Chinese DRAM maker CXMT is expected to raise its monthly capacity from 320,000 to 420,000 wafers by 2027, according to a forecast Counterpoint Research published in July. The scale is hard to ignore when considering the memory shortage affecting PCs and smartphones. In September, CXMT also announced mass production of its fifth-generation process technology, "G5," and of LPDDR5X, which is used in smartphones and similar devices. Capacity expansion and process improvements are advancing in parallel, but the number of wafers a fab processes cannot be used to directly calculate how much memory will be available in stores.
When Was the 100,000-Wafer Increase Forecast?
Counterpoint's analysis, published in July, projects that CXMT's monthly capacity will grow from 320,000 wafers at the time of publication to 420,000 by 2027. It explains that new fabs in Shanghai and Beijing, along with the manufacturing base in Hefei, will drive the expansion. This is a research firm's outlook, and it should be distinguished from an announcement by CXMT that it has actually reached that capacity.
Counterpoint's July forecast of 320,000 to 420,000 wafers a month implies a difference of 100,000 wafers, or an increase of 31.25%.
| Point in time compared in Counterpoint's forecast | Monthly DRAM capacity |
|---|---|
| At publication in July 2026 | 320,000 wafers |
| Outlook through 2027 | 420,000 wafers |
The difference is calculated as 420,000 − 320,000, and the rate of increase as (420,000 ÷ 320,000 − 1) × 100. Because this compares the monthly capacity of a single company, it does not mean global DRAM supply will grow by 31.25%. The original does not specify the month in which the capacity will be reached, so it would also be premature to read it as a deadline, such as the expansion being finished within a year from now, in October.
A wafer is the disc-shaped material on which many semiconductor chips are made together. Monthly capacity indicates a fab's processing scale, but the memory capacity that can be shipped varies with the capacity of the chips made on each wafer and with yield. A forecast that the number of wafers processed per month will rise by roughly 30% may therefore lay the groundwork for greater supply, but it does not promise that the memory available to buy will grow by the same proportion.
The "50%+ Increase" for G5 Is a Different Metric
In its announcement of G5 mass production on September 20, CXMT said the total number of dies per wafer has increased by more than 50% compared with the previous generation. A die is the unit of chip built on a wafer. CXMT is trying to boost supply not only by feeding more wafers into the fab but also by making more chips from the same area.
However, the "50%+ increase" comes with comparison conditions. According to the announcement's footnote, the baseline is the previous-generation G4, and total die counts are normalized by converting to 8Gb capacity. It is not an announcement that the number of chips shipped as good products has risen by more than 50%.
The manufacturing process uses technologies such as SAQP, which forms fine patterns. Finer patterning allows higher circuit density within the same wafer area. Whether the circuits built meet the required quality is a separate condition, though, and treating total die count as equal to the number of good products could substantially overestimate actual supply capacity.
The G5 announcement also showed that two types of 24Gb-per-die LPDDR5X are in mass production for smartphones and handheld consumer devices. The 24Gb figure is the capacity of a single die and differs from the total memory capacity installed in a device. The significance of the announcement lies in making clearer what the new process is actually producing.
So can CXMT's overall supply increase be calculated by multiplying the 31.25% capacity increase by the 50%+ rise in G5 total die counts? Answering that would require the share of total output moving to G5, plus product mix and yield. The two published figures alone do not meet those conditions. Capacity expansion and process miniaturization are separate inputs in estimating shipment volume.
General-Purpose DRAM Is Not Consumer-Only
Counterpoint expects LPDDR5 and DDR5 to approach about 75% of CXMT's production mix, while also mentioning both PC and server applications. Because DDR5 is used in both PCs and servers, an increase in general-purpose DRAM output cannot be taken directly as an increase in consumer memory. Even with a track record of mass-producing LPDDR5X for smartphones, the added capacity across the whole factory will not necessarily go to the same uses.
Allocation across customers is an issue for the market as a whole as well. According to a survey TrendForce published on September 7, manufacturers' inventories are at historically low levels, and the additional supply has mainly been directed to server applications. The firm also notes that supply of consumer DRAM has been sharply constrained. Even as memory output rises, the sense of shortage in each application does not necessarily ease in the same way.
Moreover, the memory servers use is not limited to HBM, the stacked memory used for AI. Ordinary DRAM such as DDR5 is also in demand, so a dichotomy of "the big players make HBM and CXMT makes consumer products" fails to capture the competition for supply. How much CXMT's expansion helps PCs and smartphones is likely to depend on the scale of the facilities once completed and on how their output is allocated.
TrendForce further expects that the increase in bit supply from the three major makers in 2026–2027 will come mainly from shifts to advanced manufacturing processes, with only a modest rise in wafer input. CXMT is also improving per-wafer output with G5, which makes it harder to rank companies' supply capacity, or to judge when shortages will ease, from wafer-processing capacity alone.
What to Check Before Prices: The Substance of Supply
On transactions with non-Chinese equipment makers, Counterpoint cited as a point to watch whether CXMT can turn its entry into the PC supply chain into large-scale orders. There are stages between being able to make a product and delivering the necessary volumes to a broad customer base. To judge whether consumer shortages will ease, after company-wide monthly capacity we should confirm which standards and capacities of products will ship in volume to which customers.
For example, even if new LPDDR products for smartphones increase, that alone will not fill the shortage of DDR5 modules for PCs. Conversely, even within DDR5, a larger allocation to servers could mean the volume available to PC makers grows less than total production does. The change to watch closest to consumers is not the total number of wafers but a widening of where, and in what quantities, the products they want are supplied.
CXMT's capacity-expansion forecast and G5 mass production are factors that could support that widening. But neither provides information from which the size or timing of any drop in retail prices can be calculated. When process improvements translate into stable production of good chips, and adoption and shipments for PCs and smartphones grow, the 100,000-wafer-a-month capacity figure will show up for buyers as a broader range of sourcing options.