AI Memory Super Cycle: The Tri-Force of SK Hynix, Samsung, and Micron in HBM and the Restructuring of Trillion-Dollar Valuations

On April 23, 2026, South Korean chip manufacturer SK Hynix released a quarterly earnings report capable of reshaping industry perceptions: first-quarter 2026 revenue reached 52.58 trillion won, up 198.1% year over year; operating profit surged to 37.61 trillion won, with a year-over-year increase of 405.5%; and net profit also came in at 40.35 trillion won, up 397.6% year over year. All three core metrics—revenue, operating profit, and net profit—set record highs. It was also the first time a Korean chipmaker achieved quarterly revenue exceeding 50 trillion won.

Almost simultaneously on a parallel timeline, the other two key players in the global memory chip industry—Samsung Electronics and Micron Technology—were also experiencing their own historic moments. On May 26, 2026, Micron’s stock price jumped 19.29% in a single day; its market capitalization broke through $1 trillion (approximately $1.023 trillion), making it the third company in the global semiconductor industry to enter the trillion-dollar club. As of May 28, Micron’s stock price was temporarily at $928.41, and in after-hours trading it fell slightly to around $904. Meanwhile, in the first quarter of 2026, Samsung expanded its lead over SK Hynix further with a 38% DRAM market share.

Source: Google Finance

All three companies stand at the forefront of the AI memory supercycle. This is a story about HBM (High Bandwidth Memory)—a stacked DRAM design that fits directly with GPUs and AI accelerators, and has become the most scarce strategic resource in AI infrastructure.

How HBM evolved from a niche product into a core of the industry

HBM is not a new technology, but in the era of large AI models it has gained unprecedented strategic significance. Compared with traditional DDR or LPDDR memory, HBM achieves several times the bandwidth of traditional solutions per unit area by vertically stacking DRAM chips and interconnecting them using through-silicon via (TSV) technology. This enables GPUs to process larger datasets at lower power consumption, directly determining the efficiency of AI training and inference.

To understand the competitive landscape today, it is necessary to trace a clear timeline of technological iteration and strategic divergence:

2013 to 2022: Early deployment and strategic differences

In 2013, SK Hynix became the first to launch the world’s first HBM product, maintaining its position as a technology leader across subsequent generations—HBM2, HBM2E, and HBM3. Although Samsung has long held the top spot globally in DRAM production capacity and revenue, it started later in this niche area and only significantly increased investment in the HBM3/HBM3E era. Micron chose to skip HBM3 and enter the market directly from HBM3E, using a leapfrogging strategy to capture the advantage of a late mover.

2023 to 2024: The HBM3E era takes shape

With the explosive growth of generative AI, demand for HBM surged rapidly. SK Hynix became Nvidia’s top-priority supplier of HBM3E, delivering products first to the Blackwell platform. According to Counterpoint Research data, from Q4 2024 to Q3 2025, SK Hynix’s HBM market share continued to remain above 50%. In September 2024, SK Hynix became the first to achieve mass production of HBM4 globally, further strengthening its early-mover advantage.

Second half of 2025 to early 2026: Competition in HBM4 heats up

During CES in January 2026, Nvidia officially announced its next-generation Vera Rubin GPU architecture. HBM4’s technical specifications were also clarified at this stage: the memory interface width doubled from 1024 bits to 2048 bits, the capacity per stack was increased to 48GB (16 layers of stacking), and system-level bandwidth reached 22TB/s—about three times that of early Blackwell systems. In March of the same year at GTC 2026, Samsung first publicly demonstrated physical samples of HBM4E; single-pin data rate reached 16Gbps, with total bandwidth of 4.0TB/s.

Second quarter of 2026: The trillion-dollar market-cap era arrives

As of May 2026, the market capitalizations of all three memory giants had already surpassed or were close to $1 trillion. SK Hynix’s market cap broke through $1 trillion in late May, becoming the second listed company in South Korea to reach this milestone. Micron followed closely and also joined the trillion-dollar club. Samsung, meanwhile, continued to maintain its leading position in market capitalization in the South Korean stock market.

A three-dimensional breakdown of the competitive position of the three leaders

Market share: the two-layer structure of DRAM and HBM

To understand the competitive posture of the three companies, it is necessary to distinguish two levels of market share: one is the overall DRAM market, and the other is the AI-specific HBM submarket.

According to data released by Counterpoint Research on May 27, 2026, global DRAM revenue in Q1 2026 reached $97 billion, setting a historical high, up 80% quarter over quarter and up 260% year over year. In terms of market share:

| Metric | Samsung | SK Hynix | Micron | | --- | --- | --- | --- | | DRAM market share (Q1 2026) | 38% | 29% | 22% | | HBM market share (forecast for 2026) | ~28% | ~50% | ~22% |

Data sources: DRAM share from Counterpoint Research; HBM share from TrendForce forecast data (SK Hynix ~50%, Samsung ~28%, and Micron the remaining share).

The mismatch between the two sets of data is notable: Samsung ranks first with a 38% overall DRAM market share, but in the high-value HBM subsegment SK Hynix leads by a wide margin with about 50% HBM share. This mismatch reflects two very different strategic paths: Samsung pursues scale advantage and broad product coverage, while SK Hynix concentrates resources on the high-profit AI memory track.

Quarterly changes in overall DRAM market share also convey important signals: in the first quarter, Samsung’s share increased by 2 percentage points quarter over quarter (from 36% in Q4 to 38%), while SK Hynix’s share fell by 3 percentage points quarter over quarter (from 32% to 29%), indicating that Samsung is expanding its overall production capacity more aggressively.

Financial performance: divergence in profitability

The Q1 2026 financial results clearly show different qualities of profitability among the three companies amid the current supercycle:

SK Hynix: Single-quarter revenue of 52.58 trillion won, operating profit of 37.61 trillion won, and an operating profit margin as high as 72%; net profit margin of 77%. All three core indicators hit the highest records in the company’s history. DRAM ASP rose about 65% quarter over quarter (mid-60%); NAND ASP rose about 75% quarter over quarter (mid-70%); gross margin and EBITDA margin both reached 79%. Q1 net non-operating profit was 14 trillion won, including non-recurring items such as net foreign exchange gains of 1.6 trillion won and investment asset valuation gains of 9.9 trillion won, among others.

Samsung Electronics: Samsung’s overall business covers multiple segments including storage chips, foundry, smartphones, and home appliances, but the specific profit for its storage business is not separately disclosed. In a May 18 report, Nomura Securities raised Samsung’s target price from 340,000 won to 590,000 won, and predicted that AI inference demand would ignite a new round of memory demand. This suggests that the market is also highly optimistic about the profitability outlook of Samsung’s storage business, but the overall valuation is constrained by drag from non-storage segments.

Micron Technology: Micron’s FY2026 third fiscal quarter (as of May 2026) revenue guidance is approximately $33.5 billion, up more than 260% year over year. Micron’s FY2026 Q2 non-GAAP earnings per share is approximately $8.42. Micron’s 2026 HBM supply has already been fully sold out.

From a margin perspective in a horizontal comparison: SK Hynix’s 72% operating profit margin is at the leading level in the current industry cycle. KB Securities forecasts that SK Hynix’s full-year 2026 operating margin will reach 78.1%, putting it in a position to rank first globally in the semiconductor industry, surpassing Nvidia and Saudi Aramco.

Valuation framework: how target prices and pricing logic have evolved

The valuation logic for the three companies is undergoing a fundamental shift from “cyclical pricing” to “structural revaluation.” Below is a summary of the latest target prices from major brokerages as of May 28, 2026:

| Company | Brokerage | Target Price | Core Logic | | --- | --- | --- | --- | | SK Hynix | Nomura Securities (May 17) | 4 million won | AI memory demand grows on an index-level scale | | SK Hynix | KB Securities (May 15) | 3 million won | 2026 operating profit margin 78.1% | | SK Hynix | Future Asset Securities (May 27) | 3.8 million won | ROE improvement and valuation re-rating | | Micron Technology | UBS (May 26) | 1,625 dollars | LTA locks in profits + valuation framework re-assessment |

Data sources: Nomura Securities May 17 report; KB Securities May 15 report; Future Asset Securities May 27 report; UBS May 26 report.

Nomura Securities’ pricing logic is the most aggressive. Its target price of 4 million won is a substantial increase from the prior target price of 2.34 million won, arguing that AI memory demand is in an “exponential growth” channel. Major cloud service providers are signing long-term agreements (LTA) to lock in prices and production capacity through prepayments.

UBS’s revaluation of Micron is more methodologically significant. The firm’s analyst Acuri has abandoned the traditional sum-of-the-parts valuation approach and instead uses a price-to-earnings framework based on discounted future earnings. Its core assumption is that even if the industry enters a downturn, Micron can still maintain a relatively stable level of profitability, so the valuation system should shift from the low-multiple valuation typical of traditional memory manufacturers to a higher level. Based on profit stability supported by long-term agreements, it assigns a 15x forward P/E multiple, setting a target price of 1,625 dollars.

Dissecting sentiment: four main threads amid bulls-bears divergence

Current market views on the three memory giants are not uniformly optimistic, and the bulls-bears divergence mainly centers on the following four main threads:

Theme 1: Has the supercycle arrived vs the cycle has never truly disappeared

The core argument from the bull camp is a structural shift in demand. Nomura Securities explicitly proposed the concept of a “new mechanism,” arguing that the old-cycle model cannot explain the current supply-demand landscape—AI memory demand is in an exponential growth channel, while the supply side is constrained by wafer fab construction cycles, bottlenecks in the supply of EUV lithography equipment, and limitations in advanced packaging capacity, making expansion extremely limited.

In contrast, a Morningstar analysis sounded a cautious note: SK Hynix has risen by about 88% year to date in 2026; even record-breaking quarterly data may not be enough to drive the stock price substantially higher. The article points out that the key question is whether the boom-bust cyclicality of the storage industry has been permanently eliminated or is merely temporarily dormant. Notably, SK Hynix’s stock price fell on the day its earnings report was released, reflecting that the market had already absorbed expectations in advance.

Theme 2: SK Hynix leads in HBM vs Samsung’s strong counterattack

The mainstream view holds that SK Hynix has a clear first-mover advantage in the HBM arena. According to a report by the Korean Economic Daily in March 2026, SK Hynix accounted for about 70% of the allocation in Nvidia Vera Rubin’s initial HBM4 orders, while Samsung took about 30%. TrendForce predicts that in 2026 SK Hynix will continue to lead global HBM supply, with HBM bit output share of about 50%.

But Samsung’s momentum cannot be ignored. In March 2026, Samsung had already obtained HBM4 certification from Nvidia (two speed tiers: 10Gbps and 11Gbps) and began shipping HBM4 in February. Samsung also compressed its HBM R&D cycle from about two years to within one year to closely match the cadence at which customers like Nvidia iterate new products each year. At the GTC conference in March 2026, Samsung’s demonstrated physical samples of HBM4E have already achieved a 16Gbps single-pin data rate and total bandwidth of 4.0TB/s.

Theme 3: Micron is being sidelined vs an opportunity at a valuation trough

A widely discussed piece of news is that, according to a report by the Korean Economic Daily in March 2026, in Nvidia Vera Rubin flagship platform’s HBM4 supplier list, only Samsung and SK Hynix appear, and Micron is not included. However, industry analysis suggests that Micron may still supply HBM4 for lower-tier AI inference accelerators in the Rubin series (such as Rubin CPX), rather than being completely out of the picture.

UBS’s analysis provides a sharply different perspective: Micron’s valuation starting point is far lower than that of its two Korean competitors, and once its HBM business scales up and forms stable profitability, there is greater room for valuation re-rating. Micron’s 2026 HBM supply has already been fully sold out. The HBM market size is expected to grow from $35 billion in 2025 to $100 billion in 2028, with an estimated CAGR of about 40%.

Theme 4: Long-term agreements reshape valuation vs traditional cycles will eventually return

UBS’s research report notes that new long-term agreement terms are typically 3 to 5 years, employing a dual-lock mechanism for volume and price, requiring buyers to assume commitments for both prepayments and capital expenditure support. Major cloud service providers have locked in 60% to 70% of the industry’s DDR5 server shipment volume through enhanced LTAs. By 2027, about 20% to 30% of the industry’s DDR bit shipments will be covered by agreements of this type.

This trend is fundamentally changing how the market values the storage industry. Traditionally, memory makers are viewed as strong cyclical stocks—profits expand at the peak of the cycle, and shrink dramatically or even turn into losses at the trough. If the LTA mechanism truly achieves profit smoothing, storage leaders could maintain relatively stable profitability even during industry downturns, thereby supporting higher valuation multiples.

Industry impact analysis: structural reshaping of the memory sector

The AI memory supercycle is not simply a price appreciation cycle; it is reshaping the underlying logic of the global storage chip industry across the following four dimensions:

Structural tilt in capacity allocation. The die size of HBM chips is far larger than that of standard DRAM with the same capacity, meaning each HBM chip produced displaces several times the conventional DRAM capacity. According to Counterpoint Research data, since the second half of 2025, cloud service firms have accelerated investment in AI infrastructure, driving a surge in HBM demand and squeezing out large portions of conventional DRAM capacity, pushing the entire DRAM market into supply shortage. Feng Li, President of SEMI China, said that although the three major manufacturers have directed 70% of new/incrementally adjustable capacity toward HBM, the HBM capacity gap still stands at 50% to 60%. This capacity displacement effect is key to understanding why storage prices are rising across the board.

Deep customer relationship binding. Unlike the standardized product model in the traditional storage industry, HBM supply relationships are evolving into highly customized strategic cooperation. SK Hynix’s “One-Team” alliance with TSMC, Samsung synchronizing R&D cycles with Nvidia, Micron’s long-term agreements with major cloud service firms—each supply chain implies deep technical coupling and mutual commercial lock-in. This relationship structure raises the entry barriers to the HBM market well beyond traditional storage products, further stabilizing the competitive positioning of existing suppliers.

Accelerated R&D pace arms race. Samsung’s initiative to compress the HBM R&D cycle from two years to within one year signals that the entire industry has entered an “arms race” mode. AI chipmakers generally set a strategy of annual product line updates, and HBM suppliers that cannot keep up with both technical upgrades and volume-production delivery face the risk of key customer orders being shifted. This accelerated cadence implies that R&D intensity will continue to climb, and companies with larger scale effects and deeper technical accumulation will gain ever-increasing competitive advantages.

China’s attempts to break through. Beyond the global “three-way dominance” landscape, Chinese memory manufacturers are accelerating their catch-up efforts. CXMT has achieved a 1β-nanometer process breakthrough in DRAM and mass-produced DDR5. In Q1 2026, DRAM revenue grew more than 700% year over year; global market share expanded from 3% in the prior-year period to 8%, making it the world’s fourth-largest DRAM supplier. While Chinese firms still have a clear generational gap in HBM compared with the three giants, their pace of capacity expansion and technology catch-up is worth continued attention.

Conclusion

Within the grand narrative of the AI memory supercycle, the three storage giants each occupy different competitive positions. SK Hynix has built a first-mover advantage and technology moat in HBM, the most valuable niche segment; Samsung remains ahead in overall scale by virtue of the world’s most complete storage product lineup and its massive production capacity; and Micron, with a lower valuation starting point and the most aggressive expectations for value re-rating, offers investors another risk-reward profile.

Ultimately, the determining factor of investment value may not be who wins the supply competition for a specific generation of HBM. Rather, it is who can maintain technological foresight, discipline in capacity expansion, and stickiness in customer relationships throughout the long period of evolution of AI infrastructure—the combined depth of these three capabilities will ultimately define each company’s valuation boundary.