New memory LTAs raise the earnings floor, but cannot permanently lock in cycle peaks
AI summary card
New memory LTAs raise the earnings floor, but cannot permanently lock in cycle peaks
Bernstein's bull-bear debate concludes that long-term memory supply agreements can soften downcycles and improve revenue visibility, but the strength of contract protection depends on deposit size, the magnitude of spot-price declines, coverage rates, and counterparties' willingness to perform.
- LTA protection is effective only when the cost of breaching exceeds the savings a customer would gain by switching to lower-priced spot purchases.
- The new generation of memory LTAs differs from historical semiconductor LTAs by including upfront financial commitments, back-end weighted deposits, or third-party guarantees.
- Hyperscalers and AI infrastructure customers have stronger balance sheets and a stronger strategic motivation for supply security, making performance credibility better than in historical cases.
- The bear case argues that disclosed deposits remain small relative to the scale of multi-year revenue protection, with the current US$33B of deposits amounting to only about 0.6% of the amount to be protected.
- The report remains constructive on near- to medium-term memory supply and demand, but disagreement is concentrated on whether NAND can continue benefiting from AI demand in the same way as DRAM/HBM.
Report interpretation
Overview
This report summarizes Bernstein's July 9 webinar, "New Memory LTAs: glass half full or half empty?", including the bull-bear debate on new long-term memory supply agreements, related slides, Q&A, and investment implications. The core issue is whether new memory LTAs can change the highly volatile and highly cyclical earnings structure of the memory industry. The report's balanced view is that LTAs will not eliminate memory cycles, but may materially reduce the depth of downcycles and improve visibility on suppliers' revenue, profits, and valuations.
Core views
The report argues that the value of new memory LTAs lies not in permanently locking in peak earnings, but in lifting the earnings floor. The bull case emphasizes that the new agreements feature upfront financial commitments, back-end weighted deposits, stronger-credit hyperscaler/AI infrastructure customers, and structural demand driven by real AI deployment, making them more reliable than the failed semiconductor LTAs of the past. The bear case points out that if spot prices fall sharply relative to the contract floor price, customers may still rationally breach as long as remaining purchase volumes are large; at the same time, the currently disclosed deposit scale is too small relative to multi-year revenue protection needs, and LTA coverage cannot reach the entire market.
Analysis framework
The report analyzes LTAs using a bull-bear comparison framework: on one hand, it evaluates whether customers would exit agreements based on contract economics, remaining performance obligations, price floors, remaining deposits, and incentives to breach; on the other hand, it assesses the impact of new memory LTAs on earnings volatility and valuation multiples based on historical LTA failure cases, counterparty quality, the structural nature of AI demand, memory price stress tests, and company EPS sensitivity.
Methodology notes
When the savings from switching to spot purchases exceed the remaining financial guarantee, breach may become rational.
The report provides formulas including: RPO minus market price times remaining volume exceeds the remaining financial guarantee, or the difference between the contract floor price and market price times remaining volume exceeds the remaining financial guarantee.
LTAs cannot eliminate cycles, but they can improve historically deep loss cycles into near break-even or more bearable trough earnings.
The report believes deposit forfeiture can improve trough margins, but maintaining peak earnings still requires structural supply-demand shortages.
Customers in the new LTAs are mainly hyperscalers, large OEMs, and AI infrastructure providers, with credit quality materially better than in historical cases.
Compared with financially stressed customers in historical cases such as Hemlock, counterparties in the new agreements have stronger balance sheets, more diversified businesses, and higher recoverability of shortfall payments.
AI training, inference, expanding context windows, and rising storage demand may make memory demand more structural than in past cycles.
The report believes demand in this cycle is driven more by real AI infrastructure deployment than by inventory hoarding, double ordering, or pandemic-related supply-chain distortions.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- SNDKOne of the core beneficiary names, actively signing LTAs with financial guarantees
- Strengths
- Target price US$3,000, rated Outperform; 5 LTAs are expected to cover about one-third of FY27 bit demand, and the agreement structure provides meaningful but not unlimited downside protection.
- Weaknesses
- The consumer business is more transactional, making LTAs less applicable; protection remains limited if NAND demand disappoints or prices fall sharply.
- Comparison
- Compared with Asian suppliers, SanDisk and Micron are more aggressive in advancing deposit-backed LTAs.
- Risks
- Long-term Chinese competition in NAND, insufficient LTA deposits, customer breach, or a rapid decline in spot prices.
- MUOne of the core beneficiary names, with the most proactive LTA disclosures
- Strengths
- Rated Outperform with a target price of US$1,300; among 16 LTAs, 14 have minimum-price RPO of about US$100B, with financial guarantees of about US$22B.
- Weaknesses
- Micron's equity-linked transaction with Anthropic is viewed in the report as a case of supplier financing and amplified risk.
- Comparison
- Along with SanDisk, it is among the suppliers most aggressively advancing new memory LTAs.
- Risks
- Weak protection in early-stage contracts, customer financing risk, and a rapid short-term decline in memory prices.
- Samsung ElectronicsLarge Asian memory supplier, rated Outperform
- Strengths
- Target price KRW 440,000; positioned to obtain more deposits over time and benefit from the DRAM/HBM cycle and AI demand.
- Weaknesses
- The report says Asian suppliers are more selective in advancing LTAs, and the pace of disclosure may be slower than SanDisk/Micron.
- Comparison
- Compared with the more optimistic U.S. team's view on NAND, the Asian team is more cautious on long-term NAND competition.
- Risks
- NAND competition, insufficient LTA coverage, and overly rapid supply expansion.
- SK hynixLarge Asian memory supplier, rated Outperform
- Strengths
- Target price KRW 3,300,000; benefits from HBM and AI-driven DRAM demand.
- Weaknesses
- Disclosure of LTA financial guarantees is less detailed than for Micron/SanDisk.
- Comparison
- Together with Samsung, it is a core Asian memory supplier and may continue to secure deposit support.
- Risks
- HBM price and demand volatility, insufficient LTA disclosure, and supply-demand reversal caused by industry capex.
- KIOXIAHas higher NAND exposure and is rated Underperform
- Strengths
- Plans to cover about 50% of shipments in CY28, implying some LTA protection potential.
- Weaknesses
- Rated Underperform, with a target price of JPY 40,000 below the table's current price of JPY 52,110; structural debate around NAND is greater.
- Comparison
- Compared with companies that benefit more directly from DRAM/HBM, KIOXIA faces higher NAND cyclicality and competitive pressure.
- Risks
- Declining NAND prices, competition from Chinese suppliers, and insufficient LTA protection.
Key data
- Micron LTA countA total of 16 agreements as of June 2026, including 4 very large customers and 3 mid-sized customersTable disclosure indicates that the minimum-price RPO for 14 agreements is about US$100B.
- SanDisk LTA count5 agreementsAmong them, 3 were signed in FQ3 and 2 were signed in FQ4 as of the report date; the RPO for 3 agreements is about US$42B.
- Disclosed financial guaranteesMicron about US$22B, SanDisk about US$11B, totaling about US$33BThe bear case argues that this scale represents only a small portion of the multi-year revenue scale to be protected.
- Potential protection scaleIf LTAs protect US$1.3T of annual revenue in CY27-CY28 and extend for 3-5 years, about US$5.2T would need to be protectedThe report states that US$33B of deposits is about 0.6% of this amount.
- LTA coverage assumptionsMicron is expected to place half or more of revenue under LTAs once completed; SanDisk's 5 agreements are expected to cover about one-third of bit demand in FY27; KIOXIA plans to cover about 50% of shipments in CY28Consumer business, transactional buyers, and some Chinese customers may be difficult to fully include in LTAs.
- Ratings and target pricesSNDK US$3,000; Samsung Electronics KRW 440,000; SK hynix KRW 3,300,000; Micron US$1,300; KIOXIA JPY 40,000The corresponding ratings are Outperform for SNDK, Samsung, SK hynix, and Micron, and Underperform for KIOXIA.
Impact & implications
The investment implication is that whether memory stock valuations expand depends on whether investors can believe LTAs truly reduce downside risk, rather than merely providing a narrative at the top of the cycle. If contract coverage exceeds 50%, back-end deposit protection gradually strengthens, AI demand persists, and supply remains disciplined, LTAs could support a higher earnings floor and higher valuation multiples; but if a sharp downcycle occurs in the short term, spot prices fall far below contract floor prices, or deposits are insufficient to constrain customers, share prices may still trade like a traditional memory downcycle.
Risks
- If spot memory prices fall well below contract floor prices, the savings from customer breach may exceed deposits.
- The currently disclosed deposit scale is small relative to multi-year revenue protection needs and may be insufficient to sustain peak earnings.
- LTAs are unlikely to cover all end markets; consumer, transactional buyers, and some Chinese customers may remain outside LTA procurement.
- If a downcycle occurs early in the contract life, back-end weighted deposits may not yet provide sufficient protection.
- Supplier financing or equity investments in customers may amplify credit and capital allocation risks.
- The NAND segment may face higher competition risk from China, and there is disagreement over how much AI demand will support NAND.
- If industry capex expands too quickly, LTAs cannot substitute for genuine supply-demand shortages.
What to watch
- Further disclosures from memory suppliers on LTA count, RPO, deposits, and contract duration.
- SanDisk's upcoming Analyst Day and its explanation of LTAs, NAND demand, and earnings sustainability.
- Memory ASP trends, especially whether NAND and DRAM/HBM pricing diverge.
- Whether Micron, SanDisk, Samsung, and SK hynix can raise LTA coverage to 50% or above.
- Whether AI training, inference, context windows, and storage demand continue to drive real memory consumption.
- The impact of Chinese CSPs and Chinese memory suppliers on LTA signing and the supply landscape.
- The timing of the next downcycle: if it occurs too early, the value of back-end LTA protection may not yet be visible.