Chinese DRAM maker ChangXin Memory Technologies (CXMT) reached a 10% share of global DRAM revenue in the second quarter of 2026. According to quarterly data published by Counterpoint Research on September 1, its share rose from 4% a year earlier and 8% in the previous quarter, making it an increasingly visible supplier behind the three leading companies.
That 10%, however, is not a share of the memory capacity shipped. In a market where commodity DRAM prices are rising while HBM prices are moving differently, product mix and contract timing also shift each company's slice of revenue. To judge how much more CXMT can supply and what options that gives buyers, the share trend has to be read alongside the products it actually supplies.
How the Big Three's share changed at the 10% mark
In Q2 2026, DRAM revenue shares were Samsung at 38%, SK hynix at 25%, Micron at 24%, and CXMT at 10%. Samsung rose from 33% a year earlier, and Micron from 22% to 24%. The decline was concentrated in SK hynix, which fell from 39% to 25%. The latest quarterly table lays out this shift on a single consistent edition of the data.
The combined share of the top three fell from 94% to 87%, a drop of 7 points. A year earlier it was 33+39+22=94%; in Q2 2026 it was 38+25+24=87%, so 87−94=−7 points. Not every company lost ground: Samsung and Micron both finished above their year-earlier levels. What widened was the share held by others, including CXMT and Nanya.
- Samsung
- SK hynix
- Micron
- CXMT
データを表で見る
| Samsung (%) | SK hynix (%) | Micron (%) | CXMT (%) | |
|---|---|---|---|---|
| 2025 Q2 | 33 | 39 | 22 | 4 |
| 2025 Q3 | 33 | 33 | 26 | 5 |
| 2025 Q4 | 36 | 32 | 22 | 8 |
| 2026 Q1 | 38 | 29 | 22 | 8 |
| 2026 Q2 | 38 | 25 | 24 | 10 |
The chart makes CXMT's rise and SK hynix's decline stand out, but the data does not show existing customers moving directly from one company to the other. Revenue share is each company's revenue divided by total market revenue, so it is affected by sales volume, unit prices, and product mix. The move from 4% to 10% means the share grew 2.5 times, not that CXMT's revenue grew 2.5 times. Displayed values are rounded, so totals may not equal exactly 100%.
Note: Counterpoint's August analysis put Samsung at 39%, SK hynix at 26%, and Micron at 25%, which differs from the September table. Share comparisons in this article are unified on the September 1 edition.
Why the HBM leader lost revenue share
SK hynix still leads in HBM. In Q2 2026, HBM revenue shares were SK hynix at 50%, Samsung at 33%, and Micron at 18%; the displayed total of 101% reflects rounding. Counterpoint Research's HBM table has no row for CXMT. But the absence of a row does not prove that CXMT has zero HBM revenue.
データを表で見る
| HBM売上高シェア (%) | |
|---|---|
| SK hynix | 50 |
| Samsung | 33 |
| Micron | 18 |
SK hynix holds half of HBM, yet Samsung leads in DRAM overall. Rankings based on revenue within the HBM market and rankings across the whole DRAM market diverge in just this way.
In its August analysis, Counterpoint explained that commodity DRAM prices rose quarter on quarter in the period, while HBM prices fell year on year. It said falling HBM3E prices and delays in introducing HBM4 in particular weighed on average prices. The comparison periods differ, but the analysis is that revenue grew differently for SK hynix, which depends heavily on HBM, and for companies able to capture the rise in commodity DRAM prices.
Contract timing also matters. According to Counterpoint, SK hynix signed long-term supply contracts earlier than its rivals, and the price terms in those contracts limited its revenue gains as market prices rose. Securing supply early is valuable to customers, but it does not mean a manufacturer can reflect every later price rise in its revenue. Viewing the share decline of the HBM leader solely as a loss in manufacturing-technology competition would overlook these pricing and contract effects.
AI demand reaches commodity DRAM too
In its June 2024 earnings materials, Micron showed that, for the same process generation and the same number of bits, HBM3E requires about three times the wafer input of DDR5. Because producing the same capacity uses more production resources, increasing HBM constrains the resources that can go to ordinary DRAM. The roughly threefold comparison describes HBM3E versus DDR5 at that time and is not a figure for yields at CXMT's fabs.
In June, TrendForce reported that commodity DRAM contract prices rose 93–98% quarter on quarter in Q1 2026. It also projected that in Q2, priority for high-capacity server RDIMMs would constrain supply for PCs and smartphones. Adding cleanroom space takes time, it said, and supply growth in 2026 will rest mainly on process migrations, with only limited increases in wafer input.
AI demand does not stop at the HBM attached to GPUs. CPUs also need commodity DRAM, and the more server products are expanded, the harder it becomes to allocate production between them and PC or smartphone products. Stacked HBM and DDR5 modules are not interchangeable products, and a shortage cannot be judged resolved by total supply alone.
Counterpoint cites domestic Chinese demand for commodity DRAM and expanded production capacity as drivers of CXMT's growth. For buyers, the pressing question is whether they can procure products in the standards they need. Even without leading in cutting-edge HBM, a company that can supply commodity DRAM reliably has a chance to capture demand that AI investment has broadened.
Wider supply sources in DDR5 and LPDDR5X
In October 2025, CXMT announced that it had begun mass production of LPDDR5X at 8533Mbps and 9600Mbps in May of that year. The 10667Mbps version was at the stage of customer sample shipments at the time of the announcement.
| Product / spec | Confirmed status | Date confirmed | Notes |
|---|---|---|---|
| LPDDR5X 8533Mbps / 9600Mbps | Mass production began May 2025 | Announced Oct 28, 2025 | Die capacities of 12Gb / 16Gb |
| LPDDR5X 10667Mbps | Customer sample shipments | Announced Oct 28, 2025 | Cannot conclude it is still at the sample stage |
| DDR5 up to 8000Mbps, die capacity up to 24Gb | Products exhibited | Nov 23, 2025 | UDIMM, SODIMM, RDIMM, MRDIMM and others displayed |
| LPDDR5X up to 10667Mbps | Products exhibited | Nov 23, 2025 | An exhibit does not mean all products are qualified for mass production |
The speed tiers confirmed in mass production are not the same as the top speeds shown at exhibitions. Still, showing both PC and server modules in DDR5 is evidence that CXMT is aiming at a widening range of uses. The maximum DDR5 die capacity of 24Gb likewise does not indicate the capacity of a whole module.
There are also clues on the retail side. TechInsights analyzed Gloway's 16GB×2 DDR5-6000 UDIMM and reported that it confirmed CXMT G4. Analysis that identifies chips in a product actually on sale supports the existence of products supplied to the PC market. But confirming that one module does not allow an assessment of the speed or manufacturing cost of all CXMT products.
As supply records build up product by product, buyers can add sources according to use. Conversely, even if more top-speed products are exhibited, procurement constraints will not ease unless buyers can get the needed capacity when they need it. To judge whether share gains will continue, the breadth of product types and continuity of supply must be checked together.
What shapes price declines is supply by product
CXMT's 10% revenue share points toward more options for buyers. But it is not possible to conclude from the number that retail prices in Japan will fall immediately. To see the effect on purchase prices, one needs to track shipment volumes by standard, demand from servers and consumer devices, and production allocation to HBM. Whether a rise in revenue share comes from higher prices or from more products buyers can actually get changes what it means for them.
| Metric | What it shows | What it doesn't directly show |
|---|---|---|
| Revenue share | Each company's slice of market revenue | Bit shipments, absolute revenue, supply capacity |
| Bit shipments | Total memory capacity shipped | Revenue composition including unit prices |
| Wafer input | Scale of production resources devoted to manufacturing | Finished-product volume, yields, customers |
| Mass production / sampling / exhibition | How far a product can be confirmed | Customer qualification, continuous supply, profitability |
Tracking revenue share together with bit shipments makes it easier to separate changes in prices from changes in supply volume. In evaluating any push into the HBM market, too, one must confirm mass production and customer adoption separately from the 10% of all DRAM.
Which capacities and speeds of DDR5 and LPDDR5X see shipments grow, and how shortages in server and consumer segments change, are the questions to watch. If sustained supply increases are confirmed there, CXMT's growth could give buyers more sources and more leverage in comparing supply terms. For PC users in Japan, too, the next thing to watch is less the size of the share figure than whether products in the standards they need reach store shelves reliably.