China's Memory Gamble: Why CXMT's NAND Push Changes Everything

The quiet unraveling of China’s memory strategy

For years, the story of China’s semiconductor ambition ran along two clean tracks: CXMT in DRAM, YMTC in NAND. It was an arrangement born of necessity and convenience — two companies, one country, separate lanes. The logic was straightforward enough: avoid duplicating hundreds of millions in R&D, let each firm specialize, and together you build a domestic memory supply that reduces reliance on Taiwan and Korea.

That arrangement is now fracturing. According to a report from CBS No Cut News, CXMT is preparing to build a 3D NAND research and production line at its second Beijing facility. It has opened a separate research institute focused on NAND technology and has begun discussions with potential customers about supply feasibility. The commercial production timeline is undecided. But the direction is unmistakable.

CXMT is no longer satisfied with being China’s DRAM answer. It wants to be a comprehensive memory company — the kind of player Samsung and SK Hynix have been for decades.

What CXMT just accomplished in DRAM

Before looking ahead to NAND, it is worth understanding the foundation CXMT is building on. At the 2026 World Manufacturing Conference in Hefei last month, the company announced mass production of its fifth-generation DRAM platform, designated G5.

The G5 platform uses self-aligned quadruple patterning (SAQP) to shrink the active area half-pitch of memory cells down to 11.95 nanometers. By CXMT’s own accounting, this means more than a 50 percent increase in die yield per wafer compared with the G4 generation when producing 8Gb equivalents. The company also began mass production of two 24Gb LPDDR5X products using the G5 platform, claiming a 50 percent capacity increase per die over prior generations. Earlier this month, it announced mass production of LPDDR6 with maximum transfer speeds of 12.8Gbps.

The numbers are impressive in isolation. In the globalDRAM market, CXMT captured 9.5 percent share in the second quarter, up from 7.6 percent in the first. Its quarterly DRAM revenue nearly doubled, rising 99.3 percent. But the gap with the leaders remains stark. Samsung holds 39.4 percent, SK Hynix 24.9 percent, and Micron 23.3 percent. Samsung and SK Hynix are already shipping sixth-generation 1c process nodes in volume products, including Samsung’s HBM4 since February. CXMT’s G5 is competitive but not yet at that level.

In NAND, the distance is even larger. Samsung commands 29.3 percent of global NAND revenue in Q2, while SK Hynix (including Solidigm) holds 18.2 percent. CXMT has no commercial NAND presence at all.

The structural implications of a two-front challenger

Here is what Western coverage of Chinese semiconductors tends to miss: CXMT’s move into NAND is not just another entry by a Chinese competitor. It is a fundamental restructuring of the competitive landscape for an entire class of products.

Samsung and SK Hynix have spent decades building integrated memory divisions — companies that design, manufacture, and sell both DRAM and NAND. Their customers, particularly in server and data-center segments, value this because it simplifies procurement, creates bundled pricing power, and gives them a single point of accountability for performance across memory types. A comprehensive memory supplier is not the same as a DRAM-only supplier with ambitions.

When CXMT enters NAND commercially, it does not merely add a product line. It changes the nature of the rivalry. Samsung and SK Hynix will no longer face a Chinese DRAM specialist and a Chinese NAND specialist operating in separate markets. They will face one Chinese company that can compete against them on both fronts simultaneously.

This intensifies pressure in ways that go beyond simple capacity expansion. A dual-product Chinese supplier can leverage cross-subsidization, offer bundled contracts to Chinese cloud providers and device makers, and use one business line to fund R&D in the other. That is the exact model that allowed Samsung to overtake Korean and Japanese rivals in the 1990s.

The collision course with YMTC

Perhaps the most underappreciated consequence of CXMT’s NAND pivot is what it means for YMTC, China’s dedicated NAND champion.

Under the old division-of-labor model, CXMT and YMTC were not direct competitors. Each had a clear mandate and could focus capital and talent without duplicating effort. Now their territories overlap. Both will be investing in NAND fabrication capacity. Both will be competing for the same domestic customers — Xiaomi, Huawei, Tencent, Baidu, the growing cohort of Chinese AI chip designers who need local NAND supply.

This creates a tension that Beijing faces directly. Does it encourage healthy competition between its two memory champions, risking wasteful duplication and margin compression? Or does it try to re-impose some form of coordination, potentially stifling the very urgency that drives rapid technology catch-up?

Kim Yang-paeng, a senior researcher at the Korea Institute for Industrial Economics and Trade, told reporters that CXMT’s entry into NAND signals genuine global NAND scarcity — otherwise, there would be no reason for a Chinese firm to make the gamble. He also warned that while the immediate threat to Korean companies is limited, a fully integrated memory competitor emerging from China is exactly the kind of long-term risk that could eventually undermine Samsung and SK Hynix’s market position.

The American variable

No analysis of CXMT’s trajectory is complete without addressing the elephant in the room: United States export controls on advanced semiconductor manufacturing equipment.

CXMT, like all Chinese memory makers, faces severe constraints on acquiring the most advanced lithography and process equipment needed to push further into smaller nodes — both in DRAM and in the 3D NAND architecture it is now pursuing. The controls are not absolute. There are loopholes, workarounds, and a persistent gray market. But they impose real friction on the pace of technological advancement.

The timing of this report adds weight to that uncertainty. A US-China meeting is scheduled for September 24, and while semiconductor export policy is unlikely to be a formal agenda item, any signal of relaxation would materially alter CXMT’s cost curve. Kim Yang-paeng noted that outright deregulation is unlikely, but warned that “the moment restrictions ease, our companies will face significant difficulty.”

Conversely, if the US tightens controls further — particularly targeting the equipment CXMT needs for 3D NAND production — the company’s timeline could slip by years, not months. The R&D line at the Beijing factory may remain a research project rather than a commercial asset for some time.

What happens next

The memory supercycle is already in motion. Prices for both DRAM and NAND have been rising as AI demand, data-center expansion, and automotive computing drive consumption beyond what existing capacity can absorb. Into this environment steps a Chinese company that now claims capability in both memory types and is actively building the infrastructure to prove it.

For Samsung and SK Hynix, the short-term impact is manageable. CXMT’s NAND ambitions are still in the research phase, and its DRAM share, while growing fast, remains a fraction of the leaders’. The immediate competitive response will likely involve aggressive capacity planning and customer retention strategies aimed at locking in long-term supply agreements with Chinese buyers before CXMT can scale.

The medium-term picture is less certain. If CXMT brings even a modest amount of NAND capacity online within two to three years, it will add a new marginal supplier to a market that is already running tight. That suppresses price upside for everyone — including Samsung and SK Hynix — precisely when they are counting on the supercycle to fund the next generation of HBM and high-capacity NAND development.

And if US export controls ease at any point, the calculus changes dramatically. A constrained CXMT is a manageable competitor. A well-equipped one is a structural threat to the Korean incumbents’ integrated memory model — the very model that has defined their dominance for forty years.

The old story was simple: China was catching up in DRAM, and separately in NAND. The new story is that a single Chinese company is attempting to do both at once, and the global memory industry will have to adjust to that reality.