CXMT Chases NAND as YMTC Storms Into DRAM Turf [2026]

China’s two biggest memory chipmakers are stepping onto each other’s turf. ChangXin Memory Technologies (CXMT), the country’s dominant DRAM producer, is reportedly building a NAND flash research-and-development line at a new plant near Beijing, according to a Reuters report published September 18, 2026, and confirmed in follow-up coverage from Tom’s Hardware and TrendForce. At the same time, Yangtze Memory Technologies (YMTC), China’s leading NAND flash maker, has been sending low-power DRAM (LPDDR) samples to customers, with TrendForce reporting that YMTC expects feedback by the end of 2026 before deciding whether to move to production.

Neither move is confirmed as commercial production yet. But the timing matters. Global memory prices are climbing through a supply squeeze driven by AI data center demand — TrendForce reported that CXMT’s own DRAM contract prices nearly doubled year-on-year in the first quarter of 2026 — and Chinese suppliers appear to be using that shortage as an opening to break out of their single-product lanes and start looking more like Samsung, SK Hynix, and Micron, each of which sells DRAM, NAND, and HBM under one roof.

·

Google · Preferred Sources

Don't miss new tech stories on Google

Add Tech Insider once in the Google app and our stories appear in your news suggestions.

Add Now

What CXMT is reportedly building in Beijing

Reuters, citing people familiar with the plans, reported that CXMT intends to establish a NAND-flash R&D production line at a new manufacturing facility near Beijing. Tom’s Hardware’s follow-up report described the project as tied to a research institute CXMT has set up in the capital, with NAND listed among the institute’s projects — though CXMT itself has not publicly confirmed a commercial NAND roadmap. That distinction matters: this is an R&D and trial-production effort, not an announced mass-production line with a disclosed layer count, process node, or output capacity.

CXMT’s core business today is DRAM, where it has spent the past several years closing the gap with Samsung, SK Hynix, and Micron. Tom’s Hardware reported that CXMT’s fifth-generation DRAM process, built on quadruple patterning at roughly the 12-nanometer class, could increase die capacity by approximately 50% over its prior generation, and CRN Asia reported that CXMT’s wafer capacity had already climbed to roughly 290,000 wafers per month by December 2025. The company began mass-producing LPDDR6 on August 29, 2026, by its own account, and Counterpoint Research estimated CXMT held about 10% of global DRAM revenue share in the second quarter of 2026 — evidence that its DRAM ramp is translating into real market position, not just process roadmaps. The company has also reportedly begun risk production of HBM3E, the high-bandwidth memory format used in AI accelerators, with mass production possibly arriving in 2027 — a claim Tom’s Hardware flagged as a developing report rather than an established commercial milestone.

A NAND push would put CXMT in direct competition with its domestic rival YMTC, plus Samsung, SK Hynix, Micron, Kioxia, and SanDisk — a much more crowded field than the DRAM market CXMT already fights in. According to reporting cited by TrendForce and Tekedia, one potential early customer for CXMT’s NAND output is a newly established startup seeking flash storage for AI systems and supercomputers, though that customer has not been named.

The stakes for CXMT’s home market are significant. China has spent years pushing for greater semiconductor self-sufficiency, and a domestic NAND supplier that can match even a fraction of YMTC’s output would give Chinese device makers and cloud providers another qualified source outside Samsung, SK Hynix, Micron, Kioxia, and SanDisk. That’s a strategic goal independent of whether CXMT’s NAND business ever becomes commercially competitive on price or yield with the market leaders.

YMTC’s quiet move into DRAM territory

YMTC’s side of the story moves in the opposite direction. TrendForce, summarizing an earlier Reuters report, said YMTC has sent LPDDR (low-power DRAM) samples to customers and is waiting on feedback before the end of 2026 to decide on production. A separate report attributed to DigiTimes puts a number on the ambition: an initial LPDDR5 trial production capacity of roughly 25,000 wafers per month, though that figure and timeline have not been confirmed by YMTC itself. The scale of ambition behind that sampling effort looks larger in context — Caixin Global has reported that YMTC is planning roughly $24 billion in investment to build out three production lines with combined capacity of about 300,000 wafers per month by 2026.

YMTC built its reputation on 3D NAND, where it has narrowed the technology gap with Western and Korean rivals faster than almost anyone expected a few years ago — Counterpoint Research estimated YMTC reached about 14% of global NAND flash revenue share in the second quarter of 2026, enough to rank third worldwide behind only Samsung and SK Hynix. A move into LPDDR would mirror what CXMT is reportedly attempting with NAND: turning a single-product memory maker into a broader supplier that can sell both DRAM and flash into the same customer relationships, the same way Samsung, SK Hynix, and Micron already do.

How far behind is China, really?

The most concrete technology-gap numbers in this story come from the Korea Semiconductor Industry Association, as reported by Wccftech. The association’s assessment: China trails the West by roughly one year in NAND, about two years in DRAM, and around three years in HBM. That ordering explains the strategic logic here — YMTC is closest to parity in the category it already leads (NAND), so a DRAM side bet is a bigger reach. CXMT is further behind in DRAM than YMTC is in NAND, and it’s furthest behind of all in HBM, the category central to AI accelerator supply chains.

Memory category	Estimated gap vs. Western/Korean leaders	Chinese leader
NAND flash	~1 year	YMTC
DRAM	~2 years	CXMT
HBM (high-bandwidth memory)	~3 years	CXMT (reported HBM3E risk production)

Source: Korea Semiconductor Industry Association assessment, as reported by Wccftech.

Why this is happening during a memory shortage, not despite it

The backdrop to both moves is a global memory market that has been tight for most of 2026. Reuters described flash memory demand as booming, and reported that the shortage is giving Chinese suppliers an opening to widen their customer base beyond domestic buyers. AI infrastructure spending — server DRAM, HBM for accelerators, enterprise SSDs, and high-capacity NAND for training and inference clusters — is the primary driver cited across the reporting.

That tightness has already shown up elsewhere on this site. Acer warned that PC prices could jump 20% as the memory crisis squeezes component costs, and GPU buyers have felt it too, with reporting on GPU price relief pushed out to 2028 as DRAM prices jumped a record amount. Handheld gaming hardware hasn’t been spared either — the Steam Deck price hike tied to the DRAM crisis and rising sim racing GPU prices both trace back to the same upstream supply constraints now pulling CXMT and YMTC into new product categories.

On the supply-forecast side, SK Hynix CEO Kwak Noh-jung reportedly said in July that 2027 could be the industry’s worst year yet from a supply standpoint, and TrendForce has been cited as expecting NAND tightness to ease only in the second half of the following year. If that holds, the shortage that’s currently pulling CXMT and YMTC into each other’s markets won’t resolve quickly — it’s a multi-year window, not a short-term spike.

The competitive landscape: who actually has the most to lose

Samsung, SK Hynix, and Micron remain the incumbents across DRAM, NAND, and HBM, and none of the current reporting suggests CXMT or YMTC are close to challenging them at scale. But “close” isn’t really the point yet — the point is optionality. A memory maker that only sells DRAM is exposed if DRAM pricing swings against it; a memory maker that sells DRAM and NAND has a hedge. That’s the model Samsung, SK Hynix, and Micron have run for years, and it’s the model both CXMT and YMTC appear to be reaching for now.

Kioxia and SanDisk, both NAND-focused suppliers, would face new competition if CXMT’s flash effort ever reaches volume production — though that’s a bigger “if” than YMTC’s DRAM sampling, which at least has a customer-feedback timeline attached (end of 2026). The reporting is explicit that neither company has disclosed a commercial output target, a market-share goal, or a pricing strategy for the new category it’s entering. Execution, not ambition, is the open question.

Company	Current core business	Reported 2026 move	Status
CXMT	DRAM (China’s leading domestic producer)	NAND R&D line at new Beijing facility	Reported, not formally confirmed
YMTC	3D NAND flash (China’s leading producer)	LPDDR/DRAM samples sent to customers	Customer-feedback stage, decision expected by end of 2026
Samsung	DRAM, NAND, HBM, SSDs	Incumbent across all categories	Global benchmark
SK Hynix	DRAM, HBM, NAND via Solidigm	Major AI-memory supplier	CEO warned 2027 could be worst supply year yet
Micron	DRAM, NAND, HBM	Integrated memory supplier	Competes directly with both Chinese firms
Kioxia / SanDisk	NAND and flash storage	N/A	Exposed if CXMT NAND reaches volume

Historical context: how CXMT and YMTC got here

CXMT and YMTC have followed different paths to this moment. CXMT built its position almost entirely on DRAM, using process advances like the reported fifth-generation quadruple-patterning node to gradually shrink its gap with Samsung and SK Hynix. YMTC took the opposite route, focusing on 3D NAND stacking and closing what was once a multi-year technology deficit down to roughly a year, per the Korea Semiconductor Industry Association’s assessment cited by Wccftech.

Both companies have operated for years under U.S. export restrictions on advanced chipmaking equipment, which has shaped how they source lithography and deposition tools and, by extension, how fast they can move into new memory categories. Neither Reuters nor Tom’s Hardware’s reporting attributes the current NAND and DRAM moves directly to those restrictions easing or tightening — the stated driver in the reporting is the memory shortage and the AI demand behind it, not a change in export policy.

What “R&D line” actually means in practice

It’s worth being precise about what has and hasn’t been confirmed. An R&D production line is a pilot facility used to develop and validate a manufacturing process — it produces test wafers, not commercial volume. Getting from an R&D line to qualified, sellable product typically takes multiple years even for an experienced memory maker, and CXMT has no public NAND product history to draw on. The Reuters and Tom’s Hardware reporting is careful not to characterize this as a NAND product launch; it’s a first step, and one CXMT hasn’t even confirmed on the record.

YMTC’s LPDDR sampling is a step further along in one sense — samples are already in customers’ hands — but a decision on whether to proceed to production is still pending, with a stated internal deadline of the end of 2026 for gathering that feedback. If DigiTimes’ reported 25,000-wafer-per-month figure for LPDDR5 trial production holds up, that would still be a small fraction of what established DRAM makers run at scale, positioning it as a toe in the water rather than a full pivot.

Market impact: what buyers should actually expect

For now, the direct market impact of these moves is limited, and the reporting is explicit on that point: neither company has disclosed pricing plans, output targets, or a commercial timeline for its new category. What’s more likely to move prices in the near term is the existing shortage itself, which is already reflected in reporting on GPU and PC component pricing across 2026. If CXMT and YMTC do eventually add meaningful NAND and DRAM capacity respectively, the effect would be gradual — additional qualified supply entering a tight market over a period of years, not a sudden price correction.

Enterprise and AI infrastructure buyers are the segment most exposed to this story either way. HBM remains the tightest and most strategically important category, and CXMT’s reported HBM3E risk production — with mass production possibly arriving in 2027 — is arguably more consequential to watch than its NAND ambitions, since HBM supply is currently the binding constraint on AI accelerator output industry-wide.

Consumer-facing hardware makers are watching from a different angle. PC and console manufacturers already passed rising memory costs on to buyers this year, and any easing of the DRAM and NAND squeeze — however far out — would be welcome news for a hardware market that’s spent 2026 absorbing component price shocks. Until CXMT and YMTC move from R&D and sampling to confirmed, qualified volume shipments, though, that relief stays theoretical rather than something buyers can price in.

5 predictions for the Chinese memory turf war

• YMTC’s LPDDR decision, expected by the end of 2026, is the nearer-term signal to watch — a green light would confirm China’s two biggest memory makers are both committing to cross-category expansion, not just testing the water.
• CXMT’s NAND effort will likely stay in R&D and trial production through at least 2027, given it has no prior commercial NAND product and the category gap (per the Korea Semiconductor Industry Association) is wider than YMTC’s head start in DRAM sampling.
• HBM, not NAND or general-purpose DRAM, will remain the most closely watched category, since CXMT’s reported HBM3E risk production intersects directly with AI accelerator supply chains where shortages are most acute.
• Samsung, SK Hynix, and Micron are unlikely to see meaningful share pressure from either Chinese move before 2028, given the multi-year gap between R&D/trial production and qualified volume shipments.
• The global memory shortage — flagged by SK Hynix’s CEO as potentially worst in 2027 — will likely remain the dominant price driver through 2027, with Chinese capacity additions from CXMT and YMTC acting as a slow release valve rather than a near-term fix.

What to watch next

The next concrete milestone is YMTC’s LPDDR customer feedback deadline at the end of 2026, which should clarify whether the company moves toward production or shelves the DRAM effort. On the CXMT side, watch for any public confirmation of a NAND roadmap — so far, the company has not officially confirmed the plans reported by Reuters, Tom’s Hardware, and TrendForce. Also worth tracking: whether CXMT’s HBM3E risk production progresses toward the reported 2027 mass-production target, since that’s the category with the most direct bearing on AI hardware supply chains, an area this site has also covered in the context of Nvidia B200 pricing premiums driven by GPU scarcity and the broader high-memory GPU comparisons shaped by the same supply constraints.

Frequently asked questions

What is CXMT?
ChangXin Memory Technologies is China’s leading domestic DRAM producer. According to Reuters and Tom’s Hardware reporting from September 2026, it’s reportedly building a NAND flash R&D line at a new facility near Beijing, though it hasn’t publicly confirmed a commercial NAND roadmap.

What is YMTC?
Yangtze Memory Technologies is China’s leading 3D NAND flash producer. TrendForce reported that YMTC has sent LPDDR (low-power DRAM) samples to customers and expects feedback by the end of 2026 before deciding on production.

Has CXMT confirmed it’s entering the NAND market?
No. The reporting, based on Reuters sources described as people familiar with the plans, has not been formally confirmed by CXMT itself. No commercial timeline, capacity, or product specification has been disclosed.

How far behind are Chinese memory makers compared to Samsung, SK Hynix, and Micron?
Per an assessment from the Korea Semiconductor Industry Association cited by Wccftech, China trails by roughly one year in NAND, about two years in DRAM, and around three years in HBM.

Why are CXMT and YMTC expanding into new memory categories now?
Reuters reporting attributes the timing to booming flash memory demand and a broader global memory shortage, which is giving Chinese suppliers an opening to widen their customer bases beyond their traditional single-category businesses.

Will this affect memory chip prices for consumers?
Not immediately. Neither company has disclosed pricing or output targets for its new category, and any capacity impact would take years to materialize given both efforts are still at the R&D or customer-sampling stage.

What is HBM and why does it matter to this story?
High-bandwidth memory is the memory format used in AI accelerators. CXMT has reportedly begun risk production of HBM3E, with mass production possibly arriving in 2027, according to Tom’s Hardware. HBM is currently the widest technology gap (about three years) for Chinese memory makers, and the tightest category in the current global shortage.

When will YMTC decide on DRAM production?
TrendForce reported that YMTC expects to gather customer feedback on its LPDDR samples by the end of 2026, which would inform any decision to move toward production.