What the original report claims
MakeUseOf says the memory squeeze is hitting consumers, OEMs and PC makers alike. It says Samsung and SK hynix are devoting more capacity to HBM for AI, which has tightened supply of DRAM used in smartphones and PCs. That gives CXMT room to grow.
The report's main points:
- It says CXMT expects to reach 375,000 DRAM wafers a month by the end of 2026, 420,000 in 2027 and 500,000 by the end of 2028.
- It credits the speed to cleanroom construction of about a year, against roughly 24 months for rivals. I could not independently confirm that comparison.
- It says CXMT holds 6.7% of DRAM wafer output but contributed 11.3% of supply growth.
- It cites former Samsung semiconductor chief Kyung-Hyeon Kye, via Wccftech, as saying CXMT could trigger a memory price slump in 2027 at the earliest.
- It concludes that 2027 is a turning point, with meaningful price drops more realistic in 2028.
The forecasts don't agree
The 2026 starting point varies by source. Most outlets in my search cite about 320,000 wafers a month today. Commercial Times, as reported by TrendForce, expects CXMT's monthly capacity to rise from around 320,000 wafers to 420,000 by 2027. That is a gap from MakeUseOf's 375,000 figure for year-end.
The 2028 figures vary too. Counterpoint Research and TrendForce forecast wafer output rising from about 320,000 this year to 420,000 in 2027, with a possible 500,000 to 600,000 by 2028. Treat all of these as forecasts, not production results.
Reuters reported on CXMT's expansion plans in August. CXMT shares soared almost 500% in their Shanghai debut on 27 July 2026. Reuters said the new plants in Shanghai and Hefei, plus further projects, could double capacity to more than 600,000 wafers per month when fully operational. That is a reported plan, not output.
Wafers are not saleable memory
Capacity is not the same as bits shipped. Wafer counts say little about yield, product mix or which customers get the chips. Several sources in my search put CXMT's share of global DRAM bit shipments at roughly 9% today. That is not far from its revenue share: TrendForce put CXMT's global DRAM revenue share at 9.5% in 2Q26.
Why PC builders may not feel it soon
The same TrendForce item covers Commercial Times' view that the extra output will mostly stay in China. It says:
- Additional output is expected to primarily serve China's domestic markets.
- The expansion is expected to reduce China's reliance on standard DRAM imports, but is unlikely to significantly increase global supply of high-end DRAM.
- China's expansion may not have a meaningful impact on global supply and demand until 2027 or later.
Pricing is the second problem. Reuters reported that CXMT raised the price of its 64 GB DDR5 server modules above Samsung's $1,240 benchmark and held firm, even when Huawei requested a discount. That is a server-module price, not retail RAM. It still shows a new supplier does not automatically undercut incumbents in a shortage. Digital Today reports that CXMT memory prices are known to be at a similar level to Samsung products.
MakeUseOf's own point holds: OEMs and big brands hold most of the bookings. Smaller module makers and retail get what is left.
HBM is still taking wafers
HBM competes for the same wafers as ordinary DRAM. Samsung's Kim Taewoo told Reuters that HBM is expected to account for nearly 30% of the DRAM industry's wafer capacity next year, up from about 20% currently. TrendForce's write-up adds that the wafer area of one HBM die is roughly what 3GB of DDR5 would need. That is a rough illustration, not a one-for-one conversion to consumer DIMMs.
MakeUseOf's allocation split, with mobile at 32.8%, GDDR at 14.0% and desktop RAM at 7.8%, is the original report's figures. I could not verify them or confirm the period they cover.
Other voices are less optimistic about 2027
- A Notebookcheck report says RAM is expected to become significantly more expensive again in 2027, before prices plummet the following year.
- Mezha reports that Michael Dell told The Wall Street Journal memory supply problems could continue for up to four years, until 2030.
- A FourWeekMBA summary of TechNews says analyst Charles Kau argued CXMT's capacity could leave the market in severe oversupply once fully online in two years. That is a bearish view on prices, and it points to 2028 more than 2027.
Export controls and equipment access are the other risks. They could slow CXMT's ramp and delay any price relief.
What this means for Windows buyers
- Don't expect a 2027 price cliff. The best-supported reading is that 2027 brings more supply, and 2028 is when prices could fall meaningfully.
- CXMT parts may show up in more products. More of it may appear in OEM laptops and prebuilt PCs than in retail kits.
- Prices depend on AI spending. If AI demand plateaus while fabs keep expanding, prices could fall sharply. MakeUseOf calls that the most optimistic scenario.
- The buy-now advice is MakeUseOf's opinion. It says to buy what you need now and not wait for 2027, citing Black Friday and Christmas deals. The production evidence doesn't prove that advice, and I haven't checked current retail prices.
Bottom line
CXMT is a genuine new force in conventional DRAM, and its capacity growth is well documented as a forecast. But much of the new output is expected to serve China's AI and cloud demand, and its pricing looks close to Samsung's. HBM keeps pulling wafers away from standard RAM, and Samsung, SK hynix and Micron still hold most of the market. Relief for PC buyers is plausible by 2028, not guaranteed in 2027.
References
- RAMageddon might end in 2027, and it’s all because of a company you’ve never heard of MakeUseOf · 2026-11-17T18:30:17+00:00
- CXMT’s RAM expansion through 2028 is unlikely to help most consumers mezha.net
- CXMT’s “budget” DDR5 RAM is now pricier than Samsung’s as AI demand keeps memory prices climbing - Notebookcheck News notebookcheck.net