The memory cycle has entered a debated phase of “extension rather than collapse”
AI summary card
The memory cycle has entered a debated phase of “extension rather than collapse”
Morgan Stanley frames the memory investment debate around three lines of analysis—AI capex, LTA re-rating, and the memory cycle position—arguing that DRAM prices may peak in 4Q26, while NAND could remain tight through 2028 if AI spending continues and YMTC maintains disciplined capacity expansion.
- Monetization of AI infrastructure does not equal excess compute capacity, and 2Q26 capex is seen as a key observation point.
- The market’s pricing of memory earnings is described as rational rather than euphoric, and LTAs have not yet driven a valuation re-rating.
- The YoY growth rate of DRAM contract prices is falling from cyclical highs, and the report judges that prices may peak around 4Q26.
- Under the base-case scenario, NAND may still remain tight, but faster greenfield expansion by YMTC is the main oversupply risk.
Report interpretation
Overview
This is a Morgan Stanley Greater China technology hardware research report for Asia Pacific investors, centered on the memory cycle. The report breaks the investment debate into three parts: whether AI spending still supports demand, whether long-term agreements should drive a valuation re-rating, and whether the current memory cycle is before or after the peak or in an extended phase. The material also discusses the impact of YMTC capacity expansion on NAND supply and demand, and includes a scenario-analysis heading on Largan’s FAU opportunity.
Core views
The tone of the report is “constructive but not euphoric.” AI capex remains a key variable supporting memory demand, but the market has not simply re-rated memory earnings based on an optimistic scenario; LTAs have not led to a clear re-rating; and the YoY rate of change in DRAM prices has begun to retreat from high levels, with a price peak possibly appearing around 4Q26. Rather than a cyclical collapse, the report emphasizes a lengthening cycle. For NAND, if AI capex continues to grow and YMTC maintains discipline in capacity expansion, supply and demand could remain tight through 2028; if greenfield expansion accelerates, oversupply risk would emerge.
Analysis framework
The report uses a debate framework rather than a single earnings-forecast framework: it first looks at AI spending indicators such as hyperscaler capex, top-tier LLM intensity, financing, and ARR, then compares LTAs against inventory and renegotiation risks in historical cycles, and finally uses DRAM contract price YoY changes, inventory, and valuation metrics to judge cycle positioning. The NAND section stress-tests supply-demand conditions through YMTC capacity scenarios and growth scenarios for non-AI NAND and AI SSD demand.
Methodology notes
AI spending, LTA re-rating, cycle position
The report breaks memory investment judgment into three debates: where the money is being spent, how the market prices long-term agreements, and what stage the cycle is currently in.
Capacity expansion and AI SSD demand sensitivity
The report compares YMTC’s base to maximum capacity scenarios, as well as 2028 non-AI NAND and AI SSD demand growth, to judge whether NAND will shift into oversupply.
Price peak and valuation re-rating
The report notes that DRAM contract price YoY is retreating from the cycle peak, while NTM P/B has not clearly re-rated, consistent with the view that prices will peak around 4Q26.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- DRAMCore cyclical asset
- Strengths
- Demand from AI servers and high-performance computing may still support a longer cycle.
- Weaknesses
- YoY momentum in contract prices has already retreated from cyclical highs.
- Comparison
- Compared with NAND, the report more explicitly indicates that DRAM prices may peak around 4Q26.
- Risks
- An earlier price peak, rising inventory, and failure of valuation to re-rate.
- NANDSupply-demand scenario asset
- Strengths
- If AI capex continues to grow and YMTC maintains discipline in capacity expansion, supply and demand could remain tight through 2028.
- Weaknesses
- Highly sensitive to AI SSD demand and supply discipline.
- Comparison
- Compared with DRAM, the NAND conclusion depends more on YMTC capacity expansion and AI SSD demand scenarios.
- Risks
- YMTC greenfield expansion faster than expected leads to oversupply.
- Greater China technology hardware equitiesCoverage scope
- Strengths
- Covered companies include multiple hardware chains such as optics, connectors, servers, PCBs, displays, and the smartphone supply chain.
- Weaknesses
- The main body of the report does not focus on deep valuation analysis company by company.
- Comparison
- The industry view is In-Line; single-stock ratings should refer to the latest company research.
- Risks
- Industry demand, AI capex, inventory cycles, and investment-banking conflict disclosures all need to be incorporated into investment judgment.
- Largan PrecisionExample of a stock opportunity
- Strengths
- The report title shows an assessment of the potential positive impact of the FAU opportunity, with bull/base/bear scenario analysis.
- Weaknesses
- The current input retains only title-level information and lacks full valuation and target price evidence.
- Comparison
- In this material, it is part of the discussion on key stocks, but the available body-text evidence is limited.
- Risks
- The extent to which the FAU opportunity materializes and the scenario assumptions are unclear, requiring a reading of the full report.
Key data
- Report Date2026-07-14Date corresponding to the file name and disclosure information.
- Industry ViewAsia Pacific Industry View In-LineAsia Pacific industry view disclosed on the front page.
- DRAM Price ViewMay peak in 4Q26The report says DRAM contract price YoY is retreating from cyclical highs, while valuation has not yet re-rated.
- YMTC Potential Global NAND ShareAbout 24%Assumes all five announced fabs are devoted to NAND, with Fab4 and Fab5 each at about 100kwpm.
- YMTC Capacity Scenario310kwpm to 470kwpmThe report provides a 2028 base-to-maximum capacity range.
- 2028 Non-AI NAND Demand+5% YoYThe report’s base-case demand assumption.
- 2028 AI SSD Demand+30-60% YoYThe AI SSD demand assumption used in the report’s NAND supply-demand scenario analysis.
Impact & implications
For investors, the report suggests that the memory chain should not be treated in a binary way as either “AI demand keeps being revised infinitely upward” or “the cycle is about to collapse.” More important is tracking whether AI capex is delivered, whether DRAM price YoY momentum continues to decline, whether inventory accumulates, and whether YMTC’s capacity expansion breaks discipline. If AI demand holds and supply remains disciplined, the NAND chain is more resilient; if expansion accelerates or inventory pressure rises, cycle turning-point risks will be exposed earlier.
Risks
- A slowdown in AI capex weakens support for memory demand.
- DRAM prices peak earlier than expected or YoY price momentum continues to decline.
- Inventory rises, especially inventory accumulation driven by module makers.
- LTA renegotiation or forced inventory causes the market to withhold a valuation re-rating.
- YMTC greenfield expansion is faster than expected, triggering NAND oversupply.
- The report discloses that Morgan Stanley has or seeks investment banking and other service relationships with multiple covered companies, and investors should consider potential conflicts of interest.
What to watch
- 2Q26 and subsequent hyperscaler capex.
- Changes in top-tier LLM compute intensity, financing, and ARR.
- Whether DRAM contract price YoY and NTM P/B show a re-rating.
- DRAM and NAND inventory levels, especially module-maker inventory.
- The pace of YMTC Fab4, Fab5, and subsequent greenfield expansion.
- Whether 2028 AI SSD demand growth falls within the +30-60% YoY range.