sufficiency. If it continues expanding capacity during a downturn, the traditional Big Three may have to bear a greater share of production cuts. In NAND, YMTC's breakthrough stems more from architecture. Its Xtacking technology uses hybrid bonding; its 267-layer product is already in mass production, while products with more than 300 layers are under development. Its PCIe 5.0 enterprise TLC SSD achieves sequential read bandwidth of 14.0–14.2GB/s and random read performance of 3.1 million–3.3 million IOPS. Its QLC-based PE501 offers capacity of up to 122.88TB, a rated 5,000 program/erase cycles, and 0.6 drive writes per day, demonstrating progress in density, endurance, and system integration. However, product launches do not mean it has secured broad qualification from major cloud providers, nor do they indicate that its enterprise product portfolio has reached the level of global leaders. YMTC is the largest swing factor in NAND supply and demand after 2028. Fab 4 and Fab 5 are both under construction, with planned capacity of 100,000 wafers per month each. If all five announced fabs are used for NAND, its eventual global share could reach 24%. The report's scenario analysis assumes that non-AI NAND demand grows 5% year over year in 2028, AI SSD demand grows 30%–60% year over year, and YMTC's monthly capacity ranges from 310,000 wafers in the current base case to a potential upper limit of 470,000 wafers across five fabs. If AI capital expenditure continues growing and YMTC maintains relatively disciplined capacity expansion, NAND supply could remain tight through 2028. If YMTC or other manufacturers accelerate greenfield expansion, oversupply could emerge. Equipment and materials localization is the third key theme. Except for DUV equipment, CXMT's Gen 4B primarily uses domestically produced equipment, and domestic suppliers are progressing through qualifications in etching, deposition, cleaning, CMP, coating and developing, and inspection. Localization penetration at YMTC's new fabs has approached 60%. However, immersion DUV, advanced metrology, manufacturing software, spare parts and maintenance, as well as HBM packaging and qualification, remain the least localized areas. U.S. semiconductor equipment export restrictions have already affected process-node migration, while Chinese manufacturers are mitigating these constraints by collaborating with domestic wafer fabrication equipment suppliers. Consequently, China's memory capacity expansion will continue to increase demand for domestically produced deposition, cleaning, inspection, and polishing equipment, while also intensifying competition in these equipment segments. Industry structure and returns will change accordingly. The number of major DRAM participants fell from 24 in 1996 to 3 in 2020 and will increase to 4 in 2026 with CXMT's entry, although the industry remains highly concentrated. Samsung, SK hynix, and Micron still account for more than 80% of the global DRAM market by revenue, while CXMT accounts for approximately 6%. Strategic memory—including HBM and advanced server DRAM—remains concentrated among the traditional Big Three, while China's influence in conventional DRAM and NAND is increasing. The report expects absolute memory industry capital expenditure to rise to record levels during 2026–2028 because of AI demand and China's capacity expansion. However, capital expenditure as a percentage of industry revenue will decline from its 2022 peak to the low teens by 2028 due to significant price increases. Traditional leaders must continue investing in EUV DRAM, HBM, advanced packaging, and 3D NAND to maintain their advantages in technology, cost, and product mix. The longer-term cost is a decline in sustainable returns. The report believes the current DRAM operating margin of more than 80% is at risk and assumes that through-cycle DRAM operating margins normalize to 50% in the outer years of its ten-year residual income forecast. Given YMTC's progress and more intense competition, NAND margins are expected to decline more rapidly to 15%–20%. Higher absolute capital expenditure and lower sustainable margins will reduce return-on-capital capacity and ROE. The industry may ultimately see segmentation rather than comprehensive substitution: China will increasingly influence marginal pricing in mainstream memory, while the traditional Big Three preserve differentiated, high-margin profit pools through HBM, advanced server DRAM, and enterprise storage. A credible threat to high-end profit pools would require CXMT to achieve competitive capabilities in advanced server DRAM and HBM, or YMTC to secure broad enterprise SSD qualification. From a valuation perspective, the report distinguishes cyclical peak rerating from long-term normalized levels. Strong memory price increases, AI demand, and improved earnings visibility from long-term agreements once pushed DRAM companies' average next-12-month price-to-book ratio to approximately 4.0 times, significantly above the typical historical peak of approximately 1.8–2.1 times, before it declined to the current approximately 2.4 times. Current valuations correspond to an ROE of approximately 50% in 2026, compared with a sustainable ROE of 10%–15%, similar to the 2010–2025 average. This indicates that China's market entry may not constrain peak-cycle valuations when upward revisions to prices and earnings are sufficiently strong. The more important question is where valuations will normalize once CXMT and YMTC scale up and competition in mainstream DRAM and NAND reduces through-cycle margins, sustainable ROE, and returns on capital. The report therefore believes Chinese supply is more likely to affect the long-term normalized valuation range and downside floor than to prevent rerating at cyclical peaks.

Analysis framework

Morgan Stanley first reviews the historical shifts in memory industry leadership among the United States, Japan, and South Korea, identifying the roles of scale, yields, cost per bit, and process migration in shaping the competitive landscape. It then separately estimates CXMT's DRAM and HBM capacity, bit shipment share, and supply-demand impact, as well as NAND scenarios for YMTC under different capacity expansion rates and AI SSD demand assumptions. The report then compares Chinese manufacturers with global leaders in cost, equipment, processes, packaging, and customer qualification, before translating supply changes into implications for industry capital expenditure, margins, ROE, and normalized price-to-book levels.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  DRAM and NAND Supply-Demand Balance Model

  The report incorporates AI demand, HBM's crowding out of capacity, new Chinese capacity, and capacity expansion by traditional manufacturers into a unified supply-demand model to assess the severity of shortages during 2026–2028 and determine when incremental supply will begin to affect prices.

• Industry/Sector Analysis FrameworkVolume-price decomposition

  Decomposition of Wafer Capacity, Bit Shipments, and Marginal Pricing

  The report distinguishes wafer scale from actual bit output and further considers yields, die area, process steps, and the HBM bit-output penalty, explaining why similar capacity does not imply equivalent costs or effective supply.

• Industry/Sector Analysis FrameworkCost curve analysis

  Comparison of Cost per Bit and Manufacturing Capabilities

  The report compares the cost positions of Chinese manufacturers and traditional leaders based on process integration, defect control, yields, mask counts, production cycle times, and bit output per wafer.

• Industry/Sector Analysis Framework

  2028 NAND Scenario Stress Test

  The report combines non-AI NAND demand growth, AI SSD demand growth, and YMTC's monthly capacity into multiple scenarios to determine the conditions for tight supply or oversupply.

• Valuation MethodRIM Residual Income Model

  Ten-Year Residual Income Forecast

  The report assumes gradual normalization of through-cycle DRAM and NAND margins in its ten-year forecast to assess the impact of intensified competition on long-term returns and valuations.

• Valuation MethodPB valuation

  Comparison of Next-12-Month Price-to-Book Ratio and Sustainable ROE

  The report compares current, recent peak, and historical peak price-to-book ratios and combines them with 2026 ROE and long-term sustainable ROE to distinguish cyclical peak rerating from long-term normalized valuation.

• Competition and Strategy FrameworkEconomies of scale / learning curve

  Scale-First Strategy and Manufacturing Learning Curve

  The report believes CXMT and YMTC are first accumulating yield, process, and supply-chain capabilities through domestic demand and continued capacity expansion. Even without comprehensive technological leadership, they can reshape industry competition through scale.

Asset mapping & comparison

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

• CXMT

  By expanding conventional DRAM and advancing domestically produced HBM, it is becoming an important variable in China's import substitution and global DRAM marginal pricing.

  Strengths

  Domestic demand, policy and capital support, Gen 4B yields exceeding 90%, rapid capacity expansion, and progress in DDR5, LPDDR5X, and HBM3E products.

  Weaknesses

  Reliance on immersion DUV and multiple patterning, with disadvantages in cost per bit, cycle time, high-grade output, HBM packaging, and customer qualification.

  Comparison

  It is expected to potentially surpass Micron in 2028, but still falls short of the traditional Big Three in HBM technological depth, global customer trust, and coverage.

  Risks

  Equipment export restrictions, process complexity, yields, and qualification progress could constrain node migration and high-end product volume growth.

• YMTC

  Its NAND architectural progress and fabs under construction make it the largest swing factor in global NAND supply and demand after 2028.

  Strengths

  Xtacking hybrid-bonding architecture, mass production of 267-layer products, development of products with more than 300 layers, and progress in enterprise SSD capacity, bandwidth, and endurance.

  Weaknesses

  It has not yet demonstrated the ability to secure broad qualification from major cloud providers, and its enterprise product portfolio has not reached the level of global leaders.

  Comparison

  The report expects it could become the world's second-largest NAND manufacturer in 2028. If all five fabs are used for NAND, its eventual global share could reach 24%.

  Risks

  If capacity expands faster than AI SSD demand, it could trigger NAND oversupply.

• Samsung, SK hynix, and Micron

  They still control strategic memory and more than 80% of global DRAM revenue, but China's capacity expansion could weaken their mainstream product share and through-cycle returns.

  Strengths

  Leadership in EUV processes, cost per bit, yields, HBM, advanced packaging, server DRAM, enterprise storage, and global customer qualification.

  Weaknesses

  HBM is highly capital- and capacity-intensive, while continued investment is required to maintain technological and product-mix advantages.

  Comparison

  Compared with Chinese manufacturers, the Big Three have clear technological and cost advantages, but their profit-oriented production discipline could leave them with greater responsibility for balancing the market if CXMT continues expanding capacity.

  Risks

  Intensified competition in mainstream DRAM and NAND could reduce sustainable margins, ROE, returns on capital, and long-term normalized valuations.

• Chinese Semiconductor Manufacturing Equipment Suppliers

  Expansion of Chinese memory fabs and equipment localization will continue to increase demand for deposition, etching, cleaning, CMP, coating and developing, and inspection equipment.

  Strengths

  They have participated in CXMT's Gen 4B build-out and are progressing through qualifications in multiple critical process steps; localization penetration at YMTC's new fabs is approaching 60%.

  Weaknesses

  Immersion DUV, advanced metrology, manufacturing software, spare parts and maintenance, and HBM packaging remain weak areas.

  Comparison

  The coverage of domestic equipment is expanding, but breakthroughs have yet to be achieved in several advanced tools and service capabilities.

  Risks

  Insufficient technical validation, customer qualification, and advanced equipment capabilities could constrain the pace of localization.

Key data

• CXMT Monthly Capacity180,000 wafers in 2025, 300,000 in 2026, 500,000 in 2028, and 800,000 in 2031The report expects approximately 100,000 wafers per month of annual capacity additions after 2026
• CXMT's 2026 Global DRAM Share13% of wafer capacity and 11% of bit shipmentsCorresponding to approximately 300,000 wafers per month of capacity
• CXMT's 2030 Global Bit Shipment ShareApproximately 15%Report forecast
• 2028 Global DRAM Wafer Capacity3.374 million wafers per monthCXMT could account for approximately 15%
• CXMT Gen 4B ProcessApproximately 16nm-class, with yields exceeding 90%The transition to 100% production was completed in the second quarter of 2026
• HBM Bit-Output PenaltyRising from approximately 2.5 times to approximately 3 timesFrom the 2021–2023 level to the forecast 2028 level
• HBM Share of Advanced DRAM Wafer ConsumptionApproximately 6% in 2023 and 34% in 2028HBM crowds out advanced DRAM wafer capacity
• CXMT's 2027 HBM3E OutputApproximately 3 million–4 million stacksExpected to support approximately 800,000–1 million domestically produced Chinese AI GPUs
• CXMT HBM Bit Shipment Share3.8% in 2026 and 4.4% in 2027Report forecast
• DRAM Supply ShortfallApproximately 17% in 2026 and approximately 15% in 2027Includes CXMT's planned HBM capacity ramp-up
• YMTC Capacity Under Construction100,000 wafers per month each at Fab 4 and Fab 5If all five announced fabs are used for NAND, its eventual global share could reach 24%
• 2028 NAND Scenario AssumptionsNon-AI demand growth of 5% year over year, AI SSD demand growth of 30%–60% year over year, and YMTC capacity of 310,000–470,000 wafers per monthUsed to test the risks of tight supply and oversupply
• YMTC Enterprise TLC SSD PerformanceSequential reads of 14.0–14.2GB/s and random reads of 3.1 million–3.3 million IOPSPCIe 5.0 product
• YMTC PE501 Capacity and Endurance122.88TB, 5,000 program/erase cycles, and 0.6 drive writes per dayQLC-based enterprise SSD
• Number of DRAM Industry Participants24 in 1996, 3 in 2020, and 4 in 2026The number increases in 2026 due to CXMT's entry
• Global DRAM Revenue ShareMore than 80% combined for the traditional Big Three and approximately 6% for CXMT2026 estimate
• Long-Term Operating Margin Assumptions50% for DRAM and 15%–20% for NANDNormalized outer-year levels in the ten-year residual income forecast; the current DRAM operating margin exceeds 80%
• DRAM Companies' Next-12-Month Price-to-Book RatioRecent peak of approximately 4.0 times, current level of approximately 2.4 times, and typical historical peak of approximately 1.8–2.1 timesUsed to compare cyclical peak and long-term normalized valuations
• DRAM Companies' ROEApproximately 50% in 2026 and a sustainable level of 10%–15%The sustainable level is comparable to the 2010–2025 average

Impact & implications

The report believes the near-term impact of China's memory capacity expansion will primarily be reflected in capital expenditure, import substitution, and market share rather than an immediate reduction in global prices. AI demand and HBM's crowding out of advanced capacity will continue to support tight supply in 2026–2027. Beginning in 2028, if Chinese manufacturers continue expanding capacity in pursuit of strategic objectives that differ from those of traditional leaders, the industry could shift from a disciplined three-company oligopoly to four competitors with asymmetric objectives, reducing the long-term margins, ROE, and returns on capital of mainstream DRAM and NAND. Traditional leaders may still preserve high-end profit pools through HBM, advanced server DRAM, and enterprise storage, while Chinese semiconductor equipment manufacturers will benefit from continued fab construction and demand for equipment localization.

Risks

• Beginning in 2028, simultaneous capacity expansion by Chinese and traditional manufacturers could weaken supply discipline and increase the risk of declining DRAM and NAND prices or oversupply.
• If CXMT continues expanding capacity during a demand downturn to pursue import substitution and technological learning, the traditional Big Three may be forced to undertake more production cuts, reducing long-term industry margins and returns on capital.
• If AI capital expenditure or AI SSD demand falls short of scenario assumptions, or if YMTC and other manufacturers accelerate greenfield expansion, NAND could shift from tight supply to oversupply.
• U.S. equipment export restrictions and insufficient capabilities in immersion DUV, advanced metrology, manufacturing software, spare parts, and maintenance could slow node migration and yield improvements at Chinese manufacturers.
• CXMT's HBM still faces constraints in yields, advanced packaging, testing, and accelerator qualification, while YMTC's enterprise SSDs have not yet secured broad qualification from major cloud providers.
• If Chinese manufacturers further expand into advanced server DRAM, HBM, and enterprise SSDs, competition will extend from the scale-driven pool of mainstream products into the high-end profit pools of traditional leaders.

What to watch

• Monitor whether CXMT can raise capacity to the planned 500,000 wafers per month by 2028 and become a swing supplier capable of influencing global marginal pricing.
• Monitor whether CXMT continues expanding capacity when demand weakens and how the traditional Big Three adjust capital expenditure, utilization, and bit growth.
• Monitor CXMT's yields, packaging, testing, and customer qualification progress as it begins mass production of 12-layer HBM3E in the first half of 2027 and advances HBM4 in 2028.
• Monitor the construction pace of YMTC's Fab 4 and Fab 5, the actual uses of all five fabs, and where its monthly capacity falls within the 310,000–470,000-wafer scenarios.
• Monitor whether AI SSD demand can achieve 30%–60% year-over-year growth in 2028 and whether non-AI NAND demand can achieve 5% year-over-year growth.
• Monitor whether YMTC can secure broad enterprise SSD qualification from major cloud providers.
• Monitor whether domestically produced equipment can achieve breakthroughs in weak areas such as immersion DUV, advanced metrology, manufacturing software, spare-parts maintenance, and HBM packaging.
• Monitor how competition in mainstream DRAM and NAND translates into through-cycle margins, sustainable ROE, and the normalized range for next-12-month price-to-book ratios.