LTAs may reshape the memory cycle, lifting SK hynix earnings and valuation

Overview

This report discusses how long-term agreements (LTAs) are changing the business model and valuation framework of the memory industry. Morgan Stanley believes that AI data centers and hyperscale cloud customers are treating memory as a critical infrastructure bottleneck, prompting a shift from opportunistic purchasing to multi-year supply lock-ins. For memory manufacturers, LTAs can bring greater order visibility, price protection, prepayment cash flow, and margin stability, thereby cushioning the downside in the traditional DRAM, NAND, and HBM cycles.

Core views

The core view is that this round of LTAs may represent a structural change in the memory cycle rather than simply order locking during an upswing. The report argues that customer prepayments, price bands, minimum margin protection, and strong contractual terms will reduce inventory and pricing risk after capacity expansion, allowing memory companies to secure more stable long-term revenue and free cash flow. The market still values LTA-supported earnings like traditional cyclical commodity earnings, which may be too conservative; if the agreements prove to be cash-backed, sticky, and durable across cycles, both Samsung and SK hynix could see room for P/E re-rating.

Analysis framework

The report combines industry supply-demand analysis, management commentary, an LTA term framework, scenario valuation, and analogy-based research. On the supply-demand side, it focuses on AI semiconductor market growth, DRAM supply constraints from EUV equipment bottlenecks, and cloud customers' need for medium- to long-term supply assurance. On valuation, it uses a residual income model to adjust SK hynix's target price and applies LTA coverage ratios together with different P/E multiples to estimate the implied group P/E for Samsung and SK hynix.

Methodology notes

• Valuation methodsResidual income model

  Estimates SK hynix's target price using cost of equity, risk premium, beta, and terminal growth rate.

  The report's base case uses a residual income valuation model, assuming a cost of equity of 11.5%, a risk-free rate of 5%, an equity risk premium of 6.5%, beta of 1.0, and a terminal growth rate of 3%, which raises the target price to W2,600,000.

• Scenario AnalysisLTA Coverage and P/E Sensitivity Analysis

  Assigns different valuation multiples to LTA-supported earnings and non-LTA commodity memory earnings.

  The report assumes that HBM is supported by LTAs almost 100% of the time, while commodity memory has LTA coverage of 50%-80%; it applies a 6-12x P/E to the LTA portion and a 5x P/E to non-LTA commodity memory to assess whether the market is underestimating the stability of earnings.

• Industry Supply and DemandAI-Driven Memory Supply-Demand Framework

  AI infrastructure demand and EUV equipment bottlenecks together drive long-term supply lock-ins.

  The report cites TSMC's upward revision of the semiconductor TAM to USD$1.5tr and argues that the AI semiconductor market still has 50%-60% growth potential through 2030, while DRAM supply growth may be constrained by EUV bottlenecks, reinforcing the need for customers to sign LTAs.

• Analogy AnalysisApple Capital Returns and Japanese Shipping Re-rating Analogy

  Uses valuation changes in other industries after cash flow improvement and higher shareholder returns to explain the potential re-rating path for memory.

  The report uses Apple's buybacks and dividends to explain the source of long-term excess returns, and it draws an analogy to Japanese shipping companies, whose valuations were re-rated after freight rates fell post-pandemic but dividend payout ratios improved, to show that memory companies could command higher valuations if LTAs raise free cash flow and shareholder returns.

Asset mapping & comparison

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

• SK hynix

  Core beneficiary and target price adjustment object

  Strengths

  HBM leadership, demand and revenue visibility from LTAs, cyclical improvement in commodity DRAM/NAND, stronger free cash flow, and potential improvement in capital returns.

  Weaknesses

  Still exposed to HBM competition, commodity memory cycles, and end-demand volatility.

  Comparison

  The report believes the market has not fully assigned a valuation premium to SK hynix's LTA-supported earnings and free cash flow, and its current F27e P/E of about 5x is below the potential implied P/E in the scenario analysis.

  Risks

  AI compute demand digestion, HBM price competition, DRAM price declines faster than expected, high inventories, and a weaker macro environment.

• Samsung Electronics

  A comparable beneficiary in the Korean memory industry

  Strengths

  It may also lock in supply, pricing, and customer relationships through multi-year LTAs, and the scenario analysis shows room for P/E re-rating.

  Weaknesses

  Its relative benefit may be affected by HBM competitiveness and product mix.

  Comparison

  Under a scenario of 70% commodity memory LTA coverage and a 10x LTA multiple, Samsung's implied group P/E is about 8.5x; under an 80%/12x upside scenario, it is about 10.5x.

  Risks

  HBM competition, commodity memory supply-demand fluctuations, and weaker-than-expected execution of customer orders and pricing mechanisms.

• Micron

  Evidence sample for the industry-wide LTA trend

  Strengths

  It has announced a five-year Strategic Customer Agreement containing multi-year supply volumes and pricing commitments.

  Weaknesses

  The report does not make a specific target price adjustment for it.

  Comparison

  As a U.S. memory manufacturer, its announcement reinforces that LTAs are becoming an industry-wide phenomenon.

  Risks

  Agreement execution, memory cycle volatility, and changes in customer demand.

• SanDisk

  A NAND-related LTA / new business model case

  Strengths

  It describes its LTA strategy as New Business Models, has signed five long-term partnerships, and has financial guarantees or collateral supporting customer commitments.

  Weaknesses

  It serves more as evidence of the business model than as a primary valuation object in this report.

  Comparison

  Its terms show customers supporting purchase obligations with billions of dollars in collateral or financial instruments, indicating stronger LTA enforceability than in the past.

  Risks

  Customer performance, NAND demand volatility, and the sustainability of the agreements.

• Kioxia

  A case study in the Japanese memory industry

  Strengths

  Management has said that long-term supply agreements with hyperscale customers extend into fiscal 2028-2029, improving revenue visibility.

  Weaknesses

  The report does not provide a detailed valuation adjustment.

  Comparison

  Together with Micron, Samsung, SK hynix, and SanDisk, it provides evidence of LTA diffusion in the memory industry.

  Risks

  Demand realization, pricing terms, and industry competition.

• Hyperscalers

  The demand side of LTAs and the driver of AI infrastructure buildout

  Strengths

  They lock in critical memory supply through multi-year agreements, reducing supply uncertainty in AI infrastructure buildout.

  Weaknesses

  They must bear higher prepayments, pricing commitments, or supply assurance costs.

  Comparison

  Their purchasing behavior is shifting from opportunistic buying to strategic supply-chain partnerships.

  Risks

  If AI demand is weaker than expected, prepayments and purchase commitments may add cost pressure.

Key data

• LTA term3-5 yearsThe report believes this round of memory LTAs spans roughly 3-5 years and may explicitly define pricing and volume.
• Future supply bound by LTAsMore than 50%Morgan Stanley expects more than 50% of future supply to be bound by LTAs over the next 2-3 years, with non-AI customers potentially adding further upside.
• Prepayment ratio cluePeak demand in 2027: about 50%; peak demand in 2028: about 100%The report says some suppliers have already secured relatively high prepayment or downpayment ratios in LTA negotiations, creating a buffer heading into 2028.
• AI semiconductor market growth50%-60% growth by 2030The report is based on TSMC's upward revision of the semiconductor TAM forecast from USD$1tr to USD$1.5tr.
• DRAM supply constraintBit shipment annual growth may be capped at around 30%The report believes EUV tool bottlenecks from 2027 onward may limit DRAM supply expansion.
• SK hynix EPS adjustment2026E +6%; 2027E +3%; 2028E +14%The increase reflects LTA downside protection through 2028, stronger commodity memory price assumptions, and HBM price resets.
• SK hynix target priceW2,600,000Based on the residual income model, implying about 38% upside.
• Bull-case valueW3,000,000; about 60% upsideThe bull case includes HBM prices potentially doubling to US$3/Gb and remaining at a higher long-term level of roughly US$4/Gb.
• LTA valuation sensitivityAbout 5.5x in the conservative case; about 8.5x-8.6x at 70% coverage and a 10x LTA multiple; about 10.5x-10.7x at 80% coverage and 12xThese figures correspond to the implied group P/E ranges for Samsung and SK hynix, showing that the market may still not be assigning a meaningful premium to LTA-supported earnings.
• Base valuation assumptionsCost of equity 11.5%; risk-free rate 5%; equity risk premium 6.5%; beta 1.0; terminal growth rate 3%Used in the base-case residual income model for SK hynix.

Impact & implications

If LTAs are cash-backed and strongly enforceable, earnings volatility in the memory industry may be lower than in historical cycles, while companies' free cash flow visibility and capital return capacity improve. The market may gradually treat LTA-supported earnings as higher-quality revenue and assign valuation premiums. For SK hynix, its leadership in HBM, a recovery in commodity memory prices, and downside protection under LTAs together support the target price and EPS upgrades; for the industry as a whole, LTAs may also create a higher "memory tax" for non-memory customers and the AI infrastructure supply chain, meaning they pay more or more stable costs to secure supply.

Risks

• The macro environment is weaker than expected, leading to softer end demand.
• HBM competition intensifies, creating pressure on pricing, margins, and market share.
• Rising DDR5 competition triggers overinvestment on the supply side.
• Cloud customers and Chinese smartphone customers keep inventory levels elevated.
• If LTA terms are not sufficiently enforceable, prepayment realization is unclear, or customers renegotiate, the valuation re-rating logic may be impaired.
• A temporary digestion of AI compute demand may suppress 2026 demand and HBM margins.
• If the supply-demand outlook for DRAM, NAND, and HBM is too optimistic, the expected EPS and free cash flow upgrades may not materialize.

What to watch

• Whether memory manufacturers' balance sheets show material cash inflows and deferred revenue obligations in 2026, validating actual LTA prepayments.
• Further confirmation from SK hynix on HBM pricing, supply lock-ins, and customer agreements.
• Supply-demand outlooks for DRAM, NAND, and HBM, especially AI inference demand and data center product demand.
• Whether EUV equipment supply bottlenecks and ASML deliveries constrain DRAM supply expansion.
• Disclosures from Samsung, Micron, SanDisk, Kioxia, and other manufacturers regarding multi-year agreements, prepayments, and margin protection terms.
• Whether memory manufacturers' free cash flow, buybacks, dividends, and capital return policies improve.
• Whether the market begins to assign a higher P/E multiple to LTA-supported earnings than to traditional commodity memory earnings.