May 2025
Bit output from the two indigenous Chinese memory manufacturers is an important factor for what happens to the supply/demand balance in the memory markets today and in the coming years. Here are notes gathered from what is in the news about CXMT and YMTC as of late May of 2025.
ChangXin Memory Technology (CXMT):
- · CXMT will phase out DDR4 memory for servers and PCs by the middle of 2026, at the direction of Beijing. Had begun mass production of DDR4 in late 2024. Priority will be shifted to DDR5. Also, CXMT is believed to be developing an HBM product. (Tom’s Hardware)
- · CXMT has stopped DDR4 production and plans EOL by end of Q1 2025. This article was published in Q2 2025 so that comment brings the veracity of its information into question. CXMT is “said to” plan to produce 280,000 wafers per month (70,000 wafers per week) by late 2025. DDR5 products are not mature and have compromised yields. Some domestic memory manufacturers have switched back to sources from South Korea. The company is said to have HBM2 capabilities and to be developing HBM3 now. (WCFF Tech)
- · Phasing out DDR4 by mid-2026, leading to prices for 8Gb DDR4 chips doubling to $2.50. Will this be a lifeline to Nanya? CXMT is projecting to reach 280,000 wafers per month by late 2025, with upside potential to 300,000. This may be as much as 15% of global DRAM output. That may be measured by wafers, so would be much less in bits, because process technology is older. By the end of 2025, DDR5 is expected to make up more than 60% of CXMT’s output, alongside LPDDR4 and LPDDR5. Taiwanese firms said CXMT’s first quarter 2025 DDR5 samples failed key tests, specifically high and low temperatures. (DigiTimes Asia)
- · CXMT will dramatically expand DRAM output in 2025, “far surpassing previous expectations.” South Korean news outlets, citing Omdia research (~500 employees,) project CXMT’s output in 2025 to hit 2.73 million wafers (52,500 per week on average). This is up from 1.62 million wafers (31,150 per week on average) in 2024. That is a 67% increase year-over-year. Assuming the ramp rate is similar across this two year period, the output at the end of 2025 would be considerably more than 52K per week. This is an increase from the prior projection of 20% year-on-year growth. CXMT produced 100,000 wafers per month in early 2024, about one-quarter of Hynix’s output, in early 2024, primarily DDR4. By the first quarter of 2025, output had doubled to 200,000 wafers, with a forecast of 300,000 wafers by 2026. CXMT had 5% market share in 2023. This is expected to be ~10% by late 2025. Research from Qianzhan Industrial Research Institute (Chinese company with ~500 employees) says CXMT had no DRAM market share in 2020, then 5% in 2023. TrendForce projects this “could climb” to 12% by the end of 2025. (DigiTimes, April 21, 2025)
- · Current CXMT chips in production are on 16-nanometer process technology. According to Tech Insights, this technology is about three years behind the leading edge. (Bloomberg, January 28, 2025)
- · CXMT released a 16Gb DDR5 part. I presume this is an engineering sample, rather than volume manufacturing. The cell size is 0.0020 µm2. This is equivalent to the 1z DRAM tech node. This 16 nm process is below the export restriction on technology below 18 nm. Not a surprise, given how hard it is to enforce such a rule, and the innovation that affected entities always show under such constraints. (Tech Insights, January 24, 2025)
- · CXMT wafer output grew from 120,000 WOPM at the end of 2023 to 160,000 WOPM in the first quarter of 2024. They will reach 200,000 WOPM at the end of 2024, according to Nomura Securities. Their production technology, however, is two to four years behind the leaders. (SemiWiki, September 20, 2024).
- · CXMT is 3-4 years behind global leaders in HBM. They are aiming to produce HBM3 in 2026 and HBM3E in 2027. The company began mass production of HBM2 devices in the second half of 2024. CXMT stockpiled enough semiconductor manufacturing equipment for HBM and DRAM production that is likely sufficient through 2026 or 2027. (ChinaTalk, April 17, 2025)
- · Total industry DRAM bit supply is expected to grow 18% in 2025 over 2024. This would be 14% excluding China. 2024 supply growth was 16%, or 13% excluding China. Within this 2025 growth, the top three players, Samsung, Hynix, and Micron, are expected to increase 11%, 17%, and 14% year-over-year, respectively. CXMT’s 2025 wafer output in 2025 will be within 10% of Micron’s, granted on much older process technology. (TrendForce, August 2024)
- · CXMT has two fabs in Hefei. Fab 1 is DDR4 with 90% yield and 100,000 WOPM capacity. Fab 2 is DDR5 on 17nm technology, making 50,000 WOPM. (Random dude on LI, January 2025*)
- · According to Tech Insights and TrendForce, YMTC’s share of the global NAND market is 13%. These sources report CXMT’s share of the DRAM market to be 9%. (Maeil Business Newspaper, October 16, 2024)
CXMT Summary
By one estimate, the company will grow DRAM bit output by ~65% in 2025 over 2024. In late 2025, some estimates have their DRAM market share in the range of 10 – 12%, up from 5% in 2023 to 2024. Around the end of 2024, their total output was 200,000 WOPM. This may increase to 280,000 to 300,000 WOPM by the end of 2025, though estimates vary across a substantial range. The company has stockpiled enough fab equipment to execute their expansion plans through 2026 or 2027. The company released engineering sample of their first 16Gb DDR5 part in early 2025. The process technology is 1z. For comparison, Micron is the industry leader in process technology. Micron released engineering samples for their 1-gamma-based DDR5 product in February of 2025. So similar time as the 1-z process node from CXMT, putting CXMT three nodes behind the leader. This equates to three to four years in time. One analyst estimates total industry bit supply will grow 18% in 2025, year-over-year. CXMT is today producing HBM2 devices for the Chinese market, which began mass production in the second half of 2024. They are transitioning as fast as they can from DDR4 to DDR5 products.
Yangtzi River Memory Technology (YMTC):
- · YMTC lost $11.6M in the first nine months of 2024. The top seven shareholders of the company are all state-backed. The PRC’s Big Fund also holds about one-quarter of the equity, via two subsidiaries. The company spent 16.7% of revenue on R&D in 2023. Both of these numbers – the reported loss in the first three quarters of 2024 and the R&D spending – are too low, in my view. I treat numbers, including financial numbers, out of China with suspicion. Perhaps YMTC has a better cost structure than I think. Maybe their R&D spending is accounted for differently than conventional accounting rules would tally them. Revenue could be inflated by the PRC forcing buyers to pay an artificially high ASP, shifting losses off of YMTC and onto PC and mobile phone makers. (TrendForce, April 29, 2025)
- · The company has started shipping its 5th generation 3D NAND, which has 294 total layers and 232 active layers. The difference between active layers and total layers is redundancy plus some dummies. The bit density of this device is similar to the major NAND competitors. This device has been shipped without fanfare, likely to avoid drawing attention from the U.S. government, and possibly attracting further sanctions. (Tom’s Hardware, January 30, 2025)
- · A recent funding round valued YMTC at $22.5B. Some financials were disclosed as part of the deal. The company made a profit of $73.8M in 2023, followed by a loss of $11.7M in the first nine months of 2024. YMTC was originally a subsidiary of Tsinghua Unigroup and was separated after the latter was restructured in 2022. YMTC is on the U.S. entities list. The company’s performance is part of a larger enterprise. Its results are suspected to be a financial drag. The fact that an equity raise was done recently supports this idea. (DigiTimes Asia, April 29, 2025)
- · From an analyst call with Sumco, while the majority of the silicon wafers produced in China are test wafers, YMTC is thought to be consuming 400,000 to 500,000 domestically-produced wafers. Since YMTC bonds two wafers together to produce a single product wafer, that means their output may be 250,000 WOPM. The quality of these Chinese prime wafers is reportedly not very good, but the State is forcing YMTC and other domestic fab operators to purchase local silicon. (Tom’s Hardware, November 26, 2024)
- · According to Tech Insights and TrendForce, YMTC’s share of the global NAND market is 13%. These sources report CXMT’s share of the DRAM market to be 9%. (Maeil Business Newspaper, October 16, 2024)
YMTC Summary
Surprisingly, less information is available today about YMTC than CXMT. YMTC’s technology is closer to the leading edge, perhaps within a year or two. Their market share is also larger, in the low-double digits. The company reportedly made money in 2023 but then generated losses in the first nine months of 2024. This is suspect, as the NAND market was stronger in 2024 than 2023. Also, YMTC was smaller in 2023 than 2024. Scale isn’t all that matters in semiconductor manufacturing, but it matters a lot. YMTC did an equity raise in early 2025. This strongly suggests they are struggling financially; else why would they need more capital? The NAND market is in much worse shape than the DRAM market, which may be part of why expansion plans for YMTC are hard to come by at this time. The company led the industry in processing NAND on two wafers, which they call XStacking. I think they did this out of necessity, being unable to make the device function on single-wafer processing. Now the rest of the industry is following, because scaling has forced them to do so. I think YMTC’s existence and progress will continue to keep the NAND industry from reaching a healthy state. There are now two too many participants.
– S. Hughes (short MU)