CXMT Advances HBM3E Mass Production, Mounting Competitive Pressure on Micron from China's Memory Sector
nashnova research
CXMT (长鑫存储) has begun low-volume production of HBM3E memory chips, with Cambricon and Alibaba's T-Head testing them for use in custom processors as early as next year. This means → Micron's share of China's AI chip supply chain faces real, structural pressure.
What is HBM3E, and how far has CXMT come?
HBM3E — high-bandwidth memory designed for AI processors, used in Nvidia's Blackwell GPUs and Google's latest TPUs — is now in low-volume production at CXMT.
CXMT's product trails the global leaders by one generation; SK Hynix began mass-producing HBM3E in 2024.
But CXMT's overall HBM technology lags the front-runners by three to five years, and yields remain low. In plain terms = it can make the chip, but large-scale, high-yield production is still a clear distance away.
Who is testing it, and what do the orders really mean?
Cambricon (寒武纪) and Alibaba's T-Head Semiconductor (平头哥) are testing CXMT's HBM and plan to integrate it into custom processors as early as next year.
This means → the core value of these orders is not revenue scale — it is the production experience and customer feedback needed to lay the groundwork for capacity expansion in 2027.
Central and local governments are absorbing much of the cost through tax breaks and subsidies. This reflects a phase where domestic substitution in memory has moved from lab proof to "paying real money for real-world data."
How do US export controls both help and hinder CXMT?
The US restricted advanced HBM exports to China in 2024; even lower-spec memory products now require approval. CXMT's progress partly eases this supply bottleneck for Chinese firms.
Yet the same controls also block CXMT from purchasing advanced manufacturing equipment, capping its production capacity.
In plain terms = export controls are both the reason CXMT has a captive domestic market and the biggest obstacle to scaling up — a double-edged sword.
How strong is CXMT's financial position?
The company's Shanghai IPO raised RMB 66.6 billion — China's second-largest ever — and shares surged 472% on the first day.
First-half revenue grew nearly tenfold year-on-year to RMB 150.3 billion, with net profit of RMB 77.6 billion, swinging from a loss in the year-ago period.
This means → CXMT has cash and momentum, but its earnings report made no mention of HBM — the technology has not yet contributed disclosable revenue.
How will the pressure on Micron evolve?
CXMT is now the world's fourth-largest DRAM producer. Barron's argues that competition from Chinese memory chips poses pressure on Micron (MU) that "will not go away."
The pivotal variable is whether CXMT can achieve volume production despite yield and equipment constraints — that directly determines how much further Micron's share of China's AI supply chain shrinks.
In plain terms = yields and equipment are hard bottlenecks in the near term, but the direction is set: domestic HBM has moved from "can it be done?" to "how much can be made?"
市场有风险，内容仅供研究参考，不构成投资建议。