New memory LTAs raise the industry's earnings floor, but cannot end the memory cycle

Overview

This report is Bernstein's third installment on new memory long-term agreements, based on the July 9, 2026 webinar "New Memory LTAs - glass half full or half empty?", integrating presentation materials, Q&A records, and a summary of views. The report centers on a bull-bear debate over new LTAs signed between memory suppliers and customers, with the core question being whether these agreements can reduce earnings volatility in the memory industry, improve valuation multiples, and provide suppliers with greater revenue and profit visibility during the AI infrastructure demand cycle.

Core views

The report's balanced conclusion is that LTAs will not eliminate the memory cycle, nor can they guarantee that suppliers will sustain peak earnings over the long term, but they can significantly reduce the severity of downcycles. The strength of protection depends on the dynamic relationship among the contract floor price, market price, remaining purchase obligations, and remaining financial guarantees. If market prices fall sharply, customers may still choose to walk away if the savings exceed the deposit; however, the new LTAs are more binding than historical agreements because they include prepayments, letters of credit, or third-party guarantees, and protection often strengthens in the later stages of the contract.

Analysis framework

The report adopts a bull-bear debate framework: the bear case assesses the limitations of LTAs from the perspectives of contract coverage, deposit size, customer incentives to exit, and insufficient current disclosure; the bull case argues for the protective value of LTAs from the angles of agreement structure, counterparty quality, AI-driven structural demand, and stress testing. The report also combines rating and target price tables for SanDisk, Micron, Samsung, SK hynix, and KIOXIA to discuss the potential impact of LTAs on memory stocks, semiconductor equipment, and industry valuations.

Methodology notes

• Industry cycle analysisMemory LTA bull-bear framework

  Whether long-term agreements can cushion the cycle

  When evaluating LTAs, one cannot simply look at whether a contract exists; one must compare the savings from customer default with the remaining deposit, contract floor price, market price, and remaining purchase volume.

• Contract economicsExit incentive test

  When customers will exit an LTA

  If (the contract floor price or RPO price minus the market price) multiplied by the remaining purchase volume exceeds the remaining financial guarantee, the customer may choose to exit on purely economic grounds.

• Risk buffer analysisBack-end-weighted deposits

  Stronger protection in the later stage of the contract

  As remaining purchase obligations decline, if deposits or collateral remain at a relatively high level, the cost of forfeiting the deposit relative to the remaining contract value rises, making default thresholds higher in the later stage.

• Demand structure analysisAI infrastructure demand driver

  AI creates real demand rather than inventory-driven demand

  The report argues that current memory demand is driven more by AI training, inference, data center deployment, and larger context windows, rather than by stockpiling or duplicate ordering seen in historical cycles.

Asset mapping & comparison

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

• SNDK.US

  Core beneficiary, covered by the report with an Outperform rating and USD 3,000 target price

  Strengths

  Actively promotes LTAs with financial guarantees; NAND may benefit from a better supply-demand backdrop under AI storage demand and constrained supply; back-end-weighted deposits can strengthen later-stage protection.

  Weaknesses

  Compared with DRAM, NAND is more affected by Chinese competition and customer coverage constraints; if spot prices fall sharply, LTAs may still be insufficient to lock in peak earnings.

  Comparison

  Compared with Asian suppliers, SanDisk and Micron are more proactive on financial-guarantee LTAs; compared with KIOXIA, Bernstein is clearly more positive on SNDK.

  Risks

  Insufficient LTA coverage, inadequate deposit size, intensifying long-term NAND competition, rapid demand weakening, or customer exit.

• MU.US

  Industry benchmark and active LTA participant

  Strengths

  Actively pursues LTAs with financial guarantees; HBM and AI-related DRAM demand still have structural support.

  Weaknesses

  Micron's deposit return arrangements differ from SanDisk's, which may affect the strength of long-term protection; its equity investment in Anthropic reflects supplier financing risk.

  Comparison

  Like SanDisk, it belongs to the more aggressive contracting camp; compared with Samsung and SK hynix, its disclosures and path are clearer.

  Risks

  Deposit refunds reducing binding force, capex commitments extending execution duration, and cyclical downturns compressing EPS.

• Samsung Electronics

  Global memory leader and covered name in the rating table

  Strengths

  Advantages in scale, balance sheet, and product mix, with the ability to secure more deposits and LTA opportunities.

  Weaknesses

  The report indicates that Asian suppliers as a whole are more selective on LTAs, and current disclosure is less proactive than SanDisk and Micron.

  Comparison

  More cautious in advancing LTAs than SanDisk/Micron; however, as a DRAM/NAND leader, it is still supported by AI memory demand.

  Risks

  LTA coverage, pricing cycles, Chinese competition, and capex pacing.

• SK hynix

  HBM and DRAM beneficiary, rated Outperform in the rating table

  Strengths

  Strong AI/HBM demand structure, relatively high-quality counterparties, and tight industry supply-demand conditions.

  Weaknesses

  Limited disclosure of LTA financial guarantees, with protective effect dependent on subsequent contract details.

  Comparison

  Compared with names with higher NAND exposure, DRAM/HBM benefits more directly from AI demand.

  Risks

  HBM pricing cycles, supply expansion, customer bargaining power, and insufficient contract coverage.

• KIOXIA Holdings

  NAND-related name, rated Underperform in the rating table

  Strengths

  Participates in the global NAND supply system and may benefit from an improving industry pricing cycle.

  Weaknesses

  The Asia team in the report is more cautious on NAND, believing AI demand is more tilted toward DRAM, while long-term Chinese competition risk is greater.

  Comparison

  Compared with SNDK, Bernstein's stance on KIOXIA is clearly more negative.

  Risks

  NAND price declines, Chinese competition, insufficient LTA coverage, and earnings cycle volatility.

• Semiconductor equipment

  Indirect beneficiary asset class

  Strengths

  Long-term supply commitments and capex announcements from Micron and others may support equipment and construction demand.

  Weaknesses

  Benefits to equipment do not equate to sustainably high earnings for memory stocks, and the demand realization cycle is longer.

  Comparison

  In the report, Mark Li believes that if Micron's investment commitments are meaningful, they are more positive for semicap and the U.S. construction chain.

  Risks

  Capex delays, non-binding announcements, and excessive supply-demand expansion.

Key data

• SNDK rating and target priceOutperform, target price USD 3,000.00, current price USD 1,354.82The pricing table is dated 2026-07-17; the body of the report states that SNDK is rated Outperform with a target price of USD 3,000.
• Micron rating and target priceOutperform, target price US$1,300.00The report lists Micron as Outperform and discusses its investment related to Anthropic as an example of supplier financing and amplified risk.
• Samsung Electronics rating and target priceOutperform, target price KRW 440,000The report lists Samsung Electronics as Outperform.
• SK hynix rating and target priceOutperform, target price KRW 3,300,000The report lists SK hynix as Outperform.
• KIOXIA rating and target priceUnderperform, target price JPY 40,000The report lists KIOXIA as Underperform, and the Asia team is more cautious on NAND.
• Disclosed deposit sizeapproximately US$33BThe bear case points out that this is only about 0.6% relative to the multi-year revenue base that may need protection.
• Potential protected revenue scaleapproximately US$5.2TBased on market expectations for memory company revenue in CY27 and CY28 and an assumed protection period of 3-5 years.
• Upper limit of LTA coverage30%-50% of the market may never be covered by LTAsConsumer electronics, transactional customers, and some Chinese customers may prefer flexible sourcing or domestic suppliers.
• Earnings floor in a downcycleOPM may improve from historically deeply negative levels to around the 0%-30% rangeThe report emphasizes that this only makes the trough more bearable and does not mean maintaining peak earnings.

Impact & implications

For investors, LTAs enhance downside protection and visibility for proactive contracting suppliers such as SanDisk and Micron, potentially supporting higher valuation multiples; however, their value is more like earnings floor protection than peak earnings lock-in. If investors gradually come to believe that LTAs can reduce tail downside risk, memory stock valuations may benefit; if demand weakens rapidly early in the contract term, share prices may still come under pressure because deposit protection is weaker. The semiconductor equipment chain may benefit from customers' long-term supply commitments and incentives for capacity expansion, but memory stocks still require continued supply-demand shortages to sustain high earnings.

Risks

• LTA deposit size may be insufficient to cover multiple years of peak revenue or severe price declines.
• If memory spot prices fall far below contract prices, customer savings may exceed the cost of forfeiting deposits, leading them to exit.
• There is an upper limit to LTA coverage, and consumer electronics, transactional customers, and some Chinese customers may not be suitable for or willing to sign long-term agreements.
• Protection is weaker in the early stage of contracts, so if demand weakens rapidly in the near term, share prices and earnings may still face significant pressure.
• NAND faces long-term structural divergence from Chinese competition and AI demand skewing toward DRAM/HBM.
• Supplier financing or exchanging equity investments for customer commitments may amplify risk.
• If investors misread LTAs as locking in peak earnings, they may overestimate the sustainable EPS and valuation multiples of memory stocks.

What to watch

• SanDisk's upcoming Analyst Day and its disclosures on LTA terms, coverage, and deposit structure.
• Whether Micron, SanDisk, Samsung, and SK hynix subsequently disclose more LTA financial guarantees and customer scope.
• Memory ASP trends, especially whether NAND and DRAM/HBM pricing can maintain near- to medium-term strength.
• AI infrastructure capex, training and inference demand, and storage demand driven by expanding context windows.
• Changes in the share of Chinese memory suppliers in NAND and among domestic customers.
• Whether contract execution sees customer exits, renegotiations, litigation, or supplier financing arrangements.
• Whether the market raises valuation multiples for memory stocks due to a higher earnings floor.