SemiAnalysis: CXMT's Monthly Capacity May Reach 350K Wafers by End of 2026, Closing In on Micron
Claire Weston
SemiAnalysis projects CXMT will reach roughly 350,000 wafers per month by end-2026, just below Micron's 385,000 — putting China's largest DRAM maker within striking distance of the world's number-three slot by capacity.
What does 350K wafers a month actually mean?
CXMT's projected end-2026 capacity of ~350,000 wafers/month (300 mm equivalent) trails Micron's 385,000 by roughly 10%.
This means → by wafer output alone, CXMT would rank just behind the global top three: Samsung (~720K/month) and SK Hynix (~595K/month).
In plain terms = DRAM — the memory chips that give computers and phones their short-term working space — has long been a three-player oligopoly. CXMT is turning the triangle into a four-way contest.
How fast is CXMT actually expanding?
From 2026 to 2028, CXMT is expected to add roughly 85K, 70K, and 80K wafers/month per year — the largest annual increments of any major supplier.
Over the same period, Samsung adds 15K, 60K, 60K; SK Hynix 50K, 60K, 90K; Micron 30K, 90K, 115K.
This means → CXMT leads the expansion race in the near term, but by 2028 both Micron and SK Hynix ramp harder, and the gap could widen again.
With this much new capacity, will DRAM oversupply?
SemiAnalysis expects DRAM supply to remain high-single-digit percent short this year, with the deficit widening to low-to-mid double digits next year.
The tight market is projected to last through 2028 — even after CXMT's Shanghai fab ramps past 400K wafers/month.
In plain terms = demand is growing faster than capacity, so the "built too much, can't sell it" scenario is not the near-term risk.
How will global market share shift?
CXMT's share of global DRAM capacity is forecast to rise from 13% in 2025 to 17% by end-2026.
Its share of actual shipments is expected to climb from roughly 9% today to 12% by 2027.
This reflects a gap between capacity share and shipment share — the fabs are going up, but yield rates and product mix still need time to catch up.
HBM — where is CXMT's weak spot?
On HBM — high-bandwidth memory, the premium chips that feed data to AI processors — CXMT allocates only about 5,000 wafers/month to HBM as of end-2025, a tiny fraction of total output.
Its prospectus shows roughly 99% of 2025 revenue comes from DDR and LPDDR (standard PC and mobile memory), with HBM nearly negligible.
SemiAnalysis expects HBM capacity to accelerate sharply in 2027–2028, potentially reaching 100K wafers/month by end-2028 and lifting CXMT's global HBM supply share from 1% to 12%.
This means → CXMT's current growth story is built on high-volume commodity memory. HBM — where margins are far higher — is just getting started, and whether the technology ramp hits schedule is the biggest uncertainty.
How might policy shape capacity allocation?
The report notes that China's push for AI-compute self-sufficiency could steer CXMT to tilt more capacity toward HBM.
Government influence is expected to grow over time — this means → capacity allocation may not be driven purely by market pricing; policy direction becomes a key variable.
Put simply = what CXMT builds next will depend not only on what is most profitable, but also on what the state deems most necessary.
Content is for reference only, not financial advice.