Global memory industry: Memory shortage persists through 2027, but China and a 2028 normalization temper the industry outlook

Overview

This global memory update revises industry and company models for HBM, DRAM and NAND. Bernstein sees a still-elevated cycle through 2027, increasingly supported by contracted demand and capital returns, but expects a gradual pricing and margin normalization from 2028. The report combines this cyclical view with a structural warning that Chinese suppliers, especially YMTC in NAND and CXMT in DRAM, are becoming materially more competitive.

Core views

Bernstein’s central industry view is that the memory shortage remains intact through CY2027, but the rate of price appreciation should slow. It forecasts conventional DRAM and NAND ASPs to increase by high-teens to 20% quarter-on-quarter in 3QCY26, followed by mid- to high-single-digit growth in 4QCY26. Conventional DRAM prices have quadrupled since 3QCY25. However, long-term agreements (LTAs) are expected to impose price ceilings, while PC and mobile customers reduce unit shipments and memory purchases as prices rise. Thus, the report expects only mild price gains in CY2027 despite continued tight supply. The HBM outlook is revised more cautiously for CY2027. Bernstein’s July and August Korean export trackers showed stronger signals for Samsung but weaker signals from SK hynix’s HBM production location, while external reports pointed to HBM4 supply difficulties. The firm therefore trims SK hynix’s HBM price forecast, delays the HBM3E-to-HBM4 transition, and expects Samsung to gain HBM share through better HBM4 performance and greater capacity. It still expects HBM ASPs to rise in 2027, but less than previously forecast, and expects HBM to be less profitable than conventional DRAM that year. Its Rubin Ultra assumption also changes: half of chips are modeled with 8-high HBM and half with 12-high HBM, reducing blended HBM content from 1,024GB to 640GB. Bernstein believes that capacity released by this mix shift can support more conventional server DRAM, leaving total memory demand above supply and the cycle elevated. For CY2028, Bernstein retains its normalization thesis but makes the trajectory more gradual because supply-demand conditions appear more resilient than in its prior revision. More capacity is expected to reach the market, while AI infrastructure spending may encounter higher borrowing costs, safety concerns, environmental concerns and job-loss backlash. The report also sees possible government intervention if consensus expectations of roughly US$1.5-1.6T in annual memory-industry revenue and US$1.2-1.3T in operating profit are realized. Bernstein forecasts supplier gross margins to normalize from about 90% at the peak to high-70% levels for DRAM and mid-60% levels for NAND exiting CY2028, still above prior-cycle peaks. The report argues that the next stock catalyst is less likely to be large earnings revisions, as consensus has broadly caught up with the cycle. Instead, it emphasizes capital-return clarity and the durability of LTAs. Most suppliers expect at least 50% of revenue or capacity eventually to be covered by LTAs, with pricing mechanisms and, in some cases, financial guarantees. Bernstein regards the guarantees as important evidence of contract enforceability. It also highlights strong free cash flow and announced returns: Samsung has announced a KRW15T buyback and KRW30T cash dividend for 3Q26, SK hynix a KRW40T buyback, and KIOXIA a JPY800B buyback. Greater clarity on returns and LTA backing is expected to support valuations. KIOXIA remains the report’s principal negative call. Bernstein’s SOTP framework indicates 10% downside despite a raised JPY49,000 target price, and it considers YMTC’s development a larger-than-expected long-term threat to KIOXIA and NAND generally. As of 1QCY26, YMTC had overtaken Micron, KIOXIA and SanDisk in NAND revenue and was only slightly below SK hynix. YMTC’s gross margin has exceeded the peer average since CY2024 and may have been the highest in 1QCY26, which Bernstein interprets as evidence that its technology, pricing and costs are approaching or exceeding incumbent levels. KIOXIA’s JPY800B buyback could nevertheless be a near-term positive catalyst, although the report notes that expected FQ2 free cash flow would be needed to fund it while maintaining net cash. CXMT is Bernstein’s structural China opportunity. The firm forecasts capacity to surpass Micron’s and reach 530K wafers per month by end-CY2028, despite uncertainty around export controls. It says CXMT’s G4 node remains roughly three to four years behind global leaders in bit density, but is already beyond what export rules allow; G5, announced for mass production on September 20, 2026, has active-area half pitch similar to the latest-generation nodes of peers. Bernstein expects CXMT’s global bit and revenue share to rise from 9% and 10% in CY2026 to 12% and 13% in CY2028, implying 42%-58% DRAM self-sufficiency for China’s OEMs. HBM remains more challenging, but localization of AI-compute infrastructure is expected to create demand able to absorb yield and cost difficulties. Bernstein forecasts CNY4.0, CNY6.8 and CNY6.0 EPS for CY2026-28, 20%-50% above consensus in CY2026-27, and values CXMT at 10x blended forward 5Q-8Q EPS for a CNY70 target. Across incumbent suppliers, Bernstein maintains Outperform on Samsung, SK hynix and Micron, preferring Samsung for its expected HBM4 progress, capacity-driven HBM share gains and anticipated Samsung Foundry recovery. It maintains Samsung’s KRW440,000 target and Micron’s US$1,300 target. SK hynix’s target is cut from KRW3,300,000 to KRW2,700,000 and its valuation multiple from 6.2x to 5.1x forward 5Q-8Q EPS because of slower HBM4 progress and a smaller assumed HBM price increase, although Bernstein still cited 52% upside as of September 28. Micron’s forecasts are broadly raised because of its greater conventional-DRAM exposure. SanDisk remains Outperform with an unchanged US$3,000 target, while Bernstein notes risks from high near-term expectations, cyclical NAND weakness and potentially structural impairment of NAND asset value.

Analysis framework

Bernstein updates bottom-up HBM, DRAM, NAND and company models using contract-price signals, Korean export trackers, company disclosures, industry data and its own supply-demand estimates. It then links pricing, capacity, technology progress, market share, margins, free cash flow and capital-return policies to earnings forecasts and valuation outcomes. Valuation is primarily based on forward P/E for most covered suppliers, with a sum-of-the-parts framework for KIOXIA.

Methodology notes

• Industry AnalysisSupply-demand framework

  Memory supply-demand and price-cycle analysis

  The report models shortages, capacity additions, customer demand and LTA price ceilings to forecast DRAM and NAND pricing, margins and the timing of cycle normalization.

• Industry AnalysisUpstream-Midstream-Downstream Transmission

  HBM capacity allocation versus conventional DRAM supply

  Bernstein assesses how HBM specifications and capacity allocation affect conventional server DRAM availability, supplier mix, revenue and profitability.

• Valuation methodsP/E and PEG Valuation

  Forward P/E valuation

  Samsung, SK hynix, Micron, CXMT and KIOXIA price targets are derived using stated forward EPS and target P/E multiples.

• Valuation methodsSOTP (Sum-of-the-Parts) Valuation

  KIOXIA sum-of-the-parts valuation

  The report separates KIOXIA’s non-LTA business, LTA-covered business, LTA guarantee payments and excess cash to estimate equity value and downside.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Samsung Electronics (005930.KS)

  Preferred incumbent memory supplier; expected to gain HBM share through better HBM4 progress and capacity.

  Strengths

  Expected HBM4 performance, available capacity, memory-market leadership and potential Samsung Foundry profit recovery.

  Weaknesses

  EPS forecast is slightly reduced because presumed employee-bonus costs are higher than expected.

  Comparison

  Bernstein considers Samsung’s market capitalization being similar to SK hynix’s unreasonable given its memory leadership and non-memory operations.

  Risks

  Earlier end to favorable memory pricing, weaker demand, greater supply or lower valuation support.

• SK hynix (000660.KS)

  Covered incumbent memory supplier rated Outperform.

  Strengths

  Continued elevated memory-cycle exposure.

  Weaknesses

  Slower HBM4 progress and lower expected HBM price increases reduce CY2026-27 forecasts and valuation.

  Comparison

  Bernstein expects some HBM share shift from SK hynix to Samsung.

  Risks

  HBM4 delays, limited HBM price upside and a faster reversal in DRAM supply-demand.

• Micron (MU)

  Covered incumbent memory supplier rated Outperform.

  Strengths

  Relatively greater conventional-DRAM exposure leads Bernstein to raise forecasts broadly.

  Comparison

  YMTC overtook Micron in NAND revenue as of 1QCY26 according to the report.

  Risks

  Earlier end of favorable pricing, weaker demand, higher supply, investor-sentiment pressure and China’s memory progress.

• KIOXIA Holdings (285A.JP)

  Covered NAND supplier rated Underperform.

  Strengths

  JPY800B buyback and possible further capital-return announcements may provide near-term support.

  Weaknesses

  Long-term NAND outlook is pressured by YMTC’s revenue scale, margin improvement and competitive progress.

  Comparison

  YMTC had overtaken KIOXIA in revenue by 1QCY26 and was assessed as increasingly competitive with incumbents.

  Risks

  YMTC-driven supply and NAND pricing pressure; Bernstein’s SOTP framework indicates 10% downside.

• CXMT (688825.CH)

  Covered DRAM supplier rated Outperform following initiation.

  Strengths

  Rapid capacity expansion, improving nodes, localization demand and earnings forecasts above consensus for CY2026-27.

  Weaknesses

  HBM capability remains well behind leaders; G4 bit density remains three to four years behind global leaders, while yields and costs remain to be proven.

  Comparison

  CXMT is valued as cheaper than other A-share semiconductor stocks and is forecast to surpass Micron in capacity by end-CY2028.

  Risks

  Export-control uncertainty and sizeable post-IPO lock-up expirations in January and July 2027.

• SanDisk (SNDK)

  Covered NAND supplier rated Outperform by Bernstein’s US IT Hardware team.

  Weaknesses

  Near-term estimates appear high and investor communications have been confusing.

  Comparison

  YMTC had surpassed SanDisk in NAND revenue by 1QCY26 according to the report.

  Risks

  Cyclical and potentially structural NAND weakness could reduce DCF value and impair asset value below replacement cost.

Key data

• 3QCY26 DRAM and NAND ASP changeHigh-teens to 20% QoQBernstein forecast for both conventional DRAM and NAND.
• Memory-cycle normalizationCY2028Expected to begin as additional supply arrives and AI investment meets boundary conditions.
• DRAM gross margin exiting CY2028High 70s%Down from approximately 90% peak margins, but above previous-cycle peaks.
• NAND gross margin exiting CY2028Mid-60s%Bernstein’s normalized industry-margin outlook.
• CXMT global revenue share10% in CY2026 to 13% in CY2028Bernstein forecast.
• CXMT capacity530K wafers per month by end-CY2028Forecast to surpass Micron’s capacity.
• KIOXIA SOTP downside10%Bernstein estimate despite the recent correction.
• SK hynix target priceKRW2,700,000Cut from KRW3,300,000 on slower HBM4 progress and more conservative HBM pricing.

Impact & implications

Bernstein sees the cycle as supportive for memory suppliers through 2027, but expects future share-price progress to depend more on sustainable earnings evidence, LTAs and shareholder returns than on further large earnings upgrades. The report’s structural implication is differentiated: Samsung benefits from expected HBM4 share gains, CXMT benefits from capacity expansion and China localization, while KIOXIA faces a more serious long-term NAND threat from YMTC.

Risks

• An earlier end to favorable memory pricing caused by weaker demand or higher supply could reduce earnings and valuation.
• China’s progress in memory, especially NAND through YMTC, is a downside risk to incumbent suppliers.
• HBM4 supply difficulties and slower progress could weaken HBM pricing, mix and market-share assumptions.
• CXMT faces export-control uncertainty, unproven production yields and costs, and potential lock-up-related market impact in January and July 2027.
• SanDisk faces high near-term expectations, NAND cyclicality, confusing disclosures and a risk of structural NAND weakness.

What to watch

• DRAM and NAND contract pricing, particularly whether LTA price ceilings constrain further increases.
• HBM4 supply progress, shipment timing and relative HBM share trends for Samsung and SK hynix.
• Financial guarantees and coverage levels in supplier LTAs as evidence of contract enforceability.
• Supplier buybacks, dividends and other capital-return announcements.
• YMTC’s revenue, gross-margin and technology progress in NAND.
• CXMT capacity expansion, G5 execution, HBM development and post-IPO lock-up expirations.