Morgan Stanley: CXMT Could Surpass Micron by 2028 as China's Memory Sector Reshapes Global Competition
Nashnova编辑部
Morgan Stanley projects that CXMT could surpass Micron in capacity by around 2028, while YMTC may become the world's second-largest NAND maker in the same timeframe — China's memory sector is shifting from a technology catch-up story to a scale-driven reshaping of global pricing and profit structures.
How could CXMT overtake Micron?
Morgan Stanley's capacity projection for CXMT: 300K wafers/month by 2026 → 500K by 2028 → 800K by 2031. By around 2030, its global DRAM bit shipment share could reach roughly 15%.
On technology, CXMT has fully transitioned to Gen 4B — roughly a 16nm-class process — with yields above 90%. But without access to EUV lithography (etching circuits with extreme-ultraviolet light), it relies on DUV plus multi-patterning, requiring more masks and longer cycle times. This means → a structural cost-per-bit disadvantage.
The strategy is scale first, technology parity later. In plain terms = CXMT does not need to match the leaders node-for-node — once its incremental output is large enough, it can move global marginal pricing on its own.
Why does the HBM boom actually help CXMT?
HBM — high-bandwidth memory, the specialized fast memory AI chips need — carries a bit-output penalty versus standard DRAM. That penalty is expected to rise from roughly 2.5× in 2021–2023 to about 3× by 2028. HBM's share of leading-node wafer consumption is projected to jump from about 6% in 2023 to 34% by 2028.
This means → Samsung, SK Hynix, and Micron are channeling their best production lines into HBM, shrinking capacity available for mainstream DDR5 and LPDDR5X — exactly the market CXMT is scaling into.
Even with still-low yields, Morgan Stanley's research suggests CXMT could produce roughly 3–4 million HBM3E stacks by 2027, enough to support about 800K–1M domestic Chinese AI GPUs. Even including this ramp, the bank's model still shows a 15%–17% global DRAM supply gap in 2026–2027.
Where does YMTC stand?
YMTC is adding roughly 100K wafers/month per year in capacity. Morgan Stanley expects it could become the world's second-largest NAND maker by around 2028.
Equipment and supply-chain localization at its new fabs has reached nearly 60%. This reflects a deliberate effort to reduce dependence on imported equipment ahead of further export restrictions.
Why is 2028 the critical inflection point?
Before 2028, strong AI demand, HBM's capacity absorption, import-substitution cycles, and product qualification timelines should absorb much of China's incremental supply.
From 2028 onward, that buffer may erode. CXMT and YMTC could be large enough — layered on top of the incumbents' own record expansions — to directly influence global pricing and how the industry responds to downturns.
In plain terms = the real risk is not "cheap Chinese DRAM" per se. It is the industry shifting from three players with aligned profit incentives to four players with asymmetric strategic goals — CXMT's objectives also include import substitution, supply-chain resilience, and semiconductor sovereignty, meaning its behavior in a downturn could diverge sharply from the big three.
What does this mean for global memory investment?
Morgan Stanley identifies three key implications: higher absolute capex requirements, lower long-run normalized returns, and wider valuation dispersion.
Strategic memory — HBM and advanced server DRAM — remains concentrated at Samsung, SK Hynix, and Micron. But China's rising share in commodity DRAM and NAND means high-cost, low-scale players face pressure first.
This reflects a familiar pattern: memory leadership has shifted from the U.S. to Japan to Korea, each time driven by a new entrant combining scale with a change in the competitive basis. Morgan Stanley sees the ultimate test as whether CXMT becomes credibly competitive in advanced server DRAM and HBM, or YMTC wins broad enterprise SSD qualification — that would mark the shift from contesting the industry's volume pool to challenging its profit pool.
市场有风险，内容仅供研究参考，不构成投资建议。