On October 20-22, along the banks of Shanghai's Huangpu River, the 2026 Sustainable Global Leaders Conference returns once again. Nobel and Turing Award winners will take the stage to discuss the future and trends of technology, and the internet-famous professors who "rescue" people from anxiety and involution—Macfarlane and Xiang Biao—will hold a live dialogue! Seats are limited, so don't delay. Let us join together for a sustainable appointment! [Click to register] Source: Yingcai Business.
Global memory giants have recently been losing sleep over Cxmt Corporation (688825.SH). Earlier, global technology sectors underwent a sharp correction, and Cxmt Corporation (688825.SH) even lost its position as the top A-share company by market capitalization after closing yesterday. The next day, Cxmt Corporation (688825.SH) rebounded sharply, with its market value returning to 3 trillion yuan. The drivers of this global tech pullback are complex, but if A-shares are to rebound, the leader will inevitably still be technology. Not long ago, Cxmt Corporation (688825.SH) announced key progress in the R&D of its 4F² DRAM architecture, and it expects to launch a new generation of DDR5 RDIMM server memory that integrates the 4F² architecture with hybrid bonding technology by the end of 2026. After the news landed, global memory assets were collectively repriced, compounded by global capital's fear of Federal Reserve rate hikes. On October 13, Samsung Electronics Co., Ltd. (SSNLF) closed down 7.39%, SK Hynix closed down 10.09%, and the KOSPI fell 5.46% in a single day, triggering the Sidecar mechanism intraday. Foreign research reports quickly followed, with BofA defining 4F² as "the most important DRAM architecture turning point since the introduction of EUV." This technology has not yet been mass-produced—so why has it triggered an overnight repricing of global memory? If you put Cxmt Corporation (688825.SH) and Samsung Electronics Co., Ltd. (SSNLF) on the same exam paper, you will find that what may be rewritten is the time gap in mass production.
01 What Is 4F²: Taking a Different Path to Bypass Lithography
From a technical perspective, the current mainstream DRAM cell structure is 6F², where the cell area equals six times the square of the minimum feature size; 4F² compresses it to four times. As the figure below shows, 8F²→6F²→4F² reflects the ongoing path of DRAM cell area compression. Multiple data sources consistently indicate that 4F² can reduce chip area by approximately 30%, or increase the net die count by approximately 30%; in some formulations, the theoretical density improvement can reach 30%-50%. The more robust institutional view is that the core benefit and change of 4F² is a reduction in cell area, an increase in net die count of about 30%, and potentially higher theoretical density improvements under specific implementation conditions. Cxmt Corporation (688825.SH) has three breakthroughs. First, a self-developed VSG vertical surrounding gate transistor, using a vertical channel to enhance gate control, suppress leakage, and mitigate short-channel effects; second, integrated hybrid bonding, allowing peripheral circuits and memory cells to be optimized in separate layers, paving the way for 3D stacking; third, IP leadership, with more than 300 published patent families, the first paper released in 2021, and continuous disclosure of progress in 2023 and 2025. The weight of this combination must be viewed within the industry's coordinates. DRAM planar scaling is approaching its physical limits: if capacitors continue to shrink, they will be unable to hold charge, and leakage, process variation, and cost are squeezing from three sides. The industry consensus identifies three paths forward: 4F² vertical transistors, peripheral circuit underlay with wafer bonding, and ultimately 3D DRAM. The role of 4F² is as a bridge to 3D DRAM. Whoever gets across this bridge first holds the ticket to the next paradigm shift. Cxmt Corporation (688825.SH) is essentially shifting the focus of DRAM innovation from "relying purely on lithographic scaling" gradually toward device structure, materials engineering, wafer stacking, interface control, and system integration—which is why foreign equipment-side analysts are paying close attention to 4F². For the same 4F², the two sides' commitment is naturally asymmetric. For Cxmt Corporation (688825.SH), it is a mandatory question: because it does not have EUV, and must rely on DUV plus multi-patterning to continue grinding down line widths, with more masks, longer steps, and higher defect risk, architectural innovation is the only way out; while for Samsung Electronics Co., Ltd. (SSNLF), 4F² is only an optional question: EUV scaling is still in hand, and 4F² is just one of multiple options. Those facing a mandatory question go all in, while those facing an optional question run multiple tracks in parallel. Historically, the technology choices of big players were an advantage; at a turning point, they can instead become an action drag—people with too many fallback options cannot run with a decisive slope. Only by understanding this asymmetry can one see why foreign capital is nervous: Chinese manufacturers are starting to change lanes, no longer chasing grid by grid along the track drawn by Korean manufacturers. In addition, one must also see the global backdrop of this lane change. Over the past two years, AI has transformed memory from a cyclical commodity into half a growth stock: HBM consumes advanced capacity, general-purpose DRAM supply is passively tightened, and the three global memory leaders enjoy rising volumes and prices. In such a boom, any potential new source of supply will be magnified and scrutinized by the market. Cxmt Corporation (688825.SH)'s 4F² happens to land at this sensitive moment, and the imaginative space it brings has two layers: one is the possibility of pushing down per-bit costs, and the other is the possibility of securing an early position in the 3D DRAM era. The former shakes profit margins, while the latter shakes the competitive landscape.
02 Three Layers of Impact: Narrative, Positioning, Discipline
The first layer is the technology narrative being pried open. In the past, the market's pricing logic for Chinese DRAM was very stable: mature categories could be given share, while advanced architectures need not be considered. The 4F² news has torn a gap in this underlying assumption. The second layer is the attack positioning, which is shrewdly chosen. To hit the snake, strike its vital point; Cxmt Corporation (688825.SH) set its debut scenario on DDR5 RDIMM server memory. Server memory has high unit prices and thick profits, and its barriers are built on long-term validation, compatibility, and supply commitments, deeply coupled with CPU platforms. In the AI era, server DRAM is already the most marginally scarce sub-market: SO-CAMM2 demand driven by the Vera Rubin platform is expected to grow from about 21 billion Gb in 2026 to about 83 billion Gb in 2028, with its share of total DRAM demand rising from 6% to 13%. The third layer is a crack in supply discipline. Over the past decade, DRAM's high profitability has been built on three oligopolists plus restrained capital expenditure. But Cxmt Corporation (688825.SH)'s share has risen from about 4% a year ago to 9.5% in the second quarter of 2026, and was 7.67% by sales in 2025Q4, firmly ranking fourth globally. Morgan Stanley's framework is very straightforward: affecting pricing does not require technological parity; new output only needs to be large enough to move marginal prices. What is more troublesome is the asymmetry of objective functions. Samsung Electronics Co., Ltd. (SSNLF), SK Hynix, and Micron are all profit-oriented listed companies that will cut production to defend prices in downturns, and this self-restraint has been the invisible cornerstone of high profitability over the past decade; Cxmt Corporation (688825.SH)'s objective function also includes domestic substitution, the learning curve, and supply chain resilience, carrying an inherent mission-driven quality, so it may not follow the contraction in the next downturn. At that point, the three overseas players will have to bear heavier production-cut obligations, and the industry's profit center will be systematically pushed down. What this round of adjustment may be previewing is the script for after 2028, if memory returns to its cyclical nature. However, the time scale is key. HSBC and Citi are aligned: over the next 2-3 years, 4F² will have limited impact on global DRAM prices, as JEDEC standards, platform validation, and mass-production yield are all still lacking, and large-scale mass production is generally no earlier than 2028. The industry's fundamentals remain: DRAM will still be tight in 2026-2027, HBM continues to squeeze out general-purpose capacity, and SO-CAMM2 and server DDR5 demand continue to expand. What is truly being traded ahead of time is sentiment and valuation. Morgan Stanley's chain of reasoning is: more participants, weakened supply discipline, lower cross-cycle profit margins, with DRAM peak-cycle operating margins able to exceed 80%, while the long-term center may return to about 50%; the roughly 50% ROE in 2026 may have a sustainable level of only 10%-15%. This chain pressures the valuation center, not necessarily the cycle top. In other words, the market is moving the idea that "CXMT changes the long-term ROE of the memory sector" from an extreme scenario into the base scenario. Once this shift is complete, even if not a single additional chip is sold before 2028, the terminal multiples of memory leaders will first be pushed down a notch—which is also why Cxmt Corporation (688825.SH) also corrected sharply along with global memory. Stock prices always run ahead of profits; this is common in asset pricing, only this time it moved fast, completing in a few days a journey that usually takes a month or even a quarter. A technology change is read by manufacturers as a threat and by equipment makers as demand. 4F² does not reduce manufacturing difficulty; it merely moves the difficulty elsewhere: DRAM manufacturing shifts from a lithography center to a materials engineering center, with intensity rising across ALD deposition, epitaxy, and hybrid bonding, and BofA specifically pointed to the equipment side. Combined with Cxmt Corporation (688825.SH)'s capacity expansion slope from about 300,000 wafers per month in 2026 to an expected 800,000 wafers per month in 2031, the more than doubling of capacity implies full-line orders, which may bring short-term opportunities to domestic semiconductor equipment.
03 CXMT vs Samsung: The First Same-Generation Competition in History
In fact, Samsung Electronics Co., Ltd. (SSNLF) also has 4F² technology. It began research about five years ago, showcased a 16Gb 4F² DRAM prototype at ISSCC in 2026, using VCT vertical channel transistors plus wafer-level hybrid bonding, with mass production generally pointing to 2028. Cxmt Corporation (688825.SH)'s mass-production pace is roughly around the same period, around 2028. This timetable is the weight behind this repricing: for the first time in the memory industry, a pursuer and the leader are taking the same exam on the same not-yet-mass-produced architecture. In the past, Cxmt Corporation (688825.SH) chased old technologies that Samsung Electronics Co., Ltd. (SSNLF) had mass-produced for years, and Samsung Electronics Co., Ltd. (SSNLF) secured premiums by staying two or three generations ahead; now, the time gap may be reduced to zero at the 4F² node. In terms of market structure: Samsung Electronics Co., Ltd. (SSNLF) is the largest supplier of general-purpose DRAM, while Cxmt Corporation (688825.SH) is the main substitution source in the Chinese market, and the two are directly facing each other in server and general-purpose chips. Let's put the two companies' foundations side by side. Samsung Electronics Co., Ltd. (SSNLF): 39.4% global DRAM share in 2Q26, 33.96% by 2025 sales, expected capacity of about 700,000 wafers per month by the end of 2026, advanced 12-13nm-class nodes already in mass production, EUV in hand, and in addition to 4F² it is simultaneously advancing multiple routes including B1b and wafer-level bonding; foreign investors expect its HBM bit share to reach 41% in 2027. Cxmt Corporation (688825.SH): fourth globally, with a 9.5% share in the second quarter of 2026, expected capacity of about 300,000 wafers per month in 2026, accounting for about 13% of global wafer capacity and about 11% of bit shipments, with main processes at the 16-17nm level, no EUV, and already in the supply chains of Xiaomi, OPPO, vivo, Transsion, and Lenovo. Samsung Electronics Co., Ltd. (SSNLF)'s moat is mass-production experience, customer certification systems, and the redundancy of running multiple routes in parallel; its soft spot is that its business is too complex, having to manage HBM, general-purpose DRAM, NAND, and foundry at the same time, which inevitably dilutes marginal resource efficiency. It has three most sensitive nerves: pricing power in general-purpose DRAM—every bit replaced by Cxmt Corporation (688825.SH) thins its price leadership; narrative power in advanced architecture—leader premiums depend on leadership being delivered, and the 4F² first-mover narrative, even if not yet mass-produced, will force it from definer to clarifier; and the complexity of resource allocation—four businesses must be answered simultaneously, and any one falling half a step behind will be priced separately by the market. Cxmt Corporation (688825.SH)'s advantages lie in its home market, IDM integration, and expansion speed, with capacity expected to go from about 200,000 wafers per month in 2025 to 800,000 wafers per month in 2031; its shortcomings lie in the cost structure of advanced nodes, server certification cycles, and the HBM ecosystem gap. General-purpose DRAM happens to be the segment where Samsung Electronics Co., Ltd. (SSNLF) least wants a price war, yet Cxmt Corporation (688825.SH) has planted a long-term source of marginal supply disruption precisely there. A third-party perspective can complete this puzzle. The affected areas for Micron are non-US server markets and long-term share erosion, with medium-term pressure smaller than for the Korean players; SK Hynix's HBM moat is deeper, giving it a thicker cushion when general-purpose DRAM is squeezed. The valuation premium of leaders has long come from the certainty of being two generations ahead; once same-generation competition begins, the source of that premium must switch to more realistic indicators such as yield, cost, and return on invested capital. On these indicators, Samsung Electronics Co., Ltd. (SSNLF) cannot even confidently beat SK Hynix. Moreover, it is still the pricing anchor for global memory assets, with index funds, passive allocations, and supply chain valuations all anchored to it, so every bit of sentiment from moat repricing will first be reflected in it; the company with the most stable technology has instead become the first landing point of narrative shock. In summary, for Cxmt Corporation (688825.SH), 4F² is a technology valuation repricing point, while mass-production yield, cost, and certification remain hard thresholds; its identity is upgrading from a domestic substitution supplier to a participant in next-generation architecture, and whether this upgrade can be delivered will be priced first by the market after external factors weaken, with the industry verdict coming around 2028. For Samsung Electronics Co., Ltd. (SSNLF), the short-term fundamental impact is limited. HSBC maintains Buy with a target price of 450,000 Korean won, UBS judges that memory supply will remain tight before 2028, and the market expects its 3Q26 operating profit at about 10.5 trillion Korean won and 4Q26 at about 11.6 trillion, with 2Q26 revenue of 172 trillion Korean won and memory sales up 471% year over year; over the long term, it must prove again that the time gap still exists. For the sector, the short-term AI boom has not broken, while over the long term it may gradually enter a stage where valuation is repriced first and profits are verified later. Disclaimer: This article does not constitute investment advice. The content comes from public platforms and brokerage research reports; the data may be lagging and its truth and validity are not guaranteed. The content in this article should not be used as a reference for any investment decision, and the individual stocks mentioned are not recommended in any way. Markets carry risks, and investment should be cautious.
Comments