LTAs Reshape Memory Valuation Logic, Favoring Long-Term Industry Uptrend

Overview

This report provides an in-depth analysis of the current state, drivers, and potential structures of Long-Term Agreements (LTAs) in the global memory market. J.P. Morgan believes that with AI-driven memory demand surging and supply constraints persisting, LTAs will become a critical link between Cloud Service Providers (CSPs) and memory manufacturers. This not only stabilizes supply chains but could fundamentally transform the memory industry's business model from traditional 'make-to-stock' to a foundry-like 'make-to-order' approach, driving a valuation shift from cyclical P/B to more stable P/E. Based on this, the firm maintains a long-term bullish view on the memory sector and raises target prices for key covered stocks.

Core views

The rise of LTAs stems from mutual needs of both buyers and sellers. For buyers (e.g., CSPs), memory costs have surged over 3x in the past year and become a strategic asset for AI, making stable supply crucial. For sellers (memory manufacturers), after multiple downturn losses, they urgently seek LTAs to smooth cyclical volatility and gain higher demand visibility to support capital expenditures. On the supply side, the report forecasts that even with aggressive capex plans, DRAM wafer capacity will face a shortage of ~450k wafers/month (wfpm) from 2026-2030. This unprecedented shortage means initial LTA terms may favor memory makers, but CSPs will also implement safeguards to ensure supply stability. Regarding LTA structures, the report outlines three scenarios: fixed price, prepayment plus fixed price, and prepayment plus price flexibility. The 'prepayment + price flexibility' model (Scenario 3) is seen as optimal, as it locks in volume via prepayment while allowing price adjustments (e.g., no more than 20% below benchmark), balancing supplier volume guarantees with customer cost flexibility. Valuation-wise, the report sees the memory industry at an inflection point. Traditionally, due to high earnings volatility and product homogeneity, memory stocks were valued on P/B. But as LTA adoption grows, improving earnings visibility, the sector may transition to P/E valuation. Drawing parallels with TSMC's historical valuation shift, the report notes that higher earnings visibility tends to justify P/E multiples. Memory stocks currently trade at 7.3x forward P/E, below the 11-year average of 9x, suggesting re-rating potential.

Analysis framework

The report combines supply-demand balance analysis with historical valuation comparisons. First, it reverse-engineers the DRAM wafer supply needed to meet CSP AI memory demand, quantifying future shortages to justify LTA necessity. Next, it dissects LTA contract variables (e.g., prepayment, pricing mechanisms, enforcement clauses), using analogies like rental deposits to explain structural trade-offs. Finally, by reviewing TSMC's historical P/B-to-P/E transition and 2017's failed DRAM LTA attempt, it argues current LTAs are more enforceable and can support valuation re-rating.

Methodology notes

• Valuation methodsPE/PEG valuation

  Shift from P/B to P/E valuation framework

  The report argues that when industry earnings volatility declines and visibility improves, markets prefer P/E over P/B valuation. Similar to TSMC's post-Apple contract valuation shift, memory LTAs lock in long-term demand, enabling this transition.

• Industry Analysis FrameworksSupply-demand framework

  DRAM wafer shortage calculation

  By estimating HBM and advanced DRAM demand from AI servers and comparing it with announced capacity plans, the report concludes significant future shortages, underpinning LTA adoption and price hikes.

• Competition & Strategy FrameworksMoats/Competitive Advantages

  Make-To-Stock to Make-To-Order transition

  Traditional memory is 'make-to-stock,' prone to cycles; LTAs drive 'make-to-order,' akin to foundries. This structural change reduces imbalance risks, enhances pricing power and earnings stability, forming a new competitive edge.

Asset mapping & comparison

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

• Samsung Electronics (005930.KS)

  Beneficiary: As a global memory leader, it gains from LTA-driven stable demand and premiums, likely first to see valuation re-rating.

  Strengths

  Technology leadership, scale advantages, broad client base.

  Comparison

  Duopoly with SK hynix benefits from industry consolidation.

  Risks

  Capex execution risks, unfavorable LTA terms.

• SK hynix (000660.KS)

  Beneficiary: Leads in HBM, deeply tied to major CSPs, high LTA potential.

  Strengths

  HBM tech edge, close ties with NVIDIA etc.

  Comparison

  Leads in advanced AI memory, higher elasticity.

  Risks

  Tech iteration risks, competitor catch-up.

• KIOXIA Holdings (285A.T)

  Beneficiary: As a major NAND player, will join LTA trend, improving earnings stability.

  Strengths

  NAND tech expertise, potential merger synergies.

  Comparison

  Unlike DRAM duopoly, NAND has different competitive dynamics but still benefits from overall memory upturn.

  Risks

  NAND price volatility, merger uncertainty.

Key data

• DRAM Wafer Shortage Forecast2026E-2030E cumulative ~450k wafers/monthSupply-demand imbalance persists through 2030 even with aggressive expansion
• Memory Stocks YTD Gain238%Far exceeds Philadelphia Semiconductor Index (SOX) at 65%
• Current Memory Sector Forward P/E7.3x (FTM P/E)Below 11-year average of 9x (excluding 2022Q4-2023 downturn)
• Samsung New Target Price480,000 KRWDec 2026 target, Overweight rating
• SK hynix New Target Price3,000,000 KRWDec 2026 target, Overweight rating
• KIOXIA New Target Price80,000 JPYDec 2026 target, Overweight rating

Impact & implications

The report argues LTA adoption will structurally improve the memory industry's business model, reducing cyclical volatility and delivering stable cash flows and earnings visibility. This may attract stability-seeking investors, driving multiple expansion. For investors, memory stocks' thesis shifts from pure cyclical plays to growth-stability hybrids. While near-term skepticism about 'this time is different' persists, rising LTA signings and earnings delivery make valuation re-rating likely.

Risks

• AI monetization slows significantly, undermining LTA necessity.
• KVcache/tokenization tech advances reduce per-model memory needs.
• Capacity oversupply from aggressive expansions or new entrants.
• Unfavorable LTA terms hurt current earnings forecasts.
• Macro (tariffs/sanctions) or industry (balance sheet issues) disrupts AI demand.

What to watch

• Upcoming LTA details/scale from Samsung and SK hynix.
• Memory's share in CSP capex trends.
• Memory industry capex discipline and expansion pace.
• LTA's rising share in total bit shipments.