Memory stocks are entering a short-term cooling period, but the structural AI capex bull market is not over yet

Overview

This report discusses the latest controversies in Asia Pacific technology and the South Korean memory industry. The core view is that AI-driven memory demand still has structural support, but the market has begun to pre-price potential marginal changes, including the possibility that hyperscale cloud providers may be selling compute, long-term supply agreements not leading to valuation repricing, and DRAM earnings upgrades and price year-over-year growth approaching historical peaks.

Core views

Morgan Stanley maintains a long-term constructive view, with the rationale that 2027 earnings are expected to grow more than 35%-40%, and that growing Agentic AI volume will continue to drive AI infrastructure spending. However, in the short term, memory stocks are in a crowded trading zone; if hyperscale cloud providers do not provide sufficiently strong 3Q guidance after 2Q26 results, or if capex is not maintained/increased, the market may continue trading the narrative of surplus AI compute and a peak in earnings revisions.

Analysis framework

The report uses investors' recent discussion as the thread and makes a comprehensive assessment around AI compute supply and demand, long-term supply agreements, memory price and earnings-revision cycles, hyperscale cloud guidance, and DRAM/NAND price forecasts, while combining residual income valuation models, P/B cycle-peak valuation and industry risk factor assessments to evaluate stocks.

Methodology notes

• Valuation methodsResidual Income Valuation Model

  Estimates target value using cost of equity, risk-free rate, equity risk premium, beta and perpetual growth rate.

  The base case disclosed in the report uses 11.5% cost of equity, 5% risk-free rate, 6.5% equity risk premium, 1.0 beta and 3% perpetual growth rate; the related target price corresponds to about 2x P/B for 2027e, close to commodity cycle peak levels.

• cycle_analysisMemory Cycle Change-Rate Framework

  Checks whether price growth year-on-year, inventory, and breadth of earnings upgrades are near their peaks.

  The report believes the memory industry remains cyclical; current pricing year-over-year, inventory, and breadth of earnings revisions are near peak levels, so short-term stock prices may consolidate, but this does not mean a new bear market.

• industry_viewMorgan Stanley Industry View

  Attractive means the analyst expects the covered industry to be relatively more attractive than the broad market benchmark over the next 12-18 months.

  This report shows Asia Pacific Industry View for S. Korea Technology as Attractive.

Asset mapping & comparison

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

• DRAM

  Core beneficiary asset

  Strengths

  AI spending and memory bottlenecks directly support DRAM more directly; earnings upgrades are significant.

  Weaknesses

  Price year-over-year growth and earnings-revision breadth are nearing historical highs, so short-term momentum may fade.

  Comparison

  Favored relative to NAND and memory module manufacturers.

  Risks

  Hyperscale capex cuts, intensifying DDR5 competition, and oversupply on the supply side.

• NAND

  Secondary beneficiary asset

  Strengths

  Price forecasts still show sequential increases across multiple categories, with enterprise SSDs and similar segments having demand support.

  Weaknesses

  The report explicitly favors DRAM and legacy memory more than NAND.

  Comparison

  Less favored than DRAM but still affected by the AI and data center cycle.

  Risks

  End demand below expectations, high inventories, and slowing price momentum.

• Samsung Electronics (005930.KS)

  Covered stock / South Korea technology leader

  Strengths

  Beneficiary of the memory cycle and new technology development; the report discloses a current reference price of 318,000.00 KRW.

  Weaknesses

  Its products are tied to the memory cycle, and Apple and Chinese smartphone competition may create pressure.

  Comparison

  Along with SK hynix, a core coverage candidate in the Korean memory and tech chain.

  Risks

  Concentration in semiconductor earnings growth, weakening end demand, and cyclical volatility.

• SK hynix (000660.KS)

  Covered stock / Core memory beneficiary

  Strengths

  Beneficiary of a longer upward memory cycle driven by AI and hyperscale data center growth; the report discloses a current reference price of 2,343,000.00 KRW.

  Weaknesses

  Crowded positioning and high levels of earnings upgrade momentum make short-term stock moves more volatile.

  Comparison

  More directly positioned in the AI-memory beneficiary chain.

  Risks

  High customer inventories, weaker demand, and supply-side competition or overinvestment.

• Memory module makers

  Least favored area

  Strengths

  Can benefit from spillover effects of overall memory price cycles.

  Weaknesses

  The report clearly identifies this as the least favored memory-related area.

  Comparison

  Less favored than DRAM, legacy memory, and NAND.

  Risks

  Margin compression, weaker ability to pass through price, and amplified cyclicality.

Key data

• Report date2026-07-06The report cover page timestamp is July 6, 2026 07:34 PM GMT.
• 2027e Earnings growth expectationOver 35%-40%Used to support the long-term bullish view.
• Samsung Electronics reference price318,000.00 KRWReference price shown in the report.
• SK hynix reference price2,343,000.00 KRWReference price shown in the report.
• 2Q26 NAND price forecastTotal NAND Flash up 55-60%From the latest NAND price forecast in the report tables.
• 3Q26E NAND new forecastTotal NAND Flash up 10-15%Higher than the old forecast of up 8-13%.
• Valuation assumptionsCost of equity 11.5%, perpetual growth rate 3%Assumptions for the base-case residual income valuation model.

Impact & implications

For investment implications, the report suggests not equating short-term pullbacks with the end of the AI memory cycle. The more important validation points are whether hyperscale cloud providers maintain or raise capex in 2Q26 results, and whether AI token demand, open-source model substitution, and chip inflation affect second-half guidance. If capex stays strong, memory stock pullbacks may become a better entry point; if capex is cut, the oversupply compute narrative is likely to persist.

Risks

• AI compute could be in excess, especially if hyperscale cloud providers offload compute and the market interprets that as weak demand.
• If hyperscale cloud providers cut capex or provide weaker-than-expected 3Q guidance after 2Q26 results, memory stock pressure may continue.
• Token minimization, low-cost open-source LLMs, and chip inflation may squeeze AI service providers' margins and willingness to spend on capex.
• DRAM price year-over-year growth, inventory, and breadth of earnings revisions are near high levels, so short-term momentum could peak.
• Crowded positioning may amplify volatility as investors reduce net exposure from historical peaks.
• End demand below expectations, intensifying DDR5 competition, and high inventories at cloud and Chinese smartphone customers.

What to watch

• Whether hyperscale cloud providers maintain or increase capex during 2Q26 reporting season.
• AI token suppliers' 2H26 revenue and margin guidance.
• Whether enterprises continue moving from token maxing to token minimization, and the impact of open-source model substitution on frontier-model demand.
• Spot and contract prices for DRAM and NAND, and year-over-year price changes.
• Whether the breadth of DRAM earnings upgrades falls from historical highs.
• Beyond management teams of memory companies, how the market reacts to AI spenders' results and guidance.