Overview

The memory chip industry has entered a period of unprecedented growth, driven primarily by the insatiable compute and storage demands of artificial intelligence. Micron Technology and SK Hynix, two of the world’s three dominant memory manufacturers alongside Samsung, have been at the epicenter of this transformation. Both companies have seen their market capitalizations soar to historic levels—Micron crossed the $1 trillion mark on May 26, 2026, while SK Hynix surpassed 200 trillion won (approximately $150 billion) in March 2025 and now stands at roughly 225 trillion won (around $165 billion) [23]. These valuations are underpinned by a structural shift in memory consumption: AI data centers now consume an estimated 70% of global memory chip supply, and memory has been recast from a cyclical commodity into a defining strategic asset for the AI era [1][7]. This report analyzes the specific products and market segments—High Bandwidth Memory (HBM), DDR5, and NAND flash—that are driving growth at Micron and SK Hynix, examines the role of AI demand in their valuation milestones, and discusses the key risks and opportunities facing these companies and the broader memory chip sector.

Micron Technology: Product Segments and Growth Drivers

HBM Leadership and AI Demand

Micron’s most dramatic growth story has been its rapid ascent in the High Bandwidth Memory market. HBM revenue increased fivefold to $10 billion in fiscal 2025, up from approximately $2 billion in fiscal 2024 [1]. The company’s entire HBM production for 2026 was pre-sold under binding contracts well before the year began, and in March 2026 Micron signed the industry’s first five-year HBM supply agreement covering both volume and pricing [8]. This shift toward long-term contracted revenue provides unprecedented visibility and reduces the volatility historically associated with memory chip sales.

Micron is now shipping HBM4 in high volumes, designed specifically for NVIDIA’s Vera Rubin supercomputer platform, and has entered production using its most advanced 1‑gamma (1γ) DRAM process node [4]. The company’s HBM market share jumped from about 9% to 21% in a single year, closing the gap on SK Hynix [3]. Each gigabyte of HBM consumes roughly three times the wafer capacity of traditional DRAM, structurally limiting supply expansion and helping sustain high prices [5]. Hyperscalers such as Microsoft, Alphabet, Meta, and Amazon are investing directly in Micron’s HBM pipelines, providing demand visibility through 2029 [11].

DDR5 and Data Center Transformation

DDR5 memory has become the standard for AI training and inference servers, and Micron is one of the top three global suppliers alongside Samsung and SK Hynix [2]. The company’s data center revenue jumped from 35% of total business in fiscal 2024 to 56% in fiscal 2025, reflecting massive adoption of DDR5 and HBM in enterprise infrastructure [1]. Total revenue grew nearly 50% to a record $37.4 billion in fiscal 2025 [1].

The pricing environment has been extraordinary. DRAM export prices from South Korea surged 497% year-over-year as of May 2026 [3]. Contract prices for DRAM increased by 58% to 63% in the current quarter, and Gartner forecast a 125% jump in DRAM prices for the full year 2026 [5]. Samsung raised DDR5 prices by over 100% in December 2025 [3]. The migration from DDR4 to DDR5 accelerated dramatically due to AI demand, as DDR5’s higher bandwidth is essential for handling the massive data throughput of modern AI accelerators.

NAND Flash and Enterprise SSDs

While NAND contributes a smaller share of revenue compared to DRAM and HBM, Micron has made significant strides in enterprise solid-state drives. On May 5, 2026, the company began shipping the world’s highest-capacity commercially available SSD, the 245 TB Micron 6600 ION, which uses its G9 QLC NAND [12]. This drive targets AI, cloud, and hyperscale data center workloads, requiring 82% fewer racks than HDD-based deployments and consuming up to 30 watts at maximum power—half the power of comparable HDD deployments [12]. Micron is also shipping PCIe Gen6 SSDs and SOCAMM2 modules [4]. TD Cowen noted that Micron is applying the same pre-sold capacity strategy to NAND that it uses for HBM [13]. NAND export prices from South Korea rose 351.6% year-over-year as of May 2026 [3], and NAND prices are expected to rise faster than DRAM in the near term [14].

Market Capitalization Milestones and Financial Performance

Micron’s stock has appreciated roughly nine-fold over the twelve months ending May 2026, lifting its market capitalization from below $100 billion in early 2024 to above $800 billion by mid-May 2026, and finally to $1 trillion on May 26, 2026 [8][14][16]. The trigger for the final leg was an extraordinary upgrade by UBS analyst Timothy Arcuri, who tripled his price target to $1,625—the highest on Wall Street—implying a potential market cap of approximately $1.8 trillion within twelve months [8][9][10]. UBS argued there was “no reason” Micron should trade much differently from NVIDIA on a price-to-earnings basis, given the structural change AI has wrought on the memory industry [9]. Options activity on that day exceeded $2.8 billion in premium, more than the dollar amount traded in the SPY and QQQ ETFs combined [14].

Micron’s financials have kept pace with the valuation. In the fiscal second quarter of 2026, revenue reached $23.9 billion, up 196% year-over-year, with adjusted EPS of $12.20 (up eightfold) and gross margin of 75% [3][8]. The company guided Q3 FY2026 revenue to $33.5 billion, up 260% from the prior year [10][21]. Analysts project fiscal 2026 EPS of approximately $58.11 and fiscal 2027 EPS of $101.78 to $103 [5][15]. The stock trades at a forward P/E of roughly 8.4 times expected earnings, far below the S&P 500’s 21.1 times, suggesting potential for further re-rating [9].

SK Hynix: Dominance in HBM and Expansion in NAND

HBM Dominance and NVIDIA Partnership

SK Hynix remains the global leader in HBM, holding an estimated 53% share of the overall HBM market in 2024 and an even larger share—80% to 90%—in the high-growth HBM3 and HBM3E segments [4]. The company was the first to mass-produce HBM3E (8‑stack) in March 2024, using its 1b nm (fifth-generation 10 nm class) DRAM process, which delivers 10% higher yield and 20% better power efficiency than the previous node [15]. In early 2025, SK Hynix began sample shipments of 12‑stack HBM3E, with mass production targeted for mid-2025, offering 48 GB per module [3]. For HBM4, planned for 2026, SK Hynix will use a 1c nm process and has partnered with TSMC to manufacture the base die, marking a shift in advanced packaging supply chains [16][18].

SK Hynix’s close relationship with NVIDIA has been a cornerstone of its success. The company supplies approximately 80% of NVIDIA’s HBM3 and HBM3E needs [3] and signed a long-term supply agreement with NVIDIA in 2024 covering HBM3E through 2026 [16]. SK Hynix’s proprietary Mass Reflow Molded Underfill (MR‑MUF) packaging technology is a key differentiator, offering 60% higher thermal conductivity than conventional methods and enabling thinner, more efficient HBM stacks [17].

Funding from the AI boom is cascading into employee compensation: SK Hynix settled with its union to allocate 10% of annual operating profit directly to employees as performance bonuses for the next decade, with average projected payouts of $460,000 to $477,000 per worker in 2026 [3]. Meanwhile, a potential 45,000-worker strike at Samsung’s memory chip plants beginning May 21, 2026—the largest work stoppage in semiconductor history—could further entrench SK Hynix’s HBM lead [3].

DDR5 and Mobile Memory

SK Hynix competes strongly in the DDR5 market, with an estimated 35% share in Q4 2024, behind Samsung’s ~40% but ahead of Micron’s ~25% [19]. The company’s 1b nm DDR5, in volume production since Q2 2024, powers Intel’s Xeon 6 and AMD’s EPYC Turin platforms [19]. SK Hynix also introduced the world’s fastest LPDDR5T (Turbo) memory at 9.6 Gbps in early 2024, targeting AI PCs and premium mobile devices [20]. AI smartphones increasingly require 12 GB or more of DRAM, and AI PCs are moving toward 32 GB, which directly benefits SK Hynix’s mobile and compute business units.

NAND Flash and Solidigm

SK Hynix’s NAND business is executed through its subsidiary Solidigm (formerly Intel’s NAND and SSD business), which together hold about 25% of the global NAND market [7]. In November 2024, SK Hynix began mass production of 321‑layer 1 Tb TLC NAND, the highest layer count in the industry, beating Micron’s 232‑layer and Samsung’s 236‑layer technology [21]. The 321‑layer NAND uses a 4‑plane architecture to achieve data transfer speeds of 5.6 Gbps, a 20% improvement over its predecessor [21]. Solidigm’s enterprise SSDs, particularly QLC-based drives for read-intensive AI workloads, are gaining traction. In Q1 2025, NAND revenue was $3.9 billion, contributing 25% of SK Hynix’s total revenue, with operating margins improving to 18% [24].

Market Capitalization and Revenue Growth

SK Hynix’s market cap crossed 100 trillion won ($75 billion) in July 2024, 150 trillion won ($113 billion) in November 2024, and 200 trillion won ($150 billion) in March 2025 following NVIDIA’s GTC 2025 where Jensen Huang highlighted SK Hynix as a “key partner for AI memory” [23]. The company’s current valuation of roughly 225 trillion won ($165 billion) places it about $20 billion ahead of Micron [23]. SK Hynix’s revenue has expanded rapidly: Q4 2024 revenue was $14.1 billion (up 75% year-over-year), and Q1 2025 revenue reached $15.5 billion (up 82% year-over-year), with operating margins rising from 20% to 44% over that period [24]. HBM represented 42% of total DRAM revenue in Q1 2025, up from 36% in Q4 2024, underscoring the accelerating shift toward AI-driven memory [24]. Capital expenditure guidance for 2025 was raised to $15–16 billion, with about 70% allocated to HBM and advanced DRAM [25].

The Role of AI Demand in Trillion-Dollar Valuations

The trillion-dollar valuations of Micron and SK Hynix cannot be understood without recognizing the structural transformation AI has imposed on memory demand. Historically, memory chips followed a boom-bust cycle driven by consumer electronics and server refreshes. The current upcycle is fundamentally different: AI systems consume memory geometrically rather than linearly. NVIDIA’s H100 used 80 GB of HBM; the Rubin Ultra targets 512 GB per module [8]. Memory demand per AI system is no longer growing linearly with deployment but geometrically, as larger models require proportionally more bandwidth and capacity.

This demand is compounded by the manufacturing complexity of HBM. Each gigabyte of HBM consumes three times the wafer capacity of traditional DRAM, structurally capping supply expansion [5]. New fabrication facilities have long lead times, keeping supply constrained for years. Analysts believe the memory shortage could last until 2030 [7][11]. Samsung’s memory head Kim Jae‑jun stated in May 2026 that orders for 2027 are already booked, and the supply-demand gap is expected to widen further in 2027 compared to 2026 [5].

The rise of long-term supply agreements marks another structural shift. Micron signed a five-year HBM contract covering both volume and pricing, and UBS estimates that up to 30% of industry-wide DDR volumes are now locked in under 3–5 year agreements at slightly below current spot prices [8][10]. These contracts provide demand visibility, reduce pricing volatility, and transform memory companies from cyclical commodity players into quasi-utility suppliers with predictable earnings streams. UBS expects Micron to generate over $400 billion in free cash flow from 2027 to 2029 [10]. As a result, the market is re-rating memory stocks on a P/E basis more akin to semiconductor design companies like NVIDIA than to traditional cyclical memory makers.

Direct AI-attributable revenue for Micron is estimated to have reached approximately $20 billion in fiscal 2025, with a run-rate exceeding $40 billion in fiscal 2026. SK Hynix’s AI memory revenue (HBM plus AI SSDs) reached $18.5 billion in 2024, or 35% of total revenue, and is expected to approach $35 billion annualized in 2025, representing over 50% of total revenue [15][24]. Hyperscaler AI capex is expected to surpass $700 billion in 2026, up from approximately $400 billion in 2025, further fueling this trend [3][8].

Risks to the Memory Chip Sector

Cyclical Downturn and Overcapacity

Despite the current euphoria, history warns that memory cycles can turn abruptly. All three major suppliers—Samsung, SK Hynix, and Micron—are simultaneously scaling capacity at unprecedented rates. SK Hynix plans $27 billion in capex, Micron guided above $25 billion, and Samsung is also expanding aggressively [8][25]. When all three suppliers scale capacity in unison, oversupply has historically emerged within two to three years [8]. If hyperscaler AI spending slows or if AI model improvements reduce memory intensity, demand could collapse. The 2022–2023 downturn saw Micron post its largest quarterly net loss of $2.31 billion, cut 10% of its workforce, and see its stock decline roughly 50% [8]. The 2018–2019 downturn saw DRAM prices fall ~40% and Micron shares drop 57%. Investors must weigh whether the current structural changes—long-term contracts, geometric demand growth, and manufacturing complexity—are sufficient to permanently dampen the cycle or whether a correction is inevitable.

Geopolitical Tensions and Export Controls

Geopolitical risks are acute for both companies. China’s ban on Micron products from critical infrastructure procurement, imposed in May 2023, remains fully in effect, depriving Micron of a significant market that previously accounted for about 25% of sales [4]. US export controls continue to target Chinese memory makers YMTC and CXMT, restricting their access to advanced equipment, but these controls also create regulatory burdens for US and allied companies. SK Hynix’s Wuxi DRAM fab and Dalian NAND plant in China operate under US licenses that now contain national security end-use restrictions, limiting technology upgrades at Dalian to 128‑layer NAND or below [5]. Any escalation of US–China trade tensions could disrupt supply chains or force costly restructuring.

Taiwan contingency remains a top risk, as Micron has significant R&D and production in Taoyuan, and SK Hynix relies on Taiwan-based backend partners for HBM packaging. A potential workers’ strike at Samsung beginning May 21, 2026 could disrupt up to 100 trillion won ($67 billion) in production if prolonged, threatening overall memory supply and benefiting remaining players but also causing customer uncertainty [3].

Intense Competition and Technology Challenges

Samsung remains a formidable competitor. Its HBM4 chips began mass production in February 2026 and are already sold out for 2026, and Samsung is working to resolve heat and power efficiency issues that delayed qualification with NVIDIA [3][8]. Samsung’s 1c nm DRAM offers a ~20–25% die size reduction over 1b nm, and while yields are still around 60–70%, the company has deep pockets and massive scale.

Chinese memory makers YMTC and CXMT continue to advance, though constrained by US export controls. CXMT has reached 18 nm DRAM (about two generations behind Micron’s 1‑beta) and can produce DDR4 and LPDDR4 at scale, posing a price threat in lower-end segments [5]. YMTC is stuck at 128‑layer NAND due to equipment restrictions, but any future easing of controls could increase competitive pressure.

Technology scaling is also becoming more difficult and expensive. Nodes beyond 1c nm (1d nm and beyond) will require High‑NA EUV lithography, which is limited to a single supplier (ASML). Capacitor scaling in DRAM is approaching physical limits, and the industry is exploring novel architectures like 4F2 vertical transistor DRAM for 2030 and beyond [8]. NAND scaling toward 400+ layers requires extreme hybrid bonding and new materials.

Opportunities for Growth

AI Inference and On-Device AI

The next major wave of memory demand will come from AI inference—running trained models in production. Each NVIDIA Blackwell B200 GPU uses 192 GB of HBM3E, and as inference workloads expand, memory per server is expected to double from about 2 TB today to 4 TB by 2026 [5]. On-device AI is driving a step-change in memory content per device: AI PCs require a minimum of 16 GB of DRAM, moving to 32 GB by 2026, and AI smartphones require at least 12 GB (up from 8 GB) and up to 16 GB for flagship models [5]. This creates a sustained demand floor for DDR5 and LPDDR5 even if the data center boom moderates.

Emerging Technologies (CXL, Automotive, Edge)

Compute Express Link (CXL) enables memory pooling and disaggregation in data centers, allowing servers to share memory across nodes. The CXL memory market is expected to grow from $2.5 billion in 2025 to $15 billion by 2028, driven by data center disaggregation [11]. Both Micron and SK Hynix are investing in CXL products. Automotive memory is another growth vertical: ADAS and autonomous driving increase memory content per vehicle from ~4 GB to 16 GB or more, with L4/L5 vehicles potentially using 64 GB+ [5]. Edge AI in IoT and industrial applications will also boost demand for LPDDR5 and GDDR6 memory.

Structural Changes in Memory Contracting

The emergence of long-term, fixed-volume agreements covering both HBM and some DDR volumes fundamentally alters the earnings profile of memory chipmakers. These contracts provide revenue visibility, reduce pricing volatility, and make memory companies more attractive to long-term investors. If the industry can maintain a meaningful portion of output under contract even in a downturn, the severity of future cycles could be substantially reduced. SK Hynix’s 10-year profit-sharing agreement with its union also signals a shift toward stability and retention of skilled labor, which is critical for maintaining HBM yield advantages.

Conclusion

Micron and SK Hynix have been transformed by the AI boom into trillion-dollar entities, driven by explosive demand for HBM, DDR5, and advanced NAND products that are essential for training and deploying large-scale AI models. Micron’s $1 trillion market cap milestone on May 26, 2026, and SK Hynix’s valuation of over 200 trillion won reflect a structural re-rating of the memory industry as AI memory consumption grows geometrically, manufacturing complexity constrains supply, and long-term contracts provide earnings stability. However, risks remain substantial: the memory cycle has historically punished overcapacity, geopolitical tensions threaten supply chains and market access, and competition from Samsung and Chinese manufacturers persists. The opportunity for continued growth lies in the expansion of AI inference, on-device AI, CXL, and automotive memory—markets that are still in their early stages. The next few years will test whether the memory chip industry has truly broken free from its boom-bust past or whether the current valuations represent the peak of yet another cycle.