In 2026, the fiercest battleground in the global semiconductor industry will not be CPUs or GPUs, but rather the seemingly unassuming memory chips. The memory sector is poised for a historic dual transformation driven by capital and industrial dynamics: Changxin Technology, the sole domestic DRAM leader.Changxin Technology<$CXMT Corporation (688825.SH)$>has confirmed that its STAR Market IPO subscription will open on July 16, raising tens of billions of yuan to power its breakthrough; the world’s second-largest memory manufacturerSK Hynix (SKHY.US)is accelerating its push toward a U.S. listing, launching a new round of global capital expansion.
DRAM: A Strategic Linchpin in the AI Era?
DRAM (Dynamic Random-Access Memory) serves as the 'working memory' of electronic devices, temporarily storing data currently being processed by the CPU. Fast and moderately capacious, it complements NAND flash memory—which handles long-term storage—and represents the single largest product category in the semiconductor industry by scale.
Amid the AI-driven industrial revolution, DRAM’s strategic importance is being redefined. Large model inference requires loading all parameters into memory while reserving massive space for context windows. This has directly driven expectations for average server DRAM capacity to surge from 1,032GB in 2025 to 1,432GB in 2026. According to Omdia and WSTS, the global DRAM market will reach $150.5 billion in 2025, accounting for 65% of the total memory chip market.
Against this backdrop, Changxin Technology’s role stands out prominently. It is China's only integrated device manufacturer (IDM) with large-scale DRAM R&D, design, and manufacturing capabilities, operating three 12-inch wafer fabs in Hefei and Beijing. Its global market share rose from 3% in Q1 2025 to 8% in Q1 2026, surpassing Nanya Technology to solidify its position as the world’s fourth-largest DRAM supplier—trailing only Samsung, SK Hynix, andMicron Technology (MU.US), as shown in the figure below. ChangXin has already formed deep partnerships with core clients including Alibaba Cloud, ByteDance,Tencent<$TENCENT (00700.HK)$>, Lenovo <$LENOVO GROUP (00992.HK)$>, Xiaomi <$XIAOMI-W (01810.HK)$> and other key customers.

However, it should be noted that among the many DRAM sub-segments, HBM (High Bandwidth Memory) is the most critical 'hard currency' of the AI era.
Traditional DRAM communicates with chips via a bus, which faces physical bandwidth limits. In contrast, HBM uses TSV (Through-Silicon Via) technology to vertically stack multiple layers of DRAM chips and connects them to GPUs/CPUs through micro-bumps for close-proximity, high-speed interconnects, achieving data transfer bandwidth dozens of times higher than conventional DDR memory. It is precisely HBM that enables trillion-parameter large models to overcome the 'memory wall' bottleneck during inference. As AI chips from NVIDIA (NVDA.US) and AMD (AMD.US) double their computing power with each new generation, demand for HBM—particularly in terms of stacking layers and bandwidth—continues to rise sharply.
Currently, SK hynix and Micron Technology have already achieved mass production and shipment of HBM3E, while Samsung is accelerating its push to deliver HBM4E samples, firmly dominating the global high-end HBM market. ChangXin Technology is also aggressively developing HBM3 and has completed sample validation with domestic customers, but has not yet reached large-scale commercialization. It likely still lags behind global leaders by a noticeable 2–3 years in stacking processes, yield rates, and high-end product iteration cycles.

Technology Showdown: Four Entirely Different Tracks
Although AI has reshaped the competitive landscape, the fundamental rules of the memory chip industry remain unchanged: premium segments capture AI-driven upside, mid-tier competes on scale, and low-end battles on cost. The technology roadmaps and strategic choices of these four giants may ultimately determine their respective profit ceilings.
1) SK hynix: Biggest winner of the AI era, holding the industry’s 'money-printing machine'—HBM
If there’s a clear winner set to effortlessly dominate the semiconductor industry in 2026, it has to be SK hynix. Rather than pursuing broad-based balance, it has zeroed in on one premium segment—High Bandwidth Memory (HBM)—directly grasping the lifeline of global AI computing power.
SK hynix currently commands a dominant 58% share of the global HBM market and serves as NVIDIA’s most critical supplier—over 70% of the HBM memory used in top-tier AI chips like the GB200 and B200 comes from SK hynix. While competitors are still locked in price wars over conventional memory, SK hynix is already selling ‘ultra-premium’ HBM products, which typically deliver gross margins more than three times those of standard DDR memory. In Q1 2026, SK hynix’s gross margin surged by 22 percentage points year-over-year, reaching 79.3%.
On the technology front, SK hynix has completed development of the industry’s first 1cnm LPDDR6. Its 1cnm 192GB SOCAMM2, optimized for NVIDIA’s Vera Rubin platform, has entered mass production. For HBM4, SK hynix has engaged in deep collaboration with customers from the earliest development stages to ensure volume ramp-up proceeds according to schedule.
However, the company’s weaknesses are equally pronounced: it suffers from severe over-concentration. Its NAND flash market share stands at just 18%, down 4 percentage points from last quarter, leaving its entire foundation reliant on DRAM and HBM. Notably, rivals Samsung and Micron Technology have also expanded their HBM market shares this year. Should AI demand soften, customers accelerate in-house alternatives, or these two competitors overcome their technological and capacity bottlenecks, SK hynix would face the most significant impact.
2) Samsung Electronics: The true ‘price-setter’ of the industry?
If SK hynix is the specialist champion, Samsung is undoubtedly the all-around powerhouse—the only company globally that leads in both DRAM and NAND flash memory markets.
According to Counterpoint data for Q1 2026, Samsung holds a 38% share in DRAM and 29% in NAND, dominating across smartphones, PCs, servers, AI computing, and consumer electronics. It not only possesses mature 1b-node DRAM and HBM technologies but also develops its own 3D NAND stacking processes, with full vertical integration across equipment, materials, and wafer fabrication—giving it overwhelming cost advantages over rivals.

More importantly, Samsung has its own smartphone, end-device, and server businesses. During industry downturns, if external orders fall short, internal demand can absorb excess capacity, perfectly hedging against cyclical volatility. Samsung has already shipped its first HBM4E samples to reinforce its technological leadership, with HBM4 expected to begin supplying NVIDIA in the second half of 2026.
Its main weakness lies in an overly diversified business structure—consumer appliances, smartphones, and semiconductors are bundled together, diluting the profitability of its high-quality memory business with lower-margin segments. This is why its overall gross margin lags behind both SK hynix and Micron Technology, resulting in consistently lower valuation multiples. Additionally, being overtaken by SK hynix in HBM progress is the core reason capital markets question Samsung’s ‘AI purity.’

3) Micron Technology: North America’s sole representative, holding its ground through geopolitical barriers
Micron Technology lacks SK Hynix's AI-driven windfall profits and Samsung's full-spectrum scale, yet it remains an irreplaceable presence in Western markets. As the only North American memory IDM of significant scale, its core moat lies not in cutting-edge technology, but in supply chain security.
Western governments, enterprises, defense contractors, and national data centers prioritize Micron products due to stringent security requirements—a closed and highly stable market segment that Samsung and SK Hynix struggle to penetrate.
On the technology front, Micron has adopted a steady and pragmatic approach: its HBM4 12-layer high-stack product is ramping up production at twice the speed of HBM3E 12-layer, already generating over $1 billion in HBM4 revenue. Its 1γ DRAM and G9 NAND nodes are also ramping smoothly and are poised to become the highest-volume nodes in Micron’s history. Additionally, Micron pioneered Strategic Customer Agreements (SCAs), signing 16 such deals—five-year terms for data center and consumer electronics clients, and three-year terms for automotive customers—locking in approximately 20% of its DRAM shipments and one-third of its NAND shipments.
However, its weaknesses are equally evident: limited production capacity, no proprietary end-user products, and complete reliance on external customers. It cannot compete with SK Hynix for the high-end AI memory boom and lags behind the two Korean giants in scale.
4) CXMT: China’s domestic challenger—solid in mid-range, still lagging in high-end
CXMT faces a classic uphill battle—climbing against the wind and breaking through barriers step by step. As China’s only DRAM IDM capable of large-scale production, it is the sole player among the top four memory makers restricted from accessing EUV lithography tools and locked out of advanced technologies.
The good news is that it has fully penetrated the mid-range segment: employing a 'generation-skipping' R&D strategy, CXMT has successfully mass-produced across its first to fourth-generation technology platforms, offering a full portfolio including DDR4, DDR5, LPDDR4X, and LPDDR5/5X. While global peers shifted focus to high-end AI memory, CXMT captured the incremental demand in the mainstream memory market—its global market share surged from 3% in Q1 2025 to 8% in Q1 2026, making it the only one among the top four to achieve significant share gains. Rumors also suggest Apple is considering adding CXMT to its supplier list.
Yet its shortcomings remain critical: without EUV lithography, it relies solely on DUV multi-patterning, which may impose a physical ceiling on advanced node scaling. Its HBM development is still in early stages, potentially causing it to miss out on the most lucrative AI memory boom of the past two years.
Its recent STAR Market IPO fundraising appears to be a targeted 'campaign to close key gaps': RMB 7.5 billion allocated to wafer fab technology upgrades, RMB 13 billion to DRAM technology enhancement, and RMB 9 billion to forward-looking R&D. By aggressively investing to upgrade processes, develop HBM, and expand capacity, CXMT aims to break into the high-end segment from its mid-range stronghold—this will be its most critical battle over the next three years.
Capacity vs. cycle resilience: who can navigate both bull and bear markets?
If technology determines a company’s upside, then capacity deployment and business structure may determine its resilience through economic cycles. Under the new memory landscape in 2026, the capacity strategies of these four giants may somewhat reflect their risk-resilience capabilities.
Samsung has built a safety cushion through its globally largest production capacity, with coordinated wafer fab operations across multiple countries including China and South Korea. It can flexibly adjust the mix between mature and advanced production lines to smooth out price volatility. Its diversified group businesses also help hedge against short-term fluctuations in the memory sector. However, memory accounted for more than half of the group's revenue (Q1 2026 data), meaning industry cyclicality still deeply impacts overall earnings.
SK Hynix has allocated most of its advanced capacity to the HBM segment, with core production lines in South Korea and Wuxi fully dedicated to fulfilling AI memory orders, maximizing profitability during the current super-cycle in AI demand. However, its heavy concentration of advanced capacity on AI compute, coupled with general-purpose DRAM supplied solely through its Wuxi facility, highlights significant risks: regional concentration and a narrow business mix leave it vulnerable to sudden shifts in demand, with limited diversification to buffer against downturns.
Micron Technology demonstrates the most mature counter-cyclical strategy. On one hand, its globally dispersed network of wafer fabs and packaging facilities effectively mitigates geopolitical risks tied to any single region. On the other, long-term supply agreements lock in shipments, while exposure to non-cyclical niche markets—such as automotive and defense—insulates it from consumer-driven volatility. With DRAM as its dominant product and NAND as a meaningful complement, Micron captures full upside during upcycles while maintaining a stable baseline during downturns.
CXMT’s cyclical logic is unique—driven by domestic substitution. Utilization rates at its three 12-inch wafer fabs in Hefei and Beijing have steadily risen from 87.06% to 95.73%, and its capacity-based sell-through ratio has consistently remained above 90%. Its top five customers account for roughly 68% of sales, spanning diverse domestic end-markets including cloud services, consumer electronics, and smart manufacturing—all underpinned by structural demand. During upcycles, CXMT benefits from an extremely low earnings base and rapidly expanding market share, delivering exceptional earnings elasticity; the company forecasts its first-half 2026 revenue to grow more than sixfold year-over-year, with adjusted net profit attributable to shareholders surging 22.8x to 25.3x. In downturns, ongoing domestic digital infrastructure investment and policy support for localization provide a solid floor. However, its key weakness lies in lacking high-margin HBM capacity, limiting its influence in premium segments—a gap the current multi-billion-dollar fundraising round aims squarely to close.
Valuation logic in capital markets: How should 'computing power' be priced?
As of now, Samsung Electronics, SK Hynix, and Micron Technology have all surpassed a $1 trillion market capitalization (see table below). Year-to-date, their share prices have risen by 132.34%, 236.42%, and 232.62%, respectively. Despite the sharp stock rallies lifting valuation multiples, their forward P/E ratios remain relatively modest, ranging between 8x and 13x.

Compared with its overseas peers, CXMT—set to list on the STAR Market—may enjoy two key premium advantages:
First is scarcity-driven premium: CXMT is China’s only integrated device manufacturer (IDM) with full DRAM capabilities, serving as the nation’s flagship for semiconductor memory self-reliance. It enjoys deep backing from China Integrated Circuit Industry Investment Fund II (Big Fund II) and local state-owned capital, ensuring sustained policy tailwinds. Second is growth-driven premium: its DRAM market share surged from 3% to 8% within a year. With server DDR5 ramping up, a multi-billion-dollar capital raise fueling capacity expansion, and robust domestic substitution momentum converging, CXMT has substantial growth runway ahead. The company expects its 2026 adjusted net profit attributable to shareholders to grow over 22-fold.
Unlike overseas giants valued primarily on cyclical industry dynamics, CXMT’s valuation hinges not on short-term market sentiment but on the long-term certainty of domestic substitution, tangible progress in technological catch-up, and continuous gains in global market share.
Conclusion
The major transformation in the memory storage industry, driven by AI computing power, continues to unfold. The strategic choices and competitive dynamics among the four leading players are far from settled. Overseas giants, leveraging their advanced technology and global capital advantages, are holding their ground at the high end of the industry, while ChangXin has embarked on its catch-up journey through fundraising via its listing—a path of upward breakthrough for domestic memory solutions that warrants long-term observation.
Risk Disclaimer: The above content only represents the author's view. It does not represent any position or investment advice of Futu. Futu makes no representation or warranty.Read more
Comment (1)
to post a comment
2
7
