Memory Cycle Still in Uptrend, AI Demand Drives Valuation Re-rating

Overview

Morgan Stanley released an in-depth memory industry report, with the core view that the recent adjustment in memory stock prices is a "healthy reset" rather than the end of the cycle. The report points out that AI-driven structural demand makes this memory cycle longer and peak higher than previous ones, with earnings revisions remaining strong and more sustainable than generally believed by the market. Supported by three pillars: pricing locked via long-term agreements (LTA), supply constraints (cleanroom and EUV availability limits), and growing AI inference demand, the institution is bullish on the memory sector continuing its bull run through the end of 2026, and significantly raised bear case valuations for SK hynix and Samsung.

Core views

Cycle Judgment: The report emphasizes that the current memory cycle is still accelerating rather than peaking. Historical DRAM cycles usually see supply loosen after 6 quarters of uptrend, but due to AI agent demand starting in January 2026, there are at least several more quarters before the cycle top. Unlike previous cycles, this one is driven by AI structural demand rather than traditional consumer electronics, meaning demand is more persistent. Price and Valuation: Memory prices have nearly doubled since February 2026, with lead times significantly extended. The core driver is physical capacity constraints under LTA constraints. The report calculates that if LTA can account for more than 70% of total supply over the next 3-5 years, DRAM stocks can theoretically re-rate from the current 5x PE to 8-10x PE. This is based on a simple calculation of 70%+ of 2027 EPS valued at market multiples (rolling 5-year 10-14x PE) and the remaining 30% non-LTA EPS valued at historical 5x peak PE. Earnings Revisions Support Stock Price: Although memory stocks have risen 70-134% in the past 2 months, earnings forecasts were adjusted upwards synchronously, supporting the stability of the 5.2x 2027 expected PE, rather than changes in discount rate. The report expects 3Q26 DRAM price increase to exceed 20-30%, sufficient to maintain accelerating YoY growth. Target Valuation: Based on the Residual Income Model, the report raised SK hynix bear case valuation by 175% to W1,100,000 (corresponding to 2.4x 2027 expected PE), and Samsung bear case valuation by 58% to W190,000 (corresponding to 3.1x 2027 expected PE). SK hynix base target price W2,600,000 (corresponding to 5.7x 2027 expected PE), Samsung base target price W381,000 (corresponding to approx. 2x 2027 expected PB, consistent with commodity cycle peak 2.0x).

Analysis framework

Supply-Demand Framework Analysis: The report adopts the classic memory cycle supply-demand analysis framework, emphasizing that previous cycles were never a demand problem (except for the 2008 financial crisis and 2021 early pandemic shock), but supply-driven—vendors competed to fill demand gaps and over-expanded. The key to breaking the cycle is supply discipline, but this variable has never really worked. However, this round of AI structural demand combined with relatively restrained supply expansion forms a consensus scenario of a longer cycle and higher peak. LTA Pricing Stability Analysis: The report focuses on analyzing the impact of Long-Term Agreements (LTA) on pricing volatility in the memory industry. Traditional memory pricing is highly volatile, but LTA can lock in high prices through prepayments, improve earnings predictability, and reduce the risk of over-building capacity. The report monitors four cycle turning point signals: continued strength in DRAM prices, inventory adjustment cycle, capital expenditure budget cuts while capital intensity rises, and sustained earnings revisions and valuation expansion. AI Demand Elasticity Analysis: The report distinguishes the essential difference between AI demand and traditional consumer electronics demand. AI inference is growing exponentially, and AI agent usage is in the early stage and far from saturated. Each generation of GPU is limited by memory rather than compute power, with AI unit DRAM content growing 4-7 times (from 80GB for A100 to 288-768GB for Rubin GPU/Superchip). More critically, AI demand has price elasticity—lower DRAM prices reduce AI inference operating costs, making AI deployment cheaper, thereby creating new demand rather than just reducing costs for a fixed number of devices. Valuation Re-rating Logic: The report uses the Residual Income Model for valuation, assuming a cost of equity of 11.5% (5% risk-free rate, 6.5% risk premium, 1.0 beta), terminal growth rate 3%. Sensitivity analysis shows the market has not yet assigned a meaningful PE premium to earnings and free cash flow supported by memory supplier LTAs. If a 10x PE multiple is applied to LTA-supported earnings and assuming 70% commodity LTA coverage, Samsung/hynix's 2027 PE can expand from the current 5x to approx. 8.5x.

Methodology notes

• Industry/Industrial Analysis FrameworkSupply-demand framework

  Memory Cycle Supply-Demand Analysis

  Classic memory industry analysis framework: Previous cycles were never a demand problem, but supply-driven—vendors competed to fill demand gaps and over-expanded. The key to breaking the cycle is supply discipline, but this variable has never really worked. The report uses this framework to explain why this round of AI structural demand combined with relatively restrained supply will form a longer cycle and higher peak.

• Valuation MethodRIM Residual Income Model

  Residual Income Valuation Model

  The report uses the Residual Income Model to value SK hynix and Samsung, assuming a cost of equity of 11.5% (5% risk-free rate, 6.5% risk premium, 1.0 beta), terminal growth rate 3%. This is a common valuation method for cyclical stocks, suitable for industries with large earnings fluctuations but stable long-term return expectations.

• Cycle & Prosperity FrameworkProsperity Turning Point Analysis

  Four Signals Monitoring for Cycle Turning Points

  The report monitors four early signals of cycle turning points: continued strength in DRAM prices, inventory adjustment cycle, capital expenditure budget cuts while capital intensity rises, and sustained earnings revisions and valuation expansion. This helps investors identify trend changes early in the cycle, rather than confirming afterwards.

• Industry/Industrial Analysis Framework

  Impact of LTA Long-Term Agreements on Pricing Stability

  The report analyzes how Long-Term Agreements (LTA) change the high volatility pricing characteristics of the memory industry. LTA can lock in high prices through prepayments, improve earnings predictability, and reduce the risk of over-building capacity. If LTA accounts for 70%+ of total supply over the next 3-5 years, DRAM stock PE can re-rate from 5x to 8-10x. This is a unique analysis perspective for the memory industry.

• Industry/Industrial Analysis FrameworkSubstitution Effect Analysis

  AI Demand Price Elasticity Analysis

  The report distinguishes the essential difference between AI demand and traditional consumer electronics demand: AI demand has price elasticity—lower DRAM prices reduce AI inference operating costs, making AI deployment cheaper, thereby creating new demand rather than just reducing costs for a fixed number of devices. This changes the logic of demand shrinking during traditional memory cycle downturns.

• Company Fundamentals & Financial FrameworkEarnings Quality Analysis

  Valuation Premium for LTA-Supported Earnings

  The report's sensitivity analysis shows the market has not yet assigned a meaningful PE premium to earnings and free cash flow supported by memory supplier LTAs. If a 10x PE multiple is applied to LTA-supported earnings and assuming 70% commodity LTA coverage, Samsung/hynix's 2027 PE can expand from the current 5x to approx. 8.5x. This helps investors identify valuation mismatch opportunities.

• Cycle & Prosperity FrameworkCapacity/Equipment Cycle (Juglar)

  Memory Capacity Cycle Analysis

  The report analyzes the memory industry capacity cycle: New capacity from current investment plans will come online at the end of 2027, at which point memory prices will fall and the cycle will turn. However, AI-driven demand is tied to long-term infrastructure projects rather than short-term consumer cycles, giving memory producers more future demand visibility and confidence, and they will not over-build capacity aggressively like in previous cycles.

• Industry/Industrial Analysis FrameworkUpstream/Midstream/Downstream Industry Chain Transmission

  AI CPU and Memory Demand Transmission

  The report analyzes the transmission effect of AI CPU architecture changes on memory demand: Under agent AI workloads, the CPU:GPU ratio shifts from 1:8 in the training era to 1:1 or higher, with per-agent token intensity surging, supporting higher DRAM content per CPU. Data centers are moving towards laptops, smartphones, and homes, with edge devices (such as agent AI PCs) bringing additional upside space for memory demand.

Asset mapping & comparison

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

• SK hynix (000660.KS)

  Benefit: HBM leadership + commodity memory cycle uptrend, AI inference demand growth supports strong earnings in 2026-2027

  Strengths

  Leadership in HBM field, LTA provides downside protection before 2028, stronger commodity pricing assumptions, stronger 2027 HBM pricing reset

  Weaknesses

  HBM competition risk remains but market awareness is improving

  Comparison

  Compared to Samsung, SK hynix has more leadership in the HBM field, but Samsung is more diversified in overall memory and OLED business

  Risks

  End demand weaker than expected, intensified DDR5 competition leading to overspending on supply side, cloud and China smartphone customers maintain high inventory

• Samsung Electronics (005930.KS)

  Benefit: HBM improvement + commodity uptrend cycle = stock re-rating, potential for HBM market share gain from 2026

  Strengths

  HBM improvement and potential market share gains, price increases and inventory decline favorable for memory business margin recovery, 2027 expected PB approx. 2x consistent with commodity cycle peak

  Weaknesses

  Report holds relatively conservative view on HBM progress this year, but sequential improvement in 2026 and beyond supports earnings growth

  Comparison

  Compared to SK hynix, Samsung business is more diversified (memory + OLED + smartphones), but HBM leadership is slightly inferior

  Risks

  Product and memory cycle risks (including Apple and China new smartphone competition), earnings growth concentrated in semiconductors, China competition

• Samsung Electronics Preferred Shares (005935.KS)

  Benefit: Same commodity cycle and HBM improvement logic as common shares, and discount to common shares expected to narrow

  Strengths

  Based on common stock target price, assuming preferred shares trade at 20% discount to common shares (average level over past 2 years)

  Weaknesses

  Risk of discount deepening

  Comparison

  Compared to common shares, preferred shares trade at 20% discount, but enjoy same fundamental drivers

  Risks

  Same cycle risks as common shares, plus risk of discount deepening

Key data

• Memory Price Increase (since Feb 2026)Nearly doubledCore driver is physical capacity constraints under LTA constraints
• DRAM Stock Price Increase (past 2 months)70-134%Earnings forecast adjustments synchronously support valuation stability
• 3Q26 DRAM Price Expected Increase+20-30%Sufficient to maintain accelerating YoY growth
• Current DRAM Stock PE (2027 Expected)Approx. 5xSK hynix approx. 4.7x, Samsung approx. 5.2x
• PE Re-rating Target under 70%+ LTA Scenario8-10xBased on 70%+ LTA-supported earnings valued at market multiples 10-14x PE
• SK hynix Bear Case Valuation Increase+175%New bear case valuation W1,100,000, corresponding to 2.4x 2027 expected PE
• Samsung Bear Case Valuation Increase+58%New bear case valuation W190,000, corresponding to 3.1x 2027 expected PE
• HBM Price Expected Increase (from 2027)+50-100% YoYHBM pricing negotiations still ongoing
• AI-Related Product Supply ShareOver 50% in 2026, nearly 70% in 2027Consumer electronics and module vendors will face continued order squeeze
• AI NAND Bit Growth Expectation (by 2027)+50% YoYMainly driven by CSP's KV Cache offloading demand

Impact & implications

Impact on Industry: The report believes AI structural demand is reshaping the cyclical characteristics of the memory industry. Traditional memory cycles are supply-driven with relatively fixed demand, while this round of AI demand has price elasticity—lower DRAM prices reduce AI inference costs and create new demand. This means the cycle may be longer and peak higher, and demand will not shrink as sharply as in traditional cycles during downturns. Impact on Companies: SK hynix and Samsung, as major suppliers of HBM and DRAM, will benefit significantly from the growth in AI infrastructure spending. The report raised bear case valuations for both companies, reflecting increased confidence in visibility of continued shortages in 2027. SK hynix's leadership in the HBM field and Samsung's potential to gain HBM market share are both catalysts for valuation re-rating. Implications for Investors: The report hints that the current valuation of DRAM stocks at approx. 5x 2027 expected PE has not fully reflected the stability premium of LTA-supported earnings. If LTA share reaches 70%+, there is room for valuation re-rating to 8-10x PE. However, investors need to monitor four turning point signals: narrowing DRAM price increases, inventory adjustments, capital expenditure changes, and slowing earnings revisions.

Risks

• Technological breakthroughs significantly reduce memory/HBM usage, or technological changes emerge that bypass memory
• AI race derails: Need five major frontier LLMs and hyperscale AI infrastructure spending drivers to remain high
• AI demand growth rate deviates from track, no longer exponential: ARR (Annualized Run Rate) trajectory is an important indicator of enterprise API contract AI usage and agent coding tool adoption
• Chip inflation and macro impact: Whether memory price increases are passed through channels slowly enough that by the time price shock fully manifests, two years have passed
• Market liquidity: Liquidity tightens again from growth rate perspective, China actively slows money supply growth, plus rising oil prices and strong USD, liquidity environment not as loose as Q1
• 2028 AI computing digestion period leads to more aggressive margin erosion in HBM segment, faster DRAM price decline and weak demand recovery

What to watch

• Whether DRAM price increase narrows from significant rise in 4Q25-2Q26 to 8-13% QoQ (LTA brings stability rather than turning point signal)
• Supplier inventory levels: Early 2Q26 DRAM only 2-3 weeks, NAND only 4-5 weeks, at historical lows
• Capital expenditure budget cuts while capital intensity rises: AI demand may keep absolute capital expenditure levels high during downturn cycle
• Sustainability of earnings revisions and valuation expansion: Look for further earnings upside through HBM pricing negotiations
• HBM pricing negotiation progress: Prices expected to rise 50-100% YoY from 2027
• Changes in AI-related product supply share: From over 50% in 2026 to nearly 70% in 2027
• Dynamics after Korean market circuit breaker: Historically next day hit rate approx. 75%, medium-term horizon improves to 87-88%