AI demand supports a prolonged NAND shortage, while industry trends begin to diverge
AI summary card
AI demand supports a prolonged NAND shortage, while industry trends begin to diverge
Morgan Stanley expects AI-related NAND demand to continue creating a supply gap in 2027, but consumer NAND prices may be nearing their upper limit; from an investment perspective, it prefers memory suppliers with LTA protection and earnings resilience.
- The report expects AI-related NAND demand to grow about 60% YoY in 2027, with the global NAND supply-demand gap at about 9%.
- Server and AI demand remains strong, and LTAs provide suppliers with downside price protection while improving earnings visibility.
- Smartphone and PC customers on the consumer side are under pressure, with inventories rising at some channels and module makers; consumer NAND prices may soon peak.
- Tactically, it prefers DRAM over NAND; within NAND, it prefers suppliers over module makers.
- The key variables for 2028 are AI SSD demand growth and the capacity expansion discipline of manufacturers such as YMTC; if greenfield expansion accelerates, it could create oversupply risk.
Report interpretation
Overview
This report updates the global NAND supply-demand model and runs scenario tests on 2028 AI demand growth and incremental capacity expansion. The core conclusion is that AI is still reshaping the memory cycle: by 2027, AI NAND demand and server-side eSSD demand will keep the industry in shortage; however, inventory pressure is rising in consumer smartphones, PCs, and channels, creating a divergence between AI and consumer demand. The report believes memory suppliers remain the better allocation direction. Although module makers benefit from improved pricing and margins, volume growth is constrained and cyclical risk is higher.
Core views
First, AI has become the primary incremental source of NAND bit demand, with AI NAND demand expected to reach 609EB in 2027, accounting for about 41% of total NAND demand. Second, global NAND demand in 2027 is expected to be 1,484EB, while supply is 1,347EB, implying a supply-demand gap of about 9%. Third, LTAs improve supplier earnings visibility and support valuation expansion and shareholder returns. Fourth, consumer NAND and DRAM customers face weakening price affordability, with inventories rising at module makers and distributors, leaving limited room for further price increases on the consumer side. Fifth, 2028 supply and demand will depend on whether AI SSD demand can sustain high growth and whether manufacturers such as YMTC maintain capacity expansion discipline.
Analysis framework
The report uses a top-down global NAND supply-demand model, combined with assumptions for ASIC/GPGPU shipments, AI rack eSSD configurations, CSP inventory buffers, smartphone and PC demand, enterprise SSD growth, and NAND supply bit growth forecasts. It then conducts scenario tests using 2028 AI SSD demand growth rates and YMTC potential capacity ranges, and validates changes in pricing, inventories, and customer affordability through channel checks.
Methodology notes
Split NAND demand by AI and non-AI applications, and compare it with industry supply bit growth.
The model aggregates AI NAND demand, non-AI SSD, PC SSD, enterprise SSD, smartphones, tablets, flash cards, USB drives, and other applications into total demand, then calculates the sufficiency ratio against supply forecasts.
Use AI SSD demand growth and YMTC capacity expansion to test 2028 supply-demand balance.
The report sets scenarios such as AI SSD demand growing 30%-60% YoY and YMTC capacity at 310-470kwpm to assess the risk of shortages persisting or shifting to oversupply.
Validate cycle positioning through supply-chain feedback on pricing, inventories, and customer bargaining.
Channel feedback shows server products remain tight, but inventories of consumer memory products are relatively high at some Chinese distributors, and customer acceptance of further price increases is declining.
Adjust target prices based on earnings upgrades, margin assumptions, and business opportunities.
The report raises target prices for SIMO, Longsys, and Phison, reflecting stronger pricing and margin assumptions as well as AI boot drive and enterprise SSD opportunities.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- NAND suppliersCore beneficiaries
- Strengths
- Strong AI eSSD demand, tight supply, LTAs provide downside price protection, and visibility for profits and capital returns improves.
- Weaknesses
- Compared with DRAM, NAND has weaker LTA terms and demand visibility.
- Comparison
- Within NAND, the report prefers suppliers over module makers.
- Risks
- AI capex slowdown, accelerated new capacity in 2028, and consumer-side price peaks.
- DRAM suppliersRelatively preferred allocation
- Strengths
- Better LTA terms, higher demand visibility, EUV constraints support supply discipline, and HBM4E may create capacity squeeze.
- Weaknesses
- The YoY price change rate in 4Q26 may enter a plateau, and near-term catalysts may be lacking.
- Comparison
- Tactically, the report prefers DRAM over NAND.
- Risks
- AI demand slowdown, price gains peaking, and valuations already pricing in the upside.
- Module makersCyclical beneficiaries but with differentiated elasticity
- Strengths
- Low-cost inventories, rising prices, and product mix improvement support margins; Longsys's TCM model may stabilize long-term profits.
- Weaknesses
- Suppliers shifting capacity toward CSPs constrains volume growth, and the inventory advantage may be exhausted in 2026.
- Comparison
- Compared with suppliers, module makers' profits and valuations are more affected by cycles and inventory accounting.
- Risks
- Weak consumer SSD demand, limited raw NAND support, limited eSSD contribution, and customers procuring directly from NAND makers.
- SIMOBeneficiary of AI boot drive and eSSD controller opportunities
- Strengths
- Strong share in Blue Field 3 boot drive controllers, while Vera Rubin, general-purpose servers, and ASIC racks create incremental TAM for boot drives.
- Weaknesses
- Rising raw NAND costs may pressure boot drive gross margins.
- Comparison
- The report believes SIMO is better positioned than Phison to benefit from the boot drive controller opportunity.
- Risks
- Customer ramp timing, market share assumptions, NAND costs, and enterprise SSD ramp falling short of expectations.
- PhisonBeneficiary of pricing improvement but with limited long-term contribution
- Strengths
- Preliminary 2Q26 results beat expectations, and 3Q26 revenue and margins still have support.
- Weaknesses
- Consumer demand is weakening, gross margin may normalize after low-cost inventories are depleted, and eSSD revenue contribution is limited.
- Comparison
- The report raises the target price but is more cautious on its long-term addressable market and volume growth.
- Risks
- 4Q26 revenue may decline about 20% QoQ, consumer SSD demand is weak, and hyperscaler direct procurement limits room for module makers.
Key data
- 2027 AI NAND demand609EBThe report estimates AI-related NAND demand will grow about 60% YoY.
- 2027 global NAND supply-demand gap-9%Total demand is 1,484EB and supply is 1,347EB.
- 2026 global NAND supply-demand gap-15%Total demand is 1,250EB and supply is 1,058EB.
- AI share of NAND demand in 202741%It was 18% in 2025 and 32% in 2026.
- 2027 NAND supply bit growth27% YoYBased on forecasts from U.S. SPE analysts for 2026-2027 supply.
- 3Q26 TLC eSSD price trend+30% QoQThe increase in server-grade products is significantly higher than in consumer NAND.
- 3Q26 server-grade DRAM price trend+20% QoQTraditional DRAM such as DDR3/4 rose about 30%-40% QoQ due to tight supply.
- KIOXIA CY27 LTA coveragemore than 50%CY28 coverage is expected to be about 50% to preserve flexibility.
- SIMO target priceUS$400Raised from US$155, reflecting eSSD and boot drive opportunities.
- Longsys target priceRmb673Raised from Rmb300.
- Phison target priceNT$2,588Raised from NT$2,248.
Impact & implications
In investment terms, the report remains positive on the memory cycle but emphasizes structural divergence. Suppliers benefit from AI demand, low inventories, LTAs, and stronger capital returns, with better earnings visibility than in traditional cycles; module makers benefit from low-cost inventories and rising prices, but their inventory advantage will diminish over time, and volume growth is affected by suppliers shifting toward CSP customers. Over the long term, they need to rely on product mix upgrades and value-added services to improve valuation. For 2028, if AI capex continues and players such as YMTC maintain capacity expansion discipline, NAND may still remain tight; if both China restrictions and supply discipline ease at the same time, there is a risk of oversupply.
Risks
- A slowdown in AI capex could weaken server and eSSD demand.
- Accelerated greenfield expansion by YMTC or other manufacturers in 2028 could lead to NAND oversupply.
- Consumer NAND prices are nearing their ceiling, and smartphone and PC customers have reduced tolerance for further price increases.
- Rising inventories at module makers and distributors could create destocking pressure.
- While LTAs improve visibility, they may also limit suppliers' upside pricing elasticity during strong cycles.
- After low-cost inventories are depleted, module makers' margins may fall back from high levels.
What to watch
- Whether consumer NAND and DRAM prices continue rising or peak in 2H26.
- Changes in module maker inventories, shipments, and gross margins in 3Q26 and 4Q26.
- Whether AI SSD demand can sustain 30%-60% YoY growth in 2028.
- The actual production ramp timing of YMTC Fab4, Fab5, and other announced capacity.
- The results of LTA negotiations and coverage ratios between major CSPs and memory suppliers.
- The rollout progress of boot drives for NVIDIA Vera Rubin, BlueField 4, ASIC servers, and general-purpose servers.
- Capex and shareholder return policies of suppliers such as Micron, SK hynix, Samsung Electronics, KIOXIA, and SanDisk.