To The Moon

In May 2026, Gartner projected that global semiconductor revenue will surpass $1.3 trillion, with memory revenue expected to triple. DRAM and NAND prices have surged 125% and 234% respectively, signaling that memory chips have evolved from cyclical commodities into core strategic resources for AI infrastructure.

I. 📌 Memory: The "Physical Bottleneck" of AI Computing Power

Without memory, computing power is like a kitchen without rice.

The demand for memory in AI training and inference is growing exponentially. A single AI server consumes 2–3 times more memory than a traditional server, while large language models have reached hundreds of billions of parameters, requiring high-bandwidth memory (HBM) to enable high-speed data flow. Memory is no longer merely a "data warehouse" — it now operates as a parallel computing boundary alongside GPUs. When HBM supply falls short, even the largest GPU clusters will suffer from "data starvation."

The memory industry has recently entered a new round of price hikes. Market reports indicate that Micron briefly suspended DRAM quotations for approximately one week and plans to raise prices for DDR4, DDR5, and LPDDR products by roughly 20%–30%, with certain automotive-grade DRAM products potentially seeing even higher increases. Meanwhile, server DRAM prices from $SAMSUNG SEMICON(03132)$ and $SK Hynix, Inc.(HXSCL)$ have reportedly risen by approximately 60%–70%.

On the capital markets front, $Smart Sand Inc.(SND)$ has seen significant gains driven by AI storage demand and business restructuring expectations. $Western Digital(WDC)$ have also benefited from AI data center expansion and growing enterprise storage demand, with their stock performance notably outpacing the traditional semiconductor sector.

The market is reassessing the strategic value of the memory industry. As HBM (High Bandwidth Memory) becomes a critical bottleneck for AI GPUs and data centers, the memory sector is gradually shifting from its historical identity as a "highly cyclical industry" to an integral component of core infrastructure in the AI era.

II. 💸 "Memflation" May Persist Through End of 2027

Gartner forecasts that the global memory market will grow from $216.3 billion in 2025 to $633.3 billion in 2026, and further to $748.1 billion in 2027. Concurrently, Gartner expects DRAM prices to rise 125% year-over-year in 2026, and NAND Flash prices to surge 234%, with meaningful price relief unlikely before late 2027. (Gartner)

Bull Case:

• Hyperscalers' AI infrastructure spending in 2026 is projected to grow over 50%

• Continued growth in AI server and enterprise SSD demand

• HBM, DRAM, and high-end NAND supply remains tight, with no clear demand inflection point yet

Bear Case:

• Gartner warns that "Memflation" could destroy or at least delay non-AI demand until 2028

• Consumer electronics demand (PCs, smartphones) remains weak

• Excessively rapid memory price increases may suppress traditional end-market shipments and replacement cycles (Gartner)

III. 🔄 Three Dimensions of Technological Generation Leap

1️⃣ HBM (High Bandwidth Memory) Becomes the New Battleground: Traditional DDR4/DDR5 capacity is shifting toward HBM production, tightening supply of general-purpose DRAM. HBM unit prices are several times those of standard DRAM, driving memory vendors to upgrade their product mix.

2️⃣ NAND Shifts from "Capacity Competition" to "Value Competition": AI-related NAND bit consumption is expected to account for 20% of total demand while contributing up to 34% of market value. High-density NAND such as QLC/PLC is being deployed at scale in data centers, with enterprise SSDs (eSSD) becoming the primary growth engine.

3️⃣ Inventory Cycle Completely Reversed: Industry inventory has fallen from a peak of 12–16 weeks in 2023 to the current 6–8 weeks. Low inventory amplifies price elasticity, meaning any demand fluctuation will rapidly transmit to spot prices.

IV. 🏢 Oligopoly Intensifies, Capital Discipline Becomes the New Paradigm

Structural Shift: The memory industry now exhibits a "Three Superpowers, Two Strong Players" structure — Samsung, SK Hynix, and Micron dominate DRAM and advanced NAND; Western Digital/Sandisk and Kioxia form a second tier in the NAND space. In 2026, manufacturers are maintaining strict capacity discipline, with no signs of the blind expansion seen in previous upcycles.

Governance Model Shift: Capital expenditure (CapEx) has pivoted from "pursuing market share" to "pursuing profit." Vendors are prioritizing capacity allocation toward high-margin HBM and enterprise products over low-margin consumer electronics markets. This selective supply strategy essentially constitutes a "de-commoditization" of memory's commodity attributes.

V. ⚠️ Divergence and Risks

📉 Bear Case

• Demand Front-Loading Risk: Cloud vendors' AI capex growth may slow in 2027, triggering HBM order pullbacks.

• Non-AI Demand Collapse: PC shipments in 2026 are projected to grow only 3–5%, smartphone memory upgrades have stalled, and consumer-grade DRAM/NAND faces a "high prices, no buyers" scenario.

• Geopolitics: U.S. semiconductor controls on China could disrupt global memory supply chains. Technology breakthroughs by Chinese vendors such as YMTC (Yangtze Memory Technologies) could undermine oligopoly pricing power.

📈 Bull Case

• AI Inference Explosion: Beyond training demand, edge AI and real-time inference will create sustained — rather than one-time — memory demand.

• Deepening Technology Moats: HBM's 3D stacking, hybrid bonding, and other advanced packaging technologies raise industry barriers, making it difficult for new entrants to break through in the near term.

• "Storage as a Service": Cloud vendors are transitioning storage from hardware procurement to long-term service agreements, smoothing cyclical volatility.

VI. 🔮 Key Watchpoints for H2 2026

1. Micron Technology ( $Micron Technology(MU)$ , US Stock)

• 2026 YTD gain: 133.56% (as of May 6)

• 2026 full-year revenue forecast: approx. $109 billion (consensus estimate)

• Institutional target price: $1,000 (mainstream on Wall Street)

• Core drivers: full-capacity HBM sell-out, mass production of HBM4, market share recovery

2. SK Hynix ( $SK Hynix, Inc.(HXSCL)$ , South Korean Stock)

• 2026 YTD gain: +146% (as of May 7)

• 2026 revenue forecast: KRW 230 trillion (approx. $170 billion) (revised upward)

• Institutional target price: KRW 2.34 million (Nomura, highest); KRW 1.30 million (BOCOM International)

• Core drivers: 50%–60% HBM market share, mass production of HBM3E/4, profit margin of 62%

3. Samsung Electronics ( $SAMSUNG SEMICON(03132)$ , South Korean Stock)

• 2026 YTD gain: +107.43% (as of May 7)

• 2026 expected DRAM operating profit margin: 70%–80% (Goldman Sachs)

• HBM shipment forecast: triple that of 2025 to 11.1 billion Gb in 2026

• Core drivers: HBM share rebound to above 30%, 10nm DRAM, 28.8% NAND price increase

4. Western Digital ( $Western Digital(WDC)$ , US Stock)

• 2026 YTD gain: 180.46% (as of May 6)

• 2026 full-year revenue forecast: $12.8 billion (+34.7% YoY)

• Q4 FY2026 guidance: revenue of $3.65 billion (+40% YoY)

• Core drivers: AI cold storage HDDs, long-term cloud vendor agreements through 2028–2029

5. SanDisk ( $Smart Sand Inc.(SND)$ , US Stock)

• 2026 YTD gain: 493.98% (as of May 6)

• 2026 full-year revenue forecast: $10.45 billion (+42% YoY)

• Institutional target price: $2,000 (Susquehanna, highest)

• Core drivers: NAND/AI data centers, mass production of BiCS8, exploration of HBF technology

💡 One-Sentence Summary for Investors

Memory is no longer the cyclical sidekick of semiconductors — it is the rigid constraint on AI computing expansion. Bolstered by HBM technology barriers and oligopoly supply discipline, the "memory supercycle" of 2026–2027 is likely to continue, but investors should remain vigilant for a potential inflection in cloud vendor capex growth toward late 2027.

Data as of May 7, 2026