China's CXMT: The $4.3B Memory Giant That Could Reshape the DRAM World

For three decades, the global DRAM market was effectively a closed oligopoly. Samsung, SK Hynix, and Micron controlled 95%+ of supply, enforced brutal pricing discipline through collective capacity restraint, and invested billions in R&D moats that kept every would-be competitor at bay. China tried repeatedly — Tsinghua Unigroup, JHICC — and failed spectacularly, often at the hands of US export controls.

CXMT is different. ChangXin Memory Technologies, headquartered in Hefei, is not another semiconductor start-up chasing headlines. On July 16, 2026, it lists on the Shanghai Star Market in the largest A-share IPO of the year — raising RMB 29.5 billion (~$4.1 billion). Its revenue grew approximately seven-fold in the first half of the year. It is closing in on Micron's wafer capacity. And it is being named — for the first time — as a principal competitor by SK Hynix in its own SEC filings.

This is the story of how China finally cracked the memory wall, what it means for the big three incumbents, and where the real opportunity and risk lies for investors.

The Capacity Race: CXMT vs. Micron — Closer Than You Think

The most striking data point in the CXMT story is raw wafer capacity. As of mid-2026, CXMT is running approximately 265,000 wafer starts per month (wspm). By year-end, SemiAnalysis projects that figure climbs to 350,000 wspm — within striking distance of Micron's estimated 385,000 wspm. By 2028, CXMT is forecast to hit 500,000 wspm, representing roughly 17% of all global DRAM supply.

That is a seismic shift. Five years ago, CXMT barely registered. Today, it is approaching the third-largest DRAM manufacturer by wafer volume — a position historically occupied by Micron — and its growth rate dwarfs anything the incumbents are capable of at this stage of the cycle.

But here's the crucial nuance: capacity volume alone does not determine competitive impact. Not all wafers are equal — and the gap between CXMT and the Big Three is nowhere near closing in the products that matter most for the AI era.

Where CXMT Is Winning: DDR5 Conventional DRAM

CXMT's strategy has been tactically shrewd. Rather than betting everything on High Bandwidth Memory (HBM) — where Samsung, SK Hynix, and Micron have invested a decade of advanced stacking IP — CXMT has concentrated its technology on conventional DRAM: DDR5 and LPDDR5X. These are the bread-and-butter memory products that power smartphones, PCs, laptops, mainstream servers, and most of the world's non-AI computing workloads.

The timing was fortuitous. As the Big Three pivoted resources toward HBM to serve AI customers — diverting capacity and engineering talent away from conventional DRAM — they created a supply gap in DDR5 that CXMT has aggressively filled. The result: DDR5 pricing hit exceptional highs in Q1 2026. CXMT's cost-per-bit remains more than 30% above the incumbents' — a meaningful disadvantage — but DDR5 pricing was elevated enough that CXMT reportedly achieved gross margins above 70% even at a cost disadvantage.

Market share figures confirm this trajectory. CXMT's global DRAM market share grew from approximately 3.97% to ~8% in the past year, and is projected to reach 12% by 2027. That growth comes almost entirely from DDR5 and LPDDR5X into Chinese domestic customers — smartphones (Huawei, Xiaomi), PCs, servers, and consumer electronics.

Where CXMT Is NOT Winning: HBM

The most important context for memory investors is the HBM gap — and it is enormous.

While Samsung and SK Hynix are racing toward HBM4 mass production, and Micron is shipping HBM3E in volume, CXMT is targeting HBM3E mass production in 2027 — one to two product generations behind. Of CXMT's 265,000 wspm total capacity today, approximately 5,000 wspm — less than 2% — is allocated to HBM production. By end of 2026, that's projected to reach 30,000 wspm. By end of 2027, 55,000 wspm.

This matters enormously because HBM is where the real money is. HBM pricing commands a significant premium over conventional DRAM — some estimates suggest 5–8× the ASP of equivalent DDR5. Every NVIDIA H100, H200, and GB200 chip requires HBM. Every AI training cluster from Google, Meta, Microsoft, and Oracle runs on HBM. And for the foreseeable future, none of that business is going to CXMT.

The IPO: China's Largest A-Share Listing of 2026

On July 16, 2026, CXMT begins trading on the Shanghai Star Market (STAR Market). The IPO targets RMB 29.5 billion (~$4.1–4.3 billion), making it the largest A-share IPO of 2026 and one of the most significant semiconductor listings in Chinese market history.

The revenue trajectory justifies the ambition. CXMT's revenue grew approximately seven-fold in recent periods as DDR5 pricing surged and domestic Chinese demand — insulated somewhat from US export controls by focusing on non-AI conventional DRAM — expanded rapidly. SemiAnalysis analysts project CXMT could eventually generate $50 billion+ in annual revenue at full scale, though that figure assumes successful HBM entry and further capacity expansion.

IPO proceeds will fund the next phase of the capacity build-out — moving from 350,000 wspm to 500,000 wspm by 2028 — along with R&D acceleration in advanced DRAM process nodes. The capital raise also signals Beijing's continued strategic commitment to semiconductor self-sufficiency as US export controls on advanced chip equipment tighten.

Building the Vertical Supply Chain

What makes CXMT structurally different from China's previous memory failures is the vertical integration effort underway. CXMT and its aligned partners are not just building fabs — they are building an entire DRAM ecosystem:

• Photoresist and chemicals: Chinese chemical suppliers have been qualifying photoresists and CMP slurries for DRAM processes, reducing dependence on Japanese and US suppliers.
• Equipment localisation: While ASML EUV machines remain inaccessible due to export controls, CXMT has optimised its DDR5 production around DUV-compatible processes — a tactical adaptation that sacrifices some density but maintains viability.
• Packaging: Domestic advanced packaging capacity is being built to support eventual HBM stacking requirements.
• Test and assembly: Chinese OSAT companies are scaling test infrastructure for DRAM qualification.

This is not a complete supply chain — critical gaps remain in EUV equipment, advanced pellicles, and certain electronic chemicals. But it represents far more self-sufficiency than any Chinese memory attempt before it.

What This Means for Samsung, SK Hynix, and Micron

The competitive impact is not uniform across the Big Three. Investors should think about this segment by segment.

SK Hynix (000660.KS) — Most Insulated

SK Hynix has the highest HBM exposure of the three. It supplies approximately 50–55% of all HBM globally, and NVIDIA remains its anchor customer. HBM for 2026 is sold out. CXMT cannot touch this business in the near or medium term. Hynix's main vulnerability is that any DRAM commodity price deterioration from CXMT supply eventually bleeds into blended ASPs across the broader market — but its revenue mix is skewed enough toward premium HBM that the damage is mitigated. SK Hynix explicitly named CXMT as a major competitor in SEC filings — a candid acknowledgement of the long-term risk.

Samsung (005930.KS) — Dual Exposure

Samsung faces CXMT on two fronts. In conventional DRAM, Samsung's dominant global position — particularly in server DDR5 and consumer LPDDR5 — is most directly threatened by CXMT pricing pressure. In HBM, Samsung is racing to close its own gap with Hynix on HBM4 — a separate, internal challenge. The combined competitive pressures from both directions make Samsung's near-term DRAM earnings trajectory the most uncertain of the three.

Micron (MU) — Most Exposed to CXMT

Micron's market position makes it most directly vulnerable. It has the smallest HBM share (5–10%), the smallest overall scale, and the highest reliance on conventional server DRAM and consumer memory — exactly the segments where CXMT is competing. Micron is working hard to ramp HBM3E and close the gap with Hynix, but its conventional DRAM pricing power is the most sensitive to CXMT supply additions. Analysts have already noted that CXMT's expansion is a meaningful headwind for Micron's commodity DRAM business over the next 2–3 years.

The Investment Angle: Is the Threat Overstated?

Multiple analysts — including those following the IPO closely at Seoul Economic Daily — argue the near-term CXMT threat to Samsung and SK Hynix is overstated, specifically because of the HBM gap. T