CXMT’s G5 DRAM Push and Possible NAND Entry

ChangXin Memory Technologies (CXMT) is becoming a more consequential DRAM supplier. Its G5 platform has entered mass production, while a reported plan for NAND research could eventually take the company beyond working memory into storage. The distinction is important: G5 is a production announcement; NAND remains a possible next move.845

What G5 adds to CXMT’s DRAM business

DRAM is the fast, temporary working memory that holds information processors are actively using in phones, PCs, servers and AI systems. It needs power to retain that information. CXMT announced mass production of its fifth-generation G5 manufacturing platform alongside two 24-gigabit LPDDR5X products, extending its range of high-density, low-power mobile DRAM.488

The headline manufacturing claim is at least 50% more gross dies per wafer than G4, measured on an 8-gigabit-equivalent basis. That common-capacity measure allows a comparison of potential wafer output; it does not mean the newly announced LPDDR5X products are 8-gigabit chips. Nor does gross die count account for chips lost to defects, testing or packaging.37

The cost implication is therefore conditional. If processing costs and yields stayed identical, 1.5 times as many equivalent dies would lower processing cost per equivalent die by about one-third—not by 50%. Actual costs depend on yield and the expense of the new process. Earlier analysis cited by The Korea Herald estimated CXMT’s DDR5 cost per bit at more than 30% above that of the three leading suppliers; the G5 announcement alone does not show that gap has closed.19

CXMT also reports an 11.95-nanometer active-area half-pitch in G5’s memory array, achieved with self-aligned quadruple patterning.49 Half-pitch describes a particular repeated structure, not the dimensions of every feature on the chip. It should not be treated as a direct measure of whole-chip performance or manufacturing cost.

How close is CXMT to the DRAM leaders?

TrendForce figures reported in September put CXMT fourth in second-quarter 2026 DRAM revenue, at 9.5%. Samsung held 39.4%, SK hynix 24.9% and Micron 23.3%.6 A separately reported Counterpoint estimate put CXMT at 10%, a reminder that market-share figures vary by research firm.4

That growing position does not erase the technology gap. A Korea Semiconductor Industry Association assessment put China roughly two years behind Korea in DRAM and three years behind in high-bandwidth memory (HBM), broad estimates rather than measurements of G5 specifically.30 Reports describe small-scale CXMT production of HBM3E and testing with Chinese chip designers, but they do not establish competitive high-volume output or a firm timetable for later generations.3430

CXMT may also benefit from where the incumbents are investing. As Samsung, SK hynix and Micron prioritize AI-related HBM and server memory, tighter supply growth in some conventional DRAM products can leave room for smaller suppliers.22 The available evidence does not show how much of CXMT’s market-share gain that shift explains.

NAND is a research plan, not a product launch

NAND flash performs a different job from DRAM: it retains data without power and is used for persistent storage, including solid-state drives and phone storage. Entering NAND would require CXMT to develop and qualify a distinct manufacturing technology, rather than simply sell its G5 DRAM through another channel.

Reuters reported on September 18, citing people familiar with the plans, that CXMT intended to establish a NAND research-and-development production line at its new Beijing plant.45 If successful, the move would place CXMT in a market already served by established suppliers and by China’s NAND specialist, Yangtze Memory Technologies (YMTC).3545 But a research line is not a commercial-scale factory. The cited reporting does not establish a launch date, production capacity or firm customer commitments.

For a sense of the incumbent scale, TrendForce put Samsung’s second-quarter NAND share at 29.3% and SK hynix’s at 18.2%.6 Those figures describe an existing market that CXMT is considering entering—not NAND sales it has already won.

What this means for competition

CXMT’s immediate challenge to Korean memory makers is clearest in conventional DRAM: it has a larger reported revenue share and a new platform intended to improve wafer productivity.63 Tight supply and sustained strong prices could give a higher-cost challenger more opportunity to sell, invest and refine production. They would not, by themselves, prove that its yields or costs match Samsung’s or SK hynix’s.1922

The two developments should be judged on different evidence. G5’s next test is whether its density gains translate into saleable output and competitive costs. For NAND, the first test is whether the reported Beijing research effort produces a qualified commercial product at all.345