NAND Suppliers Conservative on Technology Migration and Capacity Expansion

Key Takeaways

• Strategic Pivot to HBM/DRAM: Major suppliers (Samsung, SK hynix, Micron) are structurally reducing NAND capacity, with 2027 levels projected to be 40% lower than the 2022 peak. Capital and fab space are being diverted to high-margin, AI-driven HBM and advanced DRAM.

• Supply Constraints through 2026: Despite rising AI demand, vendors remain conservative due to 2023 losses. Since significant new capacity will not arrive until after 2027, the market will face a "tight supply-demand balance" through 2026.

• Focus on Technology Node Transition: The industry consensus is to meet demand through technology node transitions to increase bit density (Gb/mm²) rather than blindly building new wafer fabs. This technical upgrade drives bit output growth while maintaining supply discipline.

I. Global Capacity Assessment—Structural Contraction and Profit-Driven Transformation

According to the latest supply chain data, the global NAND Flash industry is undergoing an unprecedented "structural capacity reconfiguration." The era of aggressive expansion for market share has ended, replaced by a strategic focus on profit-oriented capacity optimization.

Analyzing capacity projections from Q4 2022 to 2027, total global NAND wafer starts show a significant structural decline. The most notable shift occurs within the leading "Korea & USA" bloc, comprised of Samsung, SK Hynix (including Solidigm), and Micron. The combined capacity of these three core manufacturers peaked at approximately 1,100K/m in Q4 2022. Following aggressive production cuts to 650–680K/m in 2023, capacity recovered to 950–980K/m by Q4 2024. However, forecasts indicate a long-term stabilization at a lower level of 670–700K/m through 2026 and 2027. This represents a substantial contraction of nearly 40% compared to the 2022 peak for the world's primary supplier group.

The core reasons for this capacity reduction in the Korean and American bloc lie in capacity crowding effects and the reallocation of capital expenditure (Capex):

• Capacity Conversion: These Korean and American IDMs possess both DRAM and NAND technologies. Driven by the surge in AI-related demand and the high margins of HBM (High Bandwidth Memory) and advanced DRAM, these giants are proactively converting existing NAND cleanroom space and equipment to produce high-profit DRAM or repurposing them directly into HBM production bases.
• Capex Crowding: Since HBM expansion requires expensive TSV (Through-Silicon Via) equipment and advanced packaging capacity, NAND expansion budgets have been severely compressed under limited total Capex.

In contrast, China (including YMTC and SMIC) exhibits a completely different expansionary trend. Data shows that capacity in China will grow counter-cyclically from 100–110K/m in Q4 2022, and is expected to double to 210–220K/m by 2026–2027. This indicates that, driven by non-market factors and localization policies, Chinese manufacturers are filling the mid-to-low-end market gaps vacated by international giants.

II. Market Cycle Review and the Paradigm Shift in Supply Strategy

Reviewing the fluctuations of the NAND Flash industry over the past five years, we have witnessed a complete cycle from extreme optimism to inventory correction.

• 2018–2022: The Optimistic Expansion Phase: Expansion was initially driven by long-term optimism regarding Cloud and AI servers in 2018. The COVID-19 pandemic then triggered strong consumer demand for Work-From-Home (WFH) and Learn-From-Home (LFH), pushing a wave of capacity expansion from 2020 to 1H22.
• 2022–2023: Inventory Tsunami and Production Cuts: As the pandemic-driven demand subsided, a sharp drop in demand combined with inventory accumulation led to a "great wave of production cuts" from 2H22 to 2023. Prices plummeted, severely impacting the balance sheets of all memory original equipment manufacturers (OEMs).
• 2024–2025: Recovery and Conservatism: In 2024, pricing gradually recovered thanks to AI demand (largely driven by xAI), with prices nearly doubling in recent months. However, suppliers remain calm. Despite a slight recovery in early 2025, major suppliers are maintaining existing cuts or planning further ones to ensure financial security following the heavy losses of 2023.

A unique phenomenon has emerged: although 2026 capacity is almost fully booked due to AI and server demand, suppliers remain conservative regarding new production lines. The industry consensus is to meet demand through Technology Node Transition to increase bit density (Gb/mm²) rather than blindly building new wafer fabs. This explains why the market is expected to remain in balance despite the decrease in wafer input.

III. Expansion Plans of the "Big Three"—Quality Over Quantity

A detailed look at the expansion schedules of the three major memory manufacturers reveals that new capacity will not significantly open up until after 2027, confirming a "short-term supply constraint" pattern.

1. Samsung — Balancing Capacity and Profit: As the market leader, Samsung’s expansion has slowed significantly.
   • Pyeongtaek P5 Plant: Construction began in 1Q26, but production ramp-up is not expected until mid-2028, with High Volume Manufacturing (HVM) in 2029.
   • Strategic Implication: Given Samsung's share of Korea's total capacity, the new P5 capacity may simply serve to replace NAND capacity lost to DRAM conversion rather than representing a net increase.
2. SK hynix — All-in on AI Memory: SK hynix's strategy is the most aggressive toward AI, explaining the sharpest decline in projected NAND wafer input.
   • M15X Plant: Opened in 4Q25, its core mission is DRAM and HBM rather than NAND, targeting the demand from AI chip giants like NVIDIA.
   • Yong-in Plant: Expected to be completed in 2Q27, this will be the heart of its future semiconductor cluster, though it remains more than two years away from contributing actual output.
3. Micron — Long-term Layout for U.S. Manufacturing: Micron's plans align with the U.S. CHIPS Act, with relatively late timelines.
   • Idaho Plant: Targets production in 2H27.
   • New York Plant: Targets production in mid-2028.
   • Data Comparison: Micron's capacity will remain at a low of 120–140K/m until 2027, indicating strict supply discipline until these two new plants are completed.

IV. Future Outlook and Conclusion

Based on the data and manufacturer dynamics, we propose the following outlook for the NAND Flash industry over the next three years:

1. Supply-Side "Binary Divergence": The industry will develop in two directions. Korean and American giants will continue to reduce or freeze NAND wafer input to focus resources on high-margin HBM and advanced DRAM. Meanwhile, Chinese manufacturers (YMTC) will be the only players significantly increasing physical wafer capacity to expand influence in mature nodes and consumer markets.
2. A New Normal for Pricing and Profit: Since new wafer fab capacity won't arrive until after 2027 and current capacity is being diverted to DRAM, the NAND market will enter a state of structural "tight supply-demand balance" during 2025–2026. Suppliers will strive to keep prices high to repair financial damage. For buyers, the era of "price collapses" due to oversupply is unlikely to return soon.
3. Technology Upgrades Over Scale Expansion: Future growth will rely on "building taller" rather than "building more." Suppliers will accelerate 3D NAND stacking (200, 300, or even 400+ layers) to increase bit output per wafer. However, this transition will be gradual and cautious to avoid repeating the mistake of inventory surplus.

In summary, the NAND Flash industry has bid farewell to the era of "scale dividends" and entered a new cycle characterized by "Profit First, Capital Discipline, and AI Orientation." For downstream system and module manufacturers, securing long-term supply contracts and understanding the capacity crowding-out effect will be the most critical supply chain management tasks for the next two years.