AI Memory Supercycle 2026: The Three Major Growth Curves and Structural Investment Opportunities of Samsung, SK Hynix, and SanDisk

When AI data centers have “drunk dry” the world’s memory, this is not a metaphor. In 2026, global AI data centers are consuming DRAM, HBM, and NAND flash capacity at an unprecedented pace. The combined capital expenditure of the world’s nine major cloud service providers (CSPs) has been raised to about $830 billion, with an annual growth rate as high as 79%. The first industry to be swept away by this capital flood is the memory sector at the very foundation of computing infrastructure.

In the first quarter of 2026, global DRAM revenue grew 80% quarter-over-quarter to a record high of $97 billion. The combined market size of DRAM and NAND Flash reached $137.14 billion, up 81.6% quarter-over-quarter. Meanwhile, total HBM (high-bandwidth memory) demand is expected to reach 32.279 billion Gb for the full year, up about 150% year-over-year. Supplier production capacity has already been fully locked up—SK hynix and Micron’s 2026 HBM inventories have both been sold out.

But beneath the revelry of this supercycle, there is a sharply different competitive landscape and investment narrative among the three growth curves. Samsung Electronics (005930.KS), SK hynix (000660.KS), and SanDisk (SNDK.US)—three storage giants—are staging a multi-party game along the three main tracks of DRAM, HBM, and NAND, showcasing differences in supply-chain dominance, technical depth, and business-model differentiation.

Market Landscape: A Power Map Defined by DRAM Market Share

In Q1 2026, the competitive landscape in the global DRAM market further concentrates toward the top. According to Counterpoint Research data, Samsung’s DRAM market share was 38% in Q1, ranking first; SK hynix was 29%; and Micron was 22%. Measured by the CFM flash market definition, Samsung’s Q1 DRAM sales revenue was $38.214 billion with a market share of 40.5%; SK hynix was $27.925 billion with a 29.6% share; and Micron was $18.768 billion with a 19.9% share. Although the two institutions use different data-cut methodologies, the overall picture is consistent: the top three manufacturers control about 90% of global DRAM capacity. This means any supply disruption by any single manufacturer will have an amplified effect on the global AI chip supply chain.

In the NAND flash space, however, the landscape is more dispersed. In Q1, the global NAND Flash market size reached $42.815 billion, up 81.8% quarter-over-quarter. As SanDisk, spun off from Western Digital as an independent pure NAND player, becomes its data-center business turns into a key incremental variable.

Samsung’s dominance in overall DRAM volume is unquestionable, but leading in total volume does not automatically translate into structural advantage. SK hynix’s leading position in HBM—the core AI track—means the “technology content” per dollar of DRAM revenue is significantly higher than Samsung’s general DRAM revenue. This also explains why the market assigns the two companies entirely different valuation logic.

3D Comparison Table of the Three Giants

| Dimension | Samsung Electronics (005930.KS) | SK hynix (000660.KS) | SanDisk (SNDK.US) | | --- | --- | --- | --- | | DRAM Market Share (Q1 2026) | 38% (Counterpoint) / 40.5% (CFM) | 29% / 29.6% | — | | HBM Market Share (2026E) | ~28% | ~50% | — (not an HBM market participant) | | Latest Quarter Revenue | 133.9 trillion KRW (Q1 2026) | 52.58 trillion KRW (Q1 2026) | $5.95 billion (FY Q3 2026) | | YoY Revenue Growth Rate | +69% | +198% | +251% | | Operating Profit Margin | 42.8% (group) | 72% | 78.4% (Non-GAAP gross margin) | | Core Growth Engine | Both volume and pricing rise + HBM4 catch-up | Exclusive HBM supply + profit explosion | Data-center NAND premiumization + long-term contracts | | Latest Quarter Operating Profit | 57.2 trillion KRW (YoY +756%) | 37.61 trillion KRW (YoY +405%) | Net profit approx. $1.43 billion (Non-GAAP) | | HBM4 Progress | Already shipped to Nvidia, 1c DRAM process; yield below 60% | Volume ramp in H2 2026, 16-layer stacking | — (not an HBM market participant) | | Key Risks | End-market drag, disputes over internal bonus allocation | High customer concentration in HBM, pressure from capacity expansion | Doubts about the sustainability of ultra-high gross margins, floating pricing risk |

Data sources: Counterpoint Research, CFM flash market, TrendForce, company financial reports; some are forecast data (marked “E”)

Three Growth Curves: Different Paths for the Giants

Curve One: Samsung—The “Elephant Dancing” Ruler of the Full Industry Chain

In Q1 2026, Samsung delivered a set of earnings reports worthy of the semiconductor history books: consolidated revenue of 133.9 trillion KRW, up 69% year-over-year; operating profit of 57.2 trillion KRW, up a whopping 756% year-over-year.

Breaking down the growth drivers: the DS (Device Solutions) division contributed 53.7 trillion KRW in operating profit in Q1, accounting for 93.9% of the group’s total operating profit. Within that, storage business sales reached 81.7 trillion KRW, setting a record high.

Samsung’s growth logic is “full-stack”—it is not only the world’s largest DRAM and NAND supplier, but also has self-developed wafer foundry manufacturing capabilities. In the HBM4 arena, Samsung is currently the only HBM4 supplier that completes the entire chain within its own fabs: DRAM manufacturing, logic chip production, and 3D packaging. Looking at industry asset-securities predictions, Samsung’s semiconductor division is expected to deliver full-year 2026 revenue of 223 trillion KRW (up 70% year-over-year), with operating profit of 84 trillion KRW (up 259% year-over-year).

But “an elephant dancing” also comes with heavy weight. Consumer end businesses such as smartphones and TVs (DX division) have performed weakly under the squeeze from rising prices of memory chips, making the “hot-and-cold extremes” between chip and finished-product businesses more pronounced. A deeper issue lies in the enormous bonus gap between the semiconductor division and the end-consumer division: storage division employees can receive bonuses of up to 600 million KRW, while device experience division employees can only get about 6 million KRW in company stock—triggering the most severe labor-management tension at Samsung since the company was founded. Although the union ultimately approved a temporary wage agreement with 73.7% of votes in favor—avoiding a planned 18-day general strike—this incident’s exposed underlying contradictions have not been resolved.

Curve Two: SK hynix—Profit Fragmentation from the HBM Dominator

If Samsung is the “king of scale,” SK hynix is the “king of efficiency.” In Q1 2026, SK hynix achieved revenue of 52.58 trillion KRW, up 198% year-over-year and breaking the 50-trillion-KRW mark in a single quarter for the first time. Operating profit was 37.61 trillion KRW, up 405% year-over-year, with an operating profit margin as high as 72%. Net profit was 40.35 trillion KRW, up 398%.

Even more striking is the quality of SK hynix’s earnings. Although HBM accounts for only about 14% of total DRAM shipments, it contributes more than 40% of DRAM revenue. SK hynix’s profit margin far exceeds that of traditional DRAM businesses thanks to HBM pricing power. In 2025, SK hynix already held more than half of the HBM market; in 2026, it is expected to maintain about 50% market share, remaining the industry leader.

SK hynix’s growth narrative is also deeply tied to Nvidia. As Nvidia’s most core HBM supplier for GPUs, SK hynix, leveraging MR-MUF (mass reflow molding bottom fill) advanced packaging technology and 16-layer stacking capability, has maintained technical leadership along the HBM4 track—its latest demonstrated 48GB HBM4 overall bandwidth has already exceeded 2TB/s.

UBS raised its target price for SK hynix to 1.7 million KRW this April, and forecast operating profit for 2026 of as much as 286 trillion KRW (about $19.3115 billion), about 57% above market expectations. UBS analysts described this storage cycle as a “supercycle once in 30 years.”

But the other side of the coin also deserves attention. SK hynix’s high reliance on Nvidia exposes it to customer concentration risk. If Nvidia’s GPU shipment pace were to fluctuate due to supply-chain or technical reasons, SK hynix’s capacity absorption would face pressure. In addition, SK hynix has planned to cut its 2026 HBM4 shipment volume to Nvidia by about 20% to 30% from the original plan to prioritize HBM3E supply required for the Blackwell platform—an adjustment reflecting uncertainty during the product-generation transition period.

Curve Three: SanDisk—The Most Undervalued AI Storage Story

SanDisk is the one among the three growth curves with the strongest “narrative reversal” color. After SanDisk was separated from Western Digital in February 2025, it shed the “conglomerate discount” associated with HDDs and became a pure NAND flash play.

The latest financial report data is startling: in FY Q3 2026, SanDisk achieved revenue of $5.95 billion, up 251% year-over-year; Non-GAAP EPS of $23.41; and a gross margin as high as 78.4%.

SanDisk’s growth logic rests on two key pillars. First, AI data centers are seeing a surge in demand for high-capacity enterprise SSDs. The company’s data-center business has jumped about 645% year-over-year, becoming the core source of profit. Second, SanDisk is rolling out a “new business model”—by signing 3-to-5-year long-term contracts with major customers, it locks in future revenue with remaining performance obligations (RPO) of about $42 billion.

However, market divergence around SanDisk is also the most pronounced. The drivers of its revenue growth almost come entirely from pricing power and product-mix optimization rather than shipment-volume growth—bit shipments are flat year-over-year and even decline quarter-over-quarter. Some analysts issued stern warnings, arguing that the floating pricing clauses in the “new business model” cannot protect revenue during an industry down cycle. Once Samsung and SK hynix release new capacity, SanDisk’s current 78% gross margin could face significant downside pressure.

Fact vs. view: SanDisk’s 78% gross margin is a data fact, but whether it can be sustained is a highly uncertain, scenario-based question. The two must be strictly distinguished.

Structural Engine: A Supply-and-Demand Equation Between AI Data Center Capex and Memory Demand

Demand Side

In 2026, the global tech giants’ investment in AI infrastructure has reached unprecedented levels. According to TrendForce data, the combined capital expenditure of the nine major CSPs has been raised to about $830 billion, with the year-on-year growth rate increasing from 61% to 79%. Microsoft has raised its capex outlook to $190 billion, up about 130%; Google has revised upward to $180 billion to $190 billion; Meta has revised its capex range to $125 billion to $145 billion, up about 85%; and AWS expects this year’s capex to exceed $230 billion, up more than 50%.

This capital stream directly translates into memory demand. Morgan Stanley estimates that in 2026 total HBM demand will reach 32.279 billion Gb, up about 150% year-over-year, with Nvidia accounting for roughly 54%. Gartner forecasts that in 2026 the full-year prices of DRAM and NAND flash will rise by 125% and 234%, respectively.

Supply Side

The tightness on the supply side is beyond general expectations. UBS’s latest analysis states that DRAM supply shortages will persist at least until Q2 2028, later than the prior forecast of Q4 2027. NAND supply shortages are expected to continue until Q4 2027. Micron has gone even further, saying that because the growth rate of demand for high-performance memory chips far exceeds the pace of capacity expansion, supply tightness for HBM, DRAM, and NAND will continue beyond 2026.

Forecast of Global AI Data Center Capex vs Memory Demand (Dual-Axis Chart Concept)

| Indicator | 2024 (Baseline) | 2025 | 2026 (Forecast) | Year-over-Year Change | | --- | --- | --- | --- | --- | | Combined capex of the nine major CSPs (USD billions) | ~3,000 | ~4,600 | ~8,300 | +79% | | Global DRAM market size (USD billions, annualized Q1) | ~272 | ~269 | ~970 (Q1) | +260% (Q1 YoY) | | Global NAND market size (USD billions, Q1) | — | — | ~428 (Q1) | +81.8% (Q1 QoQ) | | Total HBM demand (billion Gb) | ~41 | ~129 | ~323 | +150% | | DRAM price increase | — | Recovery from the low end | +125% (full year) | — | | NAND price increase | — | — | +234% (full year) | — |

Data sources: TrendForce, Counterpoint Research, Gartner, Morgan Stanley, CFM flash market

There are three structural reasons behind the current supply-demand imbalance. First, HBM chip manufacturing is extremely complex; wafer consumption is 2 to 3 times that of standard DRAM, and large amounts of capacity are being taken up by HBM, leading to synchronized tightness in general DRAM supply. Second, memory manufacturers were conservative in capital expenditures during the 2022 down cycle, creating a long lag period for capacity expansion. Third, migrating NAND to higher stacking layer counts faces dual pressure from both capital and process ramp-up. These structural constraints mean that even if demand growth slows, the supply gap will be difficult to close in the short term.

However, an unavoidable signal is that in March 2026, the year-over-year growth rate of global integrated circuit shipments was only 9.9%, while the revenue growth rate was as high as 99.5%. The number of chips sold increased by only about one tenth, but the industry made double the money—meaning the current memory industry boom is driven largely by price increases rather than real shipment growth. Once the supply gap begins to narrow, whether ASP can hold at the current level will face a severe test.

HBM4 Competition: The Technical High Ground That Determines the Next Cycle

HBM4 is the most intensely contested technical track in the memory industry today. The competitive posture among the three major original manufacturers is as follows.

SK hynix: The Defender with Advantages in Intergenerational Bridging

With deep cooperation with Nvidia during the HBM3E era, SK hynix maintains a clear advantage in HBM4 supply bit allocation. TrendForce expects SK hynix’s 2026 HBM market share to remain at around 50%, keeping it in the leading position. Technologically, SK hynix has demonstrated HBM4 samples with 16-layer stacking and 48GB capacity; by leveraging its self-developed MR-MUF technology and 12nm logic chips in collaboration with TSMC, it further strengthens its moat in high-end packaging.

Samsung: The Chaser of Full Industry-Chain Integration

Samsung adopts a strategy of “trading technology depth for time” in the HBM4 arena. It uses a more advanced 1c DRAM process (while SK hynix and Micron still use 1b DRAM), and it is the only supplier that completes the entire chain—DRAM manufacturing, logic chips, and 3D packaging. TrendForce expects Samsung’s 2026 HBM market share to be about 28%.

However, Samsung’s overall HBM4 yields are still below 60%; whether it can improve to mature levels in the second half will be the key variable in how quickly it can catch up.

Micron: A Flank Breakout Through Energy-Efficiency Differentiation

Micron focuses on energy-efficiency differentiation in HBM4. Its 36GB 12-layer single-stack product for Nvidia’s Vera Rubin achieves per-stack bandwidth exceeding 2.8TB/s, with data rates above 11Gb/s, and it further implements power optimization through self-developed CMOS base chips. TrendForce expects Micron’s 2026 HBM market share to be about 22%, and Micron’s HBM production bit share is expected to rise from 20% last year to 28%.

Risk Panorama: The Other Side of the Supercycle

Supply- rigidity Risk

HBM wafer consumption is 2 to 3 times that of standard DRAM. The large-scale shift of capacity toward HBM is squeezing general DRAM supply. A report from Nikkei states that even by the end of 2027, suppliers are expected to be able to meet only about 60% of global DRAM demand. For downstream manufacturers, this means continued cost pressure; but for memory OEMs, pricing power remains solid.

Consumer Electronics “Backfire” Risk

The surge in storage chip prices is directly driving up the costs of PCs and smartphones. Gartner expects that in 2026 global PC shipments will decline by 10.4% and smartphone shipments will decline by 8.4%. Whether weakening end-demand will feed back to upstream supply will be one of the most critical observation variables in 2027.

Divergence Between Shipment Volume and Revenue

The most worrying signal in the current cycle is “volume-price divergence”—shipment volume growth is less than 10%, while revenue growth is approaching 100%. Once supply tightness eases, mean reversion in prices could cause sharp fluctuations in revenue.

Uncertainty in HBM4 Validation and Mass-Production Timing

Due to HBM4 validation progress lagging behind expectations, KeyBanc lowered its 2026 Nvidia Rubin GPU production forecast from the prior 2 million units to about 1.5 million units. HBM4 validation progress directly affects downstream GPU delivery schedules, which could disrupt the entire AI chip supply chain.

Structural Risk in Samsung’s Internal Bonus Allocation

The bonus gap between Samsung’s semiconductor and non-semiconductor divisions (up to a 100x difference) not only triggers labor-management conflicts in the short term, but may also affect talent stability in non-memory businesses over the medium to long term. Packaging and testing are core processes for HBM; if they are impacted by work slowdowns, it will directly hit HBM4 capacity.

Layered Investment Logic: Who’s Winning, Who’s Chasing, and Who’s Betting?

For investors focusing on AI memory chip stocks in 2026, the three giants present three completely different investment narratives.

Samsung: A Defensive Leader’s Value-Recovery Logic

Samsung’s revenue scale of 133.9 trillion KRW in Q1 2026 keeps it firmly atop the storage industry. Its catch-up pace in HBM4, improved yields in its 1c DRAM process, and full-stack layout in NAND together form a solid fundamental base. Samsung’s investment logic is closer to “a value-recovery play on an undervalued industry leader”—when the market focuses too much on SK hynix’s HBM advantage, Samsung’s large-scale dominance in general DRAM and NAND markets, its self-developed foundry capability, and its HBM4 catch-up progress may not be adequately priced. However, continued weakness in end businesses and internal labor-management contradictions pose downside risks that need ongoing monitoring.

SK hynix: The Premium Logic of a High-Visibility Cycle Leader

SK hynix is the purest AI investment target in the current storage cycle. Its 72% operating margin is far above the industry average; HBM revenue accounts for more than 40% of DRAM revenue; and its HBM market share is expected to be around 50%. UBS raised its target price to 1.7 million KRW; Huaxing Securities gives a target price of 1,949,585 KRW. SK hynix’s logic is a growth-premium valuation—so long as the HBM supply-demand gap persists, its valuation has support. But overly high customer concentration and uncertainties in the intergenerational product transition period are the core risks to watch.

SanDisk: Divergent Opportunities Under a High-Odds Narrative

SanDisk is the most controversial among the three. A 78.4% gross margin and 251% revenue growth create a highly attractive fundamentals profile; however, the sustainability of its “new business model” and the fragility of floating pricing lead skeptics to question whether this is a “narrative-driven cycle illusion.” SanDisk’s investment logic is closer to a “high-odds gamble”—if NAND’s “de-cyclicization” narrative holds, the upside could be substantial; if the cycle reverts, the downside could be equally sharp.

Conclusion: Three Curves, Three Futures

The 2026 AI memory supercycle is not a homogeneous industry-beta story. Samsung, SK hynix, and SanDisk each represent a different growth paradigm: an all-round flagship driven by scale and technical depth; a growth expert anchored in a high-barrier niche; and a new independent force powered by business-model innovation.

In the short term, supply shortages will continue to support the earnings growth of all three. But from a mid-term perspective, where the three curves head depends on the answers to three core questions: Can HBM4 mass production support Nvidia Rubin’s expected shipment volume? Can NAND’s high-gross-margin narrative withstand the test of capacity releases? Can Samsung’s “elephant” find a new balance between its end business and its semiconductor business?

In this structural leap of the memory industry, the key question is no longer “will the cycle come?” but “after the cycle, who will still be sitting at the table?”