AI-Driven Memory Uplift for Five Years, Bullish on Memory and MLCC/Substrates

Overview

This is an Asia tech outlook presentation from JP Morgan, systematically updating the cycle and growth outlook for Asian (particularly Korean) tech hardware and semiconductors, covering memory (DRAM/NAND/HBM), HBM Thermal Compression Bonding (TCB), MLCC, ABF substrates, Apple supply chain, and the relative value of Korean holding companies. The overall tone is bullish: The institution believes that the AI-centric memory upcycle is a 'five-year' long-term trend, driven by demand-side factors including token economics, growth in AI compute demand, and increasing server memory density; simultaneously, the institution also acknowledges the market's rising concerns about cyclical sustainability, with the core worry being a misalignment between Cloud Service Provider (CSP) profits and capital expenditure on the one hand and semiconductor/hardware company profits on the other. In other words, the fundamentals are strong but the market is questioning 'How long can this high growth last?'

Core views

Memory is the core bullish thematic in this outlook. Demand side: The institution believes token economics (longer contexts, multimodal AI, agents/MCP driving explosive token consumption and memory footprint), stronger AI compute demand, and increased server architecture memory density are collectively driving a roughly five-year memory upcycle; strong CPU demand is an additional tailwind, prompting the institution to raise 27E+ server bit demand by over 20%, noting that the replacement cycle for conventional server DRAM supports 26E-27E bit demand growth of +89%/+64% YoY. Supply & Pricing: The proportion of wafers allocated to HBM is rising steadily (25: 351k wafers- approx. 19% of industry; 26E: ~475k - approx. 24%; 27E: ~633k - approx. 28%), with the allocation to traditional non-HBM wafers shrinking year on year, coupled with natural per-wafer output reduction from process migrations, keeping DRAM supply/demand tight for longer; the institution assesses high upside price risk for HBM next year— although spot price increases are decelerating, WFE capital expenditure growth is seen as a signal of 'healthily alleviating constraints.' On NAND, driven by eSSD TAM expansion, KV cache offloading, and HDD shortages, the institution also offers a positive outlook. Scale & Profitability: The institution projects global memory TAM reaching ~$1.3 trillion/$1.7 trillion in 27E/28E (approx. 8x growth in three years), with memory normalized OPM staying in the high 70% range; the value share of memory in CSP capital expenditure is expected to climb to approximately 50% by 26E. On ratings: Samsung Electronics, SK Hynix, and Kioxia are all rated Overweight(OW), Nanya Technology Neutral(N), and Hanmi Semiconductor Underweight(UW); the institution emphasizes that the focus has shifted from 'momentum' to 'profit persistence' and awaits LTA renewals to validate the sustainability of multiple re-ratings. AI Hardware (Passives & Substrates): MLCC benefits from AI (server-side) penetration and continuous EV volume growth, with ASP facing tailwinds; Various ABF substrate demand and pricing strengthen, MLB demand is also robust. The institution stresses 'AI exposure is key to stock selection,' thus assigning differentiated ratings within this chain: Overweight on Samsung Electro-Mechanics (SEMCO) and ISU Petasys, Neutral on Leeno, Underweight on SKC. Cyclical Hardware: Recent upward revisions in iPhone EMS production benefit LG Innotek (OW) and LG Display (N), but the institution flags that rising materials (BOM) costs may pressure 26E margins; simultaneously, LG Electronics (N) faces headwinds in TV business margins. Relative Value (Korean Holding Companies): The institution notes that catalysts are limited recently following the third commercial law incident; Top pick is Samsung C&T (SCT, OW), citing upside potential in dividend yield plus the value of its holdings in Samsung Electronics/Samsung BioLogics; also favors SK Inc (OW), Neutral on LG Corp.

Analysis framework

The institution's analytical framework is a step-wise deduction along 'demand → supply → price → profitability → valuation,' applied consistently through the lens of the AI supply chain. Demand & Transmission: First uses token economics and increased server memory density— a 'content-driven' logic— to argue how AI amplifies memory consumption per unit of compute, then traces the industry chain transmission 'CSP CapEx → AI semiconductors → memory value share' to quantify the changing proportion of memory in CSP expenditure, thus translating macro AI narratives into concrete upward revisions for bit demand. Supply/Demand & Volume/Pricing: Internally in memory, the institution uses a typical S/D balance framework (S/D glut, HBM supply/demand table, wafer capacity allocation ratio to HBM) to gauge tightness, decomposing revenue/TAM into 'bit shipments × ASP,' then rolling quarterly forecasts by category (HBM3E/HBM4, DRAM, NAND), down to supplier share, capacity (wspm), and blended ASP. Valuation & Relative Value: Stocks and peers are compared using P/E band and P/B (e.g., Hanmi's historical premium/discount relative to peers), while Korean holding companies are analyzed with an NAV discount framework supplemented by historical discount mean and Z-score to judge whether discounts are at extreme levels, thus identifying relative value opportunities.

Methodology notes

• Industry/Sector Analysis FrameworkSupply-demand framework

  Memory/HBM Supply-Demand Balance Analysis (S/D glut, S/D gap, wafer capacity allocation to HBM)

  In cyclical memory, prices are largely set by S/D gaps. The report projects demand and supply for HBM, DRAM, NAND (including wafer allocation ratios to HBM) to determine whether markets are 'tight,' informing judgements on price direction up or down— key to understanding the 'price increase/shortage' logic this cycle.

• Industry/Sector Analysis FrameworkVolume-price decomposition

  Revenue/Market Size = Bit Shipments × Average Selling Price (ASP)

  The institution breaks memory revenue and TAM into 'shipped bits × unit price,' rolled quarterly by category. This clarifies whether growth is driven by 'volume' or 'pricing,' and facilitates quantifiable inclusion of factors like HBM price hikes and ASP premiums.

• Industry/Sector Analysis FrameworkUpstream→Midstream→Downstream Transmission

  CSP CapEx → AI Semiconductors → Memory Value Share Transmission

  The report traces downstream through the AI supply chain: start with CSP CapEx scale, then examine the proportion flowing to AI chips and memory. Positioning memory as '~50% of CSP CapEx' helps readers understand memory demand's high dependency on AI investment cycles.

• Cycle and Expansion FrameworkCapacity/Equipment Cycle (Juglar)

  WFE/CapEx Expansion Pace Sets Supply-Side Cycles

  Memory is capital-intensive; equipment investment (WFE) and capacity expansion timing dictate future supply. The report interprets 'CapEx increasing but output growth restrained' as a healthy signal easing shortages, noting supply discipline is crucial for cycle sustainability.

• Cycle and Expansion FrameworkInflection Point Analysis

  Assessing Upswing Persistence ('Five-Year Upswing' & Profit Persistence)

  The institution emphasizes a shift in focus from near-term 'momentum' to 'profit persistence,' awaiting LTA renewals for validation. This signals that, amid peak growth, the key question isn't 'how much upside' but 'for how long will this high-growth last?'

• Valuation methodsPB valuation

  Using Price-to-Book (P/B) for Historical & Peer Comparisons of Memory Leaders

  Capital-intensive, cyclical memory stocks commonly use P/B to gauge valuation within a historical context. The report compares the forward P/B of Samsung, SK Hynix, Micron to assess whether current valuations are stretched or offer further re-rating potential.

• Valuation methodsNAV Net Asset Value Method

  Korean Holding Companies' NAV Discount Analysis

  Holding company value is often derived by discounting the net value (NAV) of their holdings. The report quantifies discounts to NAV across holdings, uses historical averages and Z-score to determine if discounts are extreme, seeking relative value opportunities (e.g., favoring Samsung C&T).

• Quantitative/Factor/Portfolio TheoryOthers

  Historical Discount Z-score (Standard Score) Statistics

  Z-score measures how many standard deviations the current discount deviates from its historical distribution. An extreme value indicates the discount is far from the norm, useful for gauging 'whether the discount is overdone, signaling a potential mean-reversion opportunity.'

Asset mapping & comparison

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

• Samsung Electronics (005930.KS)

  Core beneficiary of memory (DRAM/NAND/HBM) and AI cycle, Overweight Overweight(OW)

  Strengths

  Comprehensive memory portfolio, benefits from AI-driven TAM expansion and high OPM; Increasing HBM share and capacity

  Weaknesses

  Historical volatility in HBM share, profit persistence requires validation via LTA renewals

  Comparison

  Core long thematic alongside SK Hynix; institution rates both OW

  Risks

  Sustained cycle concerns, CSP spending-semiconductor profits misalignment

• SK Hynix (000660.KS)

  HBM leader, 2026 order discussions finalized, maintains leadership position, Overweight Overweight(OW)

  Strengths

  HBM leadership and higher wallet share, benefits from DRAM upcycle

  Weaknesses

  Supplier diversification in HBM poses potential share dilution

  Comparison

  Leads Samsung in HBM share though market focus is on future share trajectory

  Risks

  Binary shifts in share due to HBM customization, cycle sustainability

• Kioxia (285A)

  NAND play, driven by eSSD expansion, Overweight Overweight(OW)

  Strengths

  Benefits from NAND rebound and eSSD TAM expansion

  Comparison

  Represents a long in the NAND direction within the memory chain

  Risks

  Potential risk in NAND content growth, price volatility

• Nanya Technology Nanya Technology (2408.TW)

  Peer, rated Neutral(Neutral(N)

  Weaknesses

  Relatively limited AI exposure

  Comparison

  Institution rates it Neutral relative to OW long longs in Samsung/Hynix/Kioxia

• Hanmi Semiconductor (042700.KS)

  HBM Thermal Compression Bonding (TCB) equipment maker, Underweight Underweight(UW)

  Strengths

  Benefits from HBM TCB market; historically enjoyed valuation premium from AI/HBM thematic

  Weaknesses

  Valuation premium historically subject to discount volatility as HBM supplier diversification occurs (notably within SKH)

  Comparison

  Institution is more cautious on this stock relative to OW names in memory chain

  Risks

  Changes in TBC cycle/supply chain life assumptions, valuation premium mean-reversion

• Samsung Electro-Mechanics Samsung Electro-Mechanics (009150.KS)

  MLCC + ABF substrates, AI server + EV beneficiary, Overweight Overweight(OW)

  Strengths

  Benefits from AI (servers) + EV driving MLCC demand/ASP tailwinds; substrate demand/pricing strengthening

  Comparison

  Within passive/substrate chain, higher AI exposure prevails over Leeno/SKC

• ISU Petasys (007660.KS)

  MLB/high-layer PCBs, driven by AI accelerators + network switches + Google TPU, Overweight Overweight(OW)

  Strengths

  Strong MLB demand; benefits from AI accelerators + network switches; growth in Google PCB TAM

  Weaknesses

  High customer concentration risk risks tied to key client Google

  Comparison

  Strategically long alongside SEMCO as top AI hardware chain picks

  Risks

  Customer concentration

• Leeno (058470.KQ)

  Test probes/IC sockets etc, rated Neutral(Neutral(N)

  Weaknesses

  Relatively limited AI exposure

  Comparison

  Institution assigns Neutral relative to OW ratings on SEMCO/ISU

• SK Materials SKC (011790.KS)

  Material play, Underweight Underweight(UW)

  Weaknesses

  Relatively weaker AI exposure

  Comparison

  Trails AI hardware chain peers; institution is most cautious here

• LG Innotek (011070.KS)

  Apple chain, benefits from iPhone EMS production revision upward, Overweight Overweight(OW)

  Strengths

  Recent iPhone EMS production revision provides a positive near-term lift

  Weaknesses

  Rising BOM materials costs may pressure 26E margins

  Comparison

  More relative beneficiary than LGD (N), yet still faces cost pressures

  Risks

  Rising BOM costs compressing margins

• LG Display (034220.KS)

  Apple chain, marginally benefits from iPhone EMS revision upward, rated Neutral(Neutral(N)

  Strengths

  Near-term positive impact from iPhone EMS revision

  Comparison

  Less benefited relative to LG Innotek

  Risks

  Rising BOM costs

• LG Electronics LGE (066570.KS)

  Consumer Electronics, TV margins under pressure, rated Neutral(Neutral(N)

  Weaknesses

  Headwinds in TV business margins

  Comparison

  Institution adopts a conservative stance on Consumer Electronics segment

  Risks

  TV margin pressures

• Samsung C&T Samsung C&T (028260.KS)

  Korea Rel Value top pick, Overweight Overweight(OW)

  Strengths

  Upside to dividend yield plus asset portfolio value embedded in Samsung Electronics/Samsung BioLogics

  Comparison

  Top relative value pick favored over Neutral-rated LG Corp

  Risks

  Some catalysts seen post-3rd commercial law event

• SK Inc (034730.KS)

  Holdco rel value play, Overweight Overweight(OW)

  Strengths

  Relative value opportunity exists with NAV discount

  Comparison

  Institution favors SK Inc Overweight, versus Neutral on LG Corp

• LG Corp (003550.KS)

  Holdco, rated Neutral(Neutral(N)

  Weaknesses

  Less rel value appeal vs SCT/SK Inc

  Comparison

  Versus SK Inc's OW, institution assigns Neutral

Key data

• Global Memory Market Size (TAM)27E ~$1.3 Trillion, 28E ~$1.7 Trillion (Fig 1,$1,337/$1,681 Billion Base)Institution cites ~8x growth over three years, normalized OPM stable in high 70% range
• Memory as % of CSP CapEx26E ~50% (Fig ~52%)Under AI industry chain transmission, memory's value share continues rising
• Server Bit Demand Revision27E+ revised upward by >20%Strong CPU serves as an extra demand tailwind
• Conventional Server DRAM Bit Demand26E-27E +89%/+64% YoYReplacement cycle support
• DRAM Industry WSPM25YE ~1.8M Wafers → 28YE ~2.6M WafersCapacity expansion trend
• Wafers Allocated to HBM25: 351k (~19% of Industry), 26E: ~475k (24%), 27E: ~633k (28%)HBM crowding out traditional DRAM capacity, intensifying tight supply conditions
• HBM Revenue (Peer Total)2025: ~$33.5 Billion → 2028E: ~$225.2 BillionInstitution estimates high upside price risk for HBM next year; blended ASP rising annually

Impact & implications

Per the institution's view, the AI-driven memory upswing is seen as structural rather than a short-term spike, hence Samsung Electronics, SK Hynix, Kioxia— as core beneficiaries in memory/HBM— are assigned Overweight ratings; HBM share, capacity, and blended ASP will be key to subsequent earnings leverage. In the passive and substrate chain, AI server and EV demand are expected to drive MLCC, ABF substrate, MLB volume/price improvements, leading the institution to favor names with higher AI exposure such as Samsung Electro-Mechanics and ISU Petasys. In cyclical hardware, the upward revision in iPhone EMS production is slightly positive for LG Innotek, but BOM cost increases may erode 26E margins, and LG Electronics' TV margins remain under pressure. On relative value, the institution sees opportunity in Korean holding companies' discounts to NAV plus dividend/asset value, selecting Samsung C&T as top pick and favoring SK Inc. It must be stressed that these are solely the institution's judgements, and the institution simultaneously flags uncertainty about cyclical sustainability— no investment recommendations are implied herein.

Risks

• Cycle Sustainability Questionable: Misalignment between CSP earnings/capEx and semiconductor/hardware profits
• Memory Profit Persistence hinges on LTA renewals for validation; risk of mean-reversion after multiple re-ratings
• HBM Supplier Diversification/Customization leads to binary changes in share allocation (impact on Hanmi et al.)
• Spot price increases deceleration; HBM3E approaching mild oversupply (versus tightness in HBM4)
• Rising cyclic hardware BOM costs pressure 26E margins; LG Electronics' TV business margins also pressured
• Weak demand in non-Apple edge devices; conservative stance on consumer electronics maintains
• Catalysts scarce for Korean holding companies after third commercial law incident

What to watch

• LTA Renewals: for validating memory profit persistence
• HBM next year pricing momentum & price hike cadence (upside risk perceived as high)
• WFE/CapEx & capacity expansion rhythm (whether supply discipline persists)
• Directionality in HBM share: diversification versus concentration, and impact of custom HBM
• iPhone EMS production schedules & BOM cost movements
• CSP CapEx & earnings alignment (whether the misalignment converges)