Morgan Stanley: “Memory Supercycle”
- Transition signal and ’26. Judgment of supply-demand imbalance.
- Reducing HBM’s margin width and changing market share are short-term burdens. However, it focuses on the trajectory of cycle turning points and profit estimates. Now, it is considered as a cyclical upward trend rather than a double-dip, and the risk of HBM is gradually reduced. The current tariff-led cyclical upward trend will accelerate further in ’26 and stock prices often fall before the first or second quarter of year-on-year price improvement. Oversupply of DRAM will become more balanced through ’26 years on the back of inventory depletion, and NAND will turn into a shortage as AI eSSD demand doubles next year.
- The upward phase of valuation needs to be judged from a long-term perspective. The rate of change in general-purpose memory prices has accelerated again against the backdrop of AI-related server and mobile DRAM demand. It believes that cycle indicators are now heading for a “cycle peak” by “27,” not a short-term slowdown.
- Structural Changes in the Memory Industry: “Lack” Everywhere
- We recently saw a surge in high-density NAND orders for delivery in ’26 years from the US Hyperscaler. The size of this order alone could exceed the overall eSSD market this year, and it is likely to spread to other areas of the NAND market given the lack of investment and limited wafer capacity in recent years. The current capital allocation to NAND suggests a significant improvement in ROIC across the cycle, after years of production cuts.
- Now, we don’t see DRAM prices being adjusted towards the end of the year, instead expecting mixed ASPs to rise 9% in Q4 2025 (previous outlook remains flat), based on urgent orders from cloud servers. Note that there are a number of new drivers of AI-driven memory strength, including SOCAMM, GDDR7, and HBM logic-based dies.
- What Has Changed: Five Surprises
1) High-density QLC eSSD demand stands out, ’26-year new order size may be equivalent to this year’s total eSSD demand
2) Demand is rebounding across the new AI-led end market, especially with regard to B40 (or RTX 6000), ‘there are approximately 2 million GDDR7s heading to China in the second half of ’25, which could increase further in ’26. Also, SOCAMM demand drives LPDDR5x with Rubin GPU launch.
3) DDR5 RDIMM server demand has significantly increased since the start of this quarter, likely driving price increases in Q4 of ’25
4) This is likely to affect customer behavior in other segments of the DRAM market, due to supply allocation and sub-normal inventories
5) DRAM facility investment is likely to increase significantly in ’26 due to HBM4 and supply tightening factors - Greater China Semiconductors
1) DDR4 shortage likely to continue into first half of 2026
- Starting in 2026, major DRAM companies will provide minimal DDR4 assistance to their core customers and focus more on DDR5 and HBM. By ’25 DDR5 is expected to account for about 80% of the market. In addition, memory companies have incentives to maintain high prices of DDR4 given the loss of DDR4 and its push to convert DDR5 (estimated to decrease by 47% in ’25/’51% in ’26, from smartphones, servers and PCs)
2) NAND Market Continues to Grow in ’26
- ‘NAND supply is expected to only increase by 15-20% year-on-year in 26. On the other hand, historical records show that demand for NAND tends to grow by 30% year-on-year on average, which could lead to a shortage of about 10% to 15%, which could further drive the rise in NAND prices.
3) HBM Base Die: Additional Growth Drivers For Foundry And ASIC Services Companies
- The HBM4/4e base die could be the next key growth engine for foundry and ASIC design services companies. ‘Estimated total foundry TAM will reach at least $2 billion in 28 years. Assuming that one-third of base dies go through design services, approximately $1 billion in TAM is possible.