The largest A-share IPO of 2026 is here. But more noteworthy than the listing itself is what is happening within the memory chip sector.
Let's Start with a Number: 29.5 Billion
On July 16, CXMT (Changxin Memory Technologies) officially launched its online and offline subscription on the STAR Market, with an issue price of RMB 8.66 per share, raising RMB 29.5 billion.
This is the largest A-share IPO in 2026 and the second-largest fundraising project in the history of the STAR Market, second only to SMIC (Semiconductor Manufacturing International Corporation)'s RMB 53.2 billion at the time.
But what you might not know is:
· From the acceptance of its IPO application to its approval, CXMT took only 148 days, setting a new record for the STAR Market.
· In Q1 2026, revenue reached RMB 50.8 billion, a year-on-year surge of 719%, with net profit attributable to shareholders reaching RMB 24.76 billion, turning a year-on-year loss into a profit.
· It is the fourth-largest global DRAM chip manufacturer (with a global market share of about 8% in Q1 2026), trailing only Samsung, SK Hynix, and Micron.
· It is the only IDM enterprise in China that has achieved large-scale mass production of DRAM.
What does this mean? The biggest shortcoming in China's memory chip sector is being made up.
But in today's article, I want to talk not only about CXMT as a company, but also about what is happening in the sector behind it—the memory chip industry.
Because the current market trend of memory chips might be even more noteworthy than the listing of CXMT itself.
I. What Exactly is the Memory Chip Business?
Let me first give a brief introduction for readers who are not in the memory sector.
The chip industry is roughly divided into two categories: logic chips and memory chips.
Logic chips are the "workers"—CPUs, GPUs, and AI chips, which are responsible for computing.
Memory chips are the "storage"—the photos on your phone, the files on your computer, and the data accessed during AI training are all stored here.
Memory chips are further divided into three main categories:
| Type | Full Name | Function | Market Leader |
| DRAM | Dynamic Random Access Memory | Memory modules; volatile, fast temporary storage | Samsung, SK Hynix, Micron |
| NAND Flash | Flash Memory | SSDs, mobile storage; non-volatile | Samsung, Kioxia, YMTC |
| HBM | High Bandwidth Memory | Dedicated for AI chips, packaged together with GPUs | SK Hynix (Dominant player) |
Memory chips have a distinct characteristic: they are the most "commodity-like" category in the chip industry.
What does this mean? Each logic chip manufacturer has its own architecture and ecosystem, resulting in strong differentiation. However, memory chips are relatively standardized, with prices fluctuating significantly based on supply and demand, making them highly cyclical.
They make a fortune when prices rise and suffer massive losses when prices fall.
And in 2026, the industry is in a phase of "skyrocketing prices."
— — — — — — — — — — — — — — — — — — — —
II. How Did the "Once-in-15-Years" Super Cycle Come About?
In the first half of 2026, the global memory chip industry entered a super boom cycle that multiple institutions have dubbed a "once-in-15-years" event.
Both DRAM and NAND prices have hit their highest levels since statistical records began in 2016. Taking an 8GB+256GB storage combination as an example, the Q1 2026 contract price surged by nearly 200% compared to the same period in 2025.
The reason is not simply a sudden explosion in demand, but a structural shift:
AI is "Devouring" Memory Capacity
This is the most core logic.
When AI chips (such as NVIDIA's H100/B200) train large models, they need to be paired with a chip called HBM (High Bandwidth Memory). HBM is essentially an advanced DRAM, stacked together with the GPU using advanced packaging technology.
The problem is:
1. HBM is extremely capacity-intensive—The wafer area required to produce HBM is 2 to 3 times that of standard DRAM. On the same production line, producing HBM yields less output.
2. HBM yields extremely high profits—Samsung, SK Hynix, and Micron are all shifting their capacity towards HBM.
3. The result—The capacity for standard DRAM and NAND is severely squeezed, leading to supply shortages.
To put it simply: the three major memory giants are all busy making HBM for AI, leaving insufficient capacity for standard memory and flash.
As a result:
· Consumer memory prices rise → The computers and phones you buy become more expensive.
· Enterprise storage prices rise → Data center costs soar.
· Module manufacturers (Longsys, Biwin, DMET) see profits surge → Because they hold low-cost inventory.
This is the core logic behind the memory chip price hike in 2026: it is not an explosion on the demand side, but the supply side being "siphoned" by AI.
III. CXMT: Why is it Worth 29.5 Billion?
Back to CXMT. Against this industry backdrop, CXMT's choice to go public at this moment could not be more precise.
Who Are They?
Headquartered in Hefei, Anhui, CXMT is the only domestic IDM enterprise that has achieved the integration of DRAM design, R&D, production, and sales. Its Chairman, Zhu Yiming, is also the Chairman of GigaDevice—the two companies have deep business synergy.
What is Their Technological Level?
| Dimension | CXMT's Current Status | Industry Benchmark (Samsung/SK Hynix) | Gap |
| Mainstream Mass Production Process | 17nm (G4, equivalent to 1z) | 11-12nm (1c node) | Lags by about 1-2 generations |
| Next-Generation Process | 15nm (G5), trial production in H2 2026 | Below 10nm already in mass production | Catching up |
| HBM | 16nm can only produce low-end HBM2 | HBM3e/HBM4 already in mass production | Lags by 1-2 generations |
| Architecture Innovation | Proprietary 4F² cell + CBA architecture | — | Density catches up with international mainstream |
Key information: CXMT uses a DUV multi-patterning solution and does not rely on EUV (Extreme Ultraviolet) Lithography machines. This means that against the backdrop of U.S. restrictions on EUV, CXMT's process route will not be choked off.
The generational gap indeed exists, but at the process nodes achievable by DUV, CXMT has already reached the limit of what can be caught up.
How is Their Profitability?
· Achieved profitability for the full year of 2025 (after consecutive annual losses in the tens of billions, with a loss of RMB 19.2 billion in 2023 and RMB 9 billion in 2024).
· In Q1 2026, revenue reached RMB 50.8 billion, a year-on-year growth of 719%, with net profit attributable to shareholders at RMB 24.76 billion—2.4 times the combined net profit of the three major A-share memory module manufacturers (Longsys, Biwin, and DMET) during the same period (approximately RMB 10.1 billion).
· The expected net profit for the first half of 2026 is RMB 66-75 billion, with revenue targeting RMB 110-120 billion.
From "losing tens of billions annually" to "earning 300 million daily," this reversal speed essentially capitalizes on the price hike dividends of the memory super cycle.
What Will the 29.5 Billion Raised Be Used For?
| Fundraising Project | Amount | Purpose |
| Technology upgrade for mass production lines of memory wafer manufacturing | About RMB 7.5 billion | Capacity expansion to capture price hike dividends |
| DRAM memory technology upgrade | Partial funding | 15nm process R&D |
| Forward-looking technology research | Partial funding | Layout of next-generation technologies such as HBM |
The logic is very clear: taking advantage of the industry boom and its own profitability, quickly raise funds to expand capacity and close the technological generational gap.
IV. More Important Than CXMT's Listing: The Formation of the "Dual Giants" Pattern in Domestic Memory
CXMT's listing is just the surface of the news; what is truly worthy of attention for industry practitioners is a larger trend:
The "Dual Giants" Pattern in Domestic Memory Has Taken Shape
| Company | Sector | Position | 2026 Developments |
| CXMT | DRAM | Fourth globally (about 8% market share in Q1), the only domestic player | Listed on the STAR Market in July, raising RMB 29.5 billion |
| YMTC | NAND Flash | Among the top globally, catching up with Samsung in 3D NAND stacking layers | Valued at over RMB 300 billion, IPO tutoring and filing have been initiated |
DRAM + NAND = the two main tracks of memory chips; China finally has one competitive player in each.
What does this mean?
4. Domestic substitution has basically completed the "0 to 1" stage—Memory chips were once the biggest shortcoming of China's semiconductor industry, with DRAM almost 100% reliant on imports. Now, CXMT and YMTC have filled this gap.
5. Entering the "1 to 10" catching-up stage—The generational gap remains (1-2 generations), but the route has been proven. What comes next is a competition of capacity expansion speed and technology iteration efficiency.
6. Supply chain driving effect—CXMT's listing is not just about one company; upstream and downstream equipment, materials, packaging and testing, and module manufacturers will all benefit.
Full Picture of the Memory Chip Supply Chain
| Level | Segment | Representative Companies |
| Upstream Materials | Silicon wafers, photoresist, specialty gases | Zing Semiconductor, Red Avenue New Materials, Yoke Technology |
| Manufacturing Equipment | Etching, thin film, cleaning, inspection | AMEC / NAURA, Piotech, ACM Research, Wuhan Jingce |
| Core Manufacturing | DRAM, NAND | CXMT (DRAM), YMTC (NAND) |
| Packaging and Testing | Packaging and testing | JCET, TFME, Hefei Xinfeng |
| Module Manufacturers | Memory modules | Longsys, Biwin Storage, DMET |
| Terminal Applications | Terminals | Mobile phones, PCs, servers, data centers |
V. Some Sobering Thoughts
Amid the hype, it is also necessary to mention a few less pleasant but crucial points:
1. The Generational Gap is Real
CXMT's mainstream process is 17nm, while competitors are already at 11-12nm. In HBM, CXMT can only produce low-end HBM2, whereas competitors have HBM3e/HBM4. A 1-2 generation gap in the memory industry means a dual disadvantage in both cost and performance.
Catching up requires time, money, and talent—RMB 29.5 billion can solve part of it, but not all.
2. Cycles Will Reverse
The biggest risk for memory chips is always the cycle. The current price hike is because AI is siphoning capacity, but if:
· AI demand growth slows down
· The three giants massively expand production of standard DRAM/NAND
· The HBM technology route undergoes changes
The price hike cycle could reverse at any time. How long can CXMT's current profitability of "earning 300 million daily" be sustained during a downturn?
3. HBM is the Real Hard Nut to Crack
CXMT has caught up to fourth place in DRAM, but HBM is a competition in another dimension. HBM requires advanced TSV (Through-Silicon Via) technology and 3D stacked packaging. This not only tests memory design but also advanced packaging capabilities—which happens to be another current shortcoming of China's semiconductor industry.
Final Thoughts
CXMT officially opened for subscription on July 16. This event itself is undoubtedly one of the most important IPOs in China's semiconductor industry in 2026.
But if you are a practitioner in the chip industry, what I want to say is: more worthy of attention than subscribing to the IPO is the structural change happening in the memory chip sector.
AI is reshaping the profit distribution across the entire semiconductor supply chain. Memory chips used to be "cyclical commodities," but now, because of HBM, they have become an irreplaceable link in the AI computing power chain.
CXMT and YMTC have caught this wave. Whether they can transition from "catching up" to "catching up and leveling the playing field" will determine the landscape of China's memory chip industry for the next 5-10 years.
In the short term, this is a hot topic for IPO subscriptions; in the long term, it is a watershed moment for domestic memory transitioning from "having it" to "being strong at it."
Paying attention to the memory chip sector is exactly the right thing to do now.
Interactive Topic: After CXMT goes public, how many more years do you think it will take for domestic DRAM to catch up with Samsung/SK Hynix? Welcome to discuss in the comments.