AI demand is supporting the continuation of NAND shortages, with industry performance becoming increasingly divergent
AI summary card
AI demand is supporting the continuation of NAND shortages, with industry performance becoming increasingly divergent
Morgan Stanley believes that AI-related storage demand will keep NAND supply tightness in place through 2027, but consumer-side pricing is approaching a ceiling, and the pace of supply expansion in 2028 will determine whether the market shifts into oversupply.
- The report expects AI-related NAND demand to grow 60% year-over-year in 2027, implying an approximately 9% supply-demand gap; the shortage is expected to remain significant.
- Suppliers are still outperforming module makers, because LTAs provide downside protection on pricing, visibility on profitability, and shareholder return potential; tactically, DRAM is preferred to NAND.
- Consumer-side performance is diverging: smartphone, PC, and distribution-channel inventories are rising, order cuts have started after the 2Q26 price increase, and mainstream consumer product pricing may hit a near-term peak quickly.
- The key variable for 2028 is the pace of greenfield capex by YMTC and other suppliers; if AI capex continues to grow and expansion remains disciplined, NAND should remain relatively tight; if expansion accelerates, oversupply risk increases.
- SIMO was significantly upgraded due to AI boot drive and enterprise SSD upside, while Longsys and Phison target prices were also increased because of improved pricing and margin assumptions.
Report interpretation
Overview
This report updates the global NAND supply-demand model and outlines expectations for 2026-2028, covering AI storage demand, supply expansion, long-term LTA agreements, consumer-side inventory pressure, and changes in module-maker business models. The core conclusion is that AI is still creating NAND scarcity, with that scarcity expected to persist into 2027; however, consumer pricing and demand are already under pressure, and the industry outlook is shifting from broad upside to divergence across suppliers, the AI chain, and module makers.
Core views
The report believes the memory cycle still has an upside foundation, and that suppliers are more attractive because of low inventories, LTA protection, capital return, and visibility into AI demand. In contrast, module makers benefit from prior price rises and low-cost inventory releases, but face constraints from allocation shifting toward CSP, weak consumer SSD demand, limited OEM support, and rising inventory carrying costs. After 2028, the supply-demand direction will depend on AI SSD demand growth, YMTC capacity ramps, and whether other greenfield capacity additions remain disciplined.
Analysis framework
The report combines a global NAND supply-demand model, AI server and ASIC/GPGPU shipment assumptions, enterprise SSD capacity assumptions, consumer electronics demand assumptions, supplier-side WFE and wafer-capacity forecasts, a 2028 scenario test, channel checks, and company-level earnings and valuation adjustments to assess the cycle, price elasticity, the ranking of beneficiaries, and target-price changes.
Methodology notes
Estimate the supply-demand gap based on AI NAND demand, non-AI demand, and supplier bit-growth.
The report uses ASIC/GPGPU shipments, eSSD deployment, smartphone and PC demand, enterprise eSSD growth, and supplier-side bit-growth forecasts to judge that there is still an estimated 9% supply-demand gap in 2027.
Use AI SSD year-over-year growth and YMTC capacity ranges to test 2028 shortage versus oversupply risk.
The base case assumes non-AI NAND demand growth of 5%, AI SSD demand growth of 30% to 60%, and YMTC capacity ranging from 310 kwpm to 470 kwpm; if expansion discipline weakens, oversupply risk rises.
Validate the supply-demand model through price, inventories, and customer acceptance.
Checks show 3Q26 TLC eSSD prices rose approximately 30% quarter-over-quarter; server demand remains tight, while consumer NAND inventories rose and clients showed resistance to further price increases.
Estimate boot drive module and controller opportunities using AI rack, GPU/ASIC server, and general server shipment data.
The report argues that while boot drives are not the main memory for training or inference, they are essential for system boot, control-plane functions, logging, telemetry, and recovery, making them a stable and scalable AI storage demand source.
Support target-price revisions using 2027e EPS or P/E multiples, company revenue contribution, and earnings adjustments.
SIMO target price was raised to US$400, corresponding to 23x 2027e EPS; Longsys target price was raised to Rmb673, corresponding to 15x 2027e P/E.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- NAND supplierscore beneficiaries
- Strengths
- AI demand, low inventories, LTA downside protection, tightness in supply-demand, and capital returns support earnings visibility.
- Weaknesses
- If greenfield expansion accelerates in 2028 or AI capex slows, supply-demand could turn loose.
- Comparison
- Compared with module makers, they have greater margin resilience; however, the report tactically views DRAM terms and demand visibility as superior.
- Risks
- YMTC and other suppliers’ expansions, AI demand below expectations, and peaking price cycles.
- DRAM supplierstactically preferred over NAND
- Strengths
- Better LTA terms, higher demand visibility, and supply discipline constrained by EUV and HBM4E capacity pressure.
- Weaknesses
- YoY price growth may plateau in 4Q26, with fewer short-term cyclical catalysts.
- Comparison
- The report clearly states that tactically DRAM is preferred over NAND.
- Risks
- AI capex slowdown and a deceleration in price uptrend slope.
- Module makerscyclical beneficiaries, but outcomes diverge
- Strengths
- Low-cost inventory, price increases, and product-mix upgrades can lift EPS, and longer-term margins could stabilize.
- Weaknesses
- Depletion of low-cost inventory, supply shifting toward CSP by suppliers, weak consumer demand, and constrained sales growth.
- Comparison
- Compared with suppliers, margin and volume outcomes are more sensitive to raw material costs and allocation decisions.
- Risks
- Weak consumer SSD demand, inventory pressure, weakening price support, and potential oversupply in 2028.
- Silicon Motion (SIMO.O)beneficiary from AI boot drive and enterprise SSD
- Strengths
- Boot drive controllers, MonTitan enterprise SSD, and rising share in mobile and PC drive revenue mix toward more AI-related areas.
- Weaknesses
- Rising raw NAND costs could compress boot drive gross margins, and execution is still needed during the scaling phase.
- Comparison
- In the boot drive TAM, SIMO is relatively more favored than Phison; the report expects AI boot drive revenue contribution of 15%/21% for 2026e/2027e.
- Risks
- CSP adoption progress, enterprise SSD customer expansion, AI platform shipments, and competition.
- Longsys (301308.SZ)valuation-upside module-maker candidate
- Strengths
- Better pricing, a TCM business model, in-house controller development, and product-mix improvements support more stable long-term margins.
- Weaknesses
- Supplier capacity shifting to CSP customers may limit near-term sales growth.
- Comparison
- Its target price was raised while rating remains EW, indicating earnings upgrades coexist with structural constraints.
- Risks
- Consumer demand, NAND supply constraints, inventory, and margin compression.
- Phison (8299.TWO)cyclical beneficiary with uncertain sustainability
- Strengths
- Stronger 2Q26 performance than expected, with 3Q26 revenue and margins supported by pricing and supply.
- Weaknesses
- Upside is more cyclically driven; after low-cost inventory is used up, margins are expected to normalize.
- Comparison
- Boot drive revenue contribution is smaller than SIMO, and the report sees it remaining relatively limited through 2027e.
- Risks
- Weak consumer technology demand, limited OEM support, limited eSSD contribution, and CSP buying directly from NAND suppliers.
- KIOXIAtop pick in Japan semiconductor
- Strengths
- AI storage product breadth, free cash flow, shareholder returns, and LTA coverage support the investment case.
- Weaknesses
- Need to balance capital allocation between growth investment and shareholder returns.
- Comparison
- Listed as the Japan team’s Top Pick among NAND suppliers.
- Risks
- ASP assumptions, capacity discipline, AI demand, and customer LTA execution.
Key data
- 2027 AI-related NAND demand growth60% YoYThe report estimates AI-related NAND demand growth would generate a roughly 9% supply-demand gap.
- 2027 NAND supply-demand gapabout 9% shortageBased on the global NAND supply-demand model, the shortage is expected to persist into 2027.
- 2027 NAND supply bit growth27% YoYBased on SPE analyst forecasts for supply-side output from 2026 to 2027 in the US.
- 3Q26 TLC eSSD price+30% QoQServer and enterprise demand is stronger, while consumer NAND showed a smaller price increase.
- 3Q26 server-class DRAM price+20% QoQTraditional DDR3/4 rose around 30% to 40% due to supply tightness and AI-related demand growth.
- KIOXIA CY27 LTA coveragemore than 50% of shipmentsCY28 coverage is expected to be around 50% to preserve flexibility.
- KIOXIA FY3/27-FY3/28 FCF¥4.0-5.0trnEven under conservative ASP assumptions, there remains strong free cash flow and potential shareholder returns.
- SIMO target priceUS$400, previously US$155Supported by AI boot drive, enterprise SSD opportunities, and improved market share.
- Longsys target priceRmb673, previously Rmb300Supported by stronger pricing and margin assumptions, though sales growth remains constrained by allocation limits.
- Phison target priceNT$2,588, previously NT$2,248Supported by 2026 pricing and profit recovery, but sustainability and consumer-side pressure remain constraints.
Impact & implications
From an investment perspective, the report remains constructive on upstream memory suppliers with AI demand, LTA protection, and shareholder return capacity. Module-makers have upside to earnings revisions, but stock sensitivity is more dependent on product-mix upgrading and new growth drivers from AI boot drive or enterprise SSD. Consumer pricing near the ceiling means the industry is no longer a uniform upcycle, but rather a divergence cycle between AI servers and consumer electronics, between suppliers and module makers, and between DRAM and NAND.
Risks
- The largest upside-cycle risk for memory stocks is a slowdown in AI capex.
- If YMTC or other suppliers accelerate greenfield expansion in 2028, NAND oversupply could emerge.
- Consumer smartphone and PC demand is under pressure, and customers’ acceptance of further price increases is weakening.
- Module maker and distributor inventories have risen, which may increase carrying-pressure on stocks and create a price ceiling.
- LTAs may limit pricing upside, and changes in customer demand can also affect shipment elasticity.
- After low-cost inventory is depleted, module maker gross margins may normalize from elevated levels.
- CSPs may buy directly from NAND suppliers or sign long-term agreements, compressing module makers’ long-term reachable market.
What to watch
- Whether 2028 AI SSD year-over-year growth remains in the 30% to 60% range.
- Actual ramp pace and NAND capacity allocation at YMTC Fab4, Fab5, and subsequent five fabs.
- Whether 3Q26 to 4Q26 consumer NAND pricing is hitting a peak, and whether order reductions are widening.
- Whether demand and tightness for server and enterprise SSDs remain constrained, especially for TLC eSSD and QLC use cases.
- DRAM and NAND LTA negotiation terms, including price caps, floors, and shipment coverage percentages.
- SIMO MonTitan customer onboarding, boot drive share, and 2026-2028 revenue contribution.
- Cash flow, buybacks, and shareholder return execution by suppliers such as KIOXIA, Micron, and SanDisk.