CXMT DRAM Expansion May Not Lower DDR5 RAM Prices as Dell Warns of Worse Shortages in 2027

Here's what the numbers say, what they leave out, and what it means for anyone pricing a Windows 11 build or a fleet refresh.

The CXMT forecast in numbers​

The forecast first circulated through a post by P Equity Research on X, which summarized estimates from Counterpoint Research and TrendForce. Wccftech and TechRadar then picked it up. The trajectory looks like this:

TechRadar gives the figures without a time unit. Wccftech describes them as monthly wafer capacity. Counterpoint's own July 2026 note uses the same baseline, saying CXMT will grow from 320,000 to 420,000 wafers per month by 2027, using new fabs in Shanghai and Beijing plus a large new cluster in Hefei. The 2028 range has not been independently confirmed. TrendForce's public CXMT research index lists recent DRAM reports but doesn't show the 500,000–600,000 figure, so treat that number as a reported analyst forecast, not settled fact.

The reason enthusiasts noticed: Samsung, SK hynix and Micron have shifted heavily toward high-bandwidth memory (HBM) for AI accelerators, while CXMT's expansion targets conventional DRAM, the kind that goes into desktops and laptops. Wccftech reports that the added capacity is aimed at the general-purpose DRAM market, which mainly serves consumer electronics.

Section summary: The forecast points to big capacity growth in conventional DRAM, the PC-relevant kind. But these are wafer-start projections, and the 2028 number is the least certain of them.

Why capacity isn't the same as cheaper RAM​

A rising wafer count is easy to read as more RAM sticks on store shelves next year. That reading skips several steps.

1. Wafers aren't bits. Wafer capacity measures manufacturing input. How many usable gigabytes come out depends on process node, yields and how fast new lines ramp up. Counterpoint tracks a different measure, bit shipments. It puts CXMT at roughly 9% of global DRAM bits today and about 11% by 2028. That's growth, but not enough to rebalance the market.
2. Product mix matters. Counterpoint expects LPDDR5 and DDR5 to approach 75% of CXMT's output, and says PC- and server-grade parts are now reaching global OEMs. LPDDR5 mostly goes into phones and soldered laptop memory, not desktop DIMMs you can upgrade.
3. Where the memory goes matters most. TechRadar argues that China's own demand is large and that Beijing is likely to prioritize domestic supply over exports. That's analysis, not a published allocation policy. Nobody has released a domestic-versus-export split for CXMT's future output.
4. Geopolitics adds risk. Counterpoint lists a possible return of CXMT to U.S. Commerce Department scrutiny as a threat to both its equipment purchases and its non-Chinese customers. TrendForce noted in June that Entity List restrictions continue to hold back CXMT's advanced capabilities.

Counterpoint also says CXMT's entry into the PC supply chain began in mid-2025 but still has to turn into volume orders from non-Chinese OEMs. Put simply, Western PC makers are buying CXMT chips, but not yet in large volumes.

Section summary: More wafers could mean more memory overall. But bit output, product mix, export allocation and sanctions all stand between a fab announcement and a cheaper 32GB kit.

Not a discount brand, either​

Some people assume a challenger brand will undercut the incumbents on price. So far, CXMT hasn't. TechRadar says the cost of CXMT's memory is reportedly about the same as Samsung's. Wccftech goes further on the Apple angle: Apple reportedly asked for a discount, CXMT reportedly refused and quoted prices similar to or higher than Samsung's. Apple's reported interest is mainly about supplying its products in China.

This is basic economics. In a seller's market, every supplier charges what the market will bear, and new suppliers have no reason to start a price war. CXMT has also just raised RMB 57.9 billion (about $8.55 billion) in its July Shanghai STAR Market listing, according to Counterpoint, and it now has investors expecting returns.

Michael Dell: "gets worse in 2027"​

The second half of the story comes from Round Rock. In a Wall Street Journal interview published October 16, Michael Dell was asked when memory constraints might ease. Xenospectrum quotes his answer: "Based on what we can see right now, I think it gets worse in 2027."

His reasoning is about factory build times. New factories take about four years to build, Dell said, meaning plants planned today may not begin production until 2029 or 2030. That's where the "until 2030" headlines come from. It's worth reading carefully: Wccftech's account of the WSJ piece says memory supply constraints could persist for up to four years. That describes a possible duration, not a firm end date.

Context also matters. Dell was speaking as the head of a company with a huge AI server business. The company reported $16.4 billion in quarterly AI server revenue and a record $95 billion backlog in September. According to Digital Today's summary of the WSJ report, he said the main constraint is not demand but a lack of memory chips needed to deliver ordered AI servers on time. Xenospectrum notes that his comment does not mean that supply of semiconductors in general will deteriorate across the board. The forecast is about AI servers, not about consumer DIMMs specifically.

PC buyers should still care, because HBM and conventional DRAM come out of the same fabs. Every wafer that goes to HBM is one that can't become a laptop module. Other parts of the supply chain show the same strain: ASML said its planned chipmaking-equipment supply would fall short of customer demand in 2027 and 2028.

Section summary: Dell's warning is a forecast from someone who buys memory at huge scale. It backs up the gloomy outlook, but it's about AI server memory and isn't a precise consumer price forecast.

The counterargument​

The strongest optimistic case goes like this: any extra conventional DRAM, wherever it's sold, loosens the global market. If CXMT supplies more of China's domestic PC and phone demand, Chinese OEMs buy less from Samsung, SK hynix and Micron, and that frees some of their output for everyone else. Memory is a global commodity, and Chinese supply doesn't have to reach Western shelves directly to have an effect.

That's a fair point, and it's why the honest conclusion isn't "CXMT doesn't matter." It's that nobody can yet show the relief arriving quickly or in large enough volume. At about 11% of bits by 2028, CXMT would still be far smaller than the three incumbents, and if Dell is right, demand is growing faster than any one supplier can add capacity.

What this means for Windows users and IT admins​

From a practical point of view (this is analysis, not a forecast from any of the firms above):

• Don't plan purchases around 2028. Projected wafer capacity at a Chinese fab doesn't guarantee cheaper RAM. If a machine needs more memory to run Windows 11 well, plan on today's market, not an analyst's spreadsheet.
• Don't panic-buy either. Nothing here supports stockpiling or predicts specific prices. It only says relief is unlikely to come soon from this source.
• For fleets, buy the RAM you'll need up front. Soldered LPDDR5 in modern laptops can't be upgraded later, so the memory you order is the memory the machine will have for its whole life. Budget for 2027 refresh cycles assuming memory stays expensive.
• Expect CXMT parts in more products. As CXMT gets further into the PC supply chain, its chips may show up in more OEM systems and retail kits. Check compatibility and firmware support like you would for any memory vendor.

The bottom line: CXMT's expansion is real and could matter in the long run. Counterpoint's longer view has the company doubling capacity by 2030 and tripling it by 2035. But for anyone pricing a 64GB workstation this winter, the forecasts give no reason to wait. The AI build-out is using up DRAM faster than anyone can add capacity, and Michael Dell's message is that 2027 looks worse than 2026.

Have you already pushed back a PC build or fleet refresh because of RAM prices? Tell us in the forum.