The new wave of storage LTAs is materially different from legacy semiconductor LTAs
AI summary card
The new wave of storage LTAs is materially different from legacy semiconductor LTAs
Bernstein argues that the new storage LTAs offer more meaningful but not unlimited downside protection for memory suppliers than legacy semiconductor LTAs, through upfront commitments, stronger counterparties, and structural AI-driven demand.
- Legacy semiconductor LTAs are often buyer-favorable, with limited upfront collateral or immediately realisable protection, and supplier remedies after buyer default mainly rely on lengthy litigation.
- The Microchip/Analog case shows that during COVID, analog and MCU orders included significant pre-buying and repeat ordering; as supply pressure eased, this quickly became inventory overhang, and so-called non-cancelable orders were ultimately suspended or delayed.
- The Hemlock polysilicon case shows that even with strong legal terms, if the buyer is financially weak and commodity prices collapse, a supplier may still fail to fully recover cash after a legal win due to customer bankruptcy.
- The key difference in the new storage LTAs is upfront financial commitments such as cash deposits, letters of credit, or third-party guarantees, and the protection is more focused on the contract tail.
- The report argues that HBM, DRAM, and NAND demand from AI and data-center buildout has a structural expansion profile, unlike the temporary pull-through seen in analog chip inventory cycles.
Report interpretation
Overview
This report is part 2 of Bernstein's new storage LTAs series. It focuses on the Microchip/Analog semiconductor supply program and the Hemlock polysilicon take-or-pay contract as historical examples, comparing why legacy semiconductor LTAs failed and explaining why current storage LTAs are structurally different. The core conclusion is that new storage LTAs do provide stronger downside protection, but this protection is not unlimited.
Core views
The report believes legacy semiconductor LTA failures stem mainly from three flaws: insufficient real upfront financial commitment, weak or overly fragmented counterparty quality, and demand that is inherently a short-term pull-through rather than structural expansion. By contrast, new storage LTAs feature cash deposits, letters of credit, or third-party guarantees, with guarantee coverage increasing at the back end as remaining obligations decline; counterparties are mostly financially sound hyperscalers and large OEMs; and demand is driven by AI and data-center expansion, with HBM, DRAM, and NAND viewed as critical AI capex items.
Analysis framework
The report uses a case-comparison method: it first revisits how Microchip’s Preferred Supply Program broke down as analog and MCU inventory overhang built up, then revisits how the Hemlock polysilicon contract led to insufficient commercial recoveries despite legal victory due to price collapse, trade frictions, and customer bankruptcy; it then structurally compares legacy LTAs and new storage LTAs across three dimensions: collateral structure, counterparty quality, and demand attributes.
Methodology notes
Upfront financial commitments and back-end weighted protection
The report emphasizes that new storage LTAs are not merely legal claims; they create immediately usable collateral through cash deposits, letters of credit, or third-party guarantees, and the share of remaining obligations covered rises toward the contract back end.
Stronger counterparties reduce the risk of one-sided legal wins
The Hemlock case shows that winning a judgment does not equal collection; new storage LTAs have more concentrated, larger, and financially stronger counterparties, reducing recovery risk after default.
Demand ceiling differences
Analog and MCU demand is constrained by end-market production, so incremental value is limited once inventories are sufficient, while AI systems’ memory capacity and bandwidth needs may continue to rise with model size, context window, and concurrent inference demand.
Asset mapping & comparison
Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).
- SanDisk Corp / SNDKCore beneficiary and valuation subject
- Strengths
- The report believes its remaining obligations are supported by over $11 billion in guarantees, and the new storage LTAs improve earnings sustainability.
- Weaknesses
- Near-term performance may be elevated, and NAND still has cyclical downside risk; company disclosures and investor communication are seen as disorganized.
- Comparison
- Compared with legacy semiconductor LTAs, SanDisk-related agreements have more meaningful upfront protection and stronger counterparties.
- Risks
- If NAND weakness extends beyond the current cycle and turns structural, DCF value and asset liquidation value could be materially below expectations.
- MicronKey reference case for new storage LTA structure
- Strengths
- The report says obligations are backed by a $18 billion cash deposit and $4 billion letter of credit, indicating the contracts have meaningful financial constraints.
- Weaknesses
- It remains exposed to storage price cycles and volatility in AI demand expectations.
- Comparison
- Compared with Microchip-style non-cancelable orders, Micron-related arrangements impose higher customer exit costs.
- Risks
- If storage prices fall sharply or AI capex slows, willingness to perform under contracts and earnings expectations could still come under pressure.
- Microchip TechnologyLegacy semiconductor LTA failure case
- Strengths
- It obtained priority supply commitments and built up 12 months of non-cancelable, non-postponable orders through the Preferred Supply Program.
- Weaknesses
- The agreements lacked sufficient legal and financial constraints, allowing customers to suspend or defer orders after inventory became excessive.
- Comparison
- Compared with new storage LTAs, the Microchip case relied more on buyer commitments than upfront collateral.
- Risks
- Once the pull-through and repeat-order cycle reverses, inventory normalization can cause a sharp contraction in revenue.
- Analog DevicesLegacy analog semiconductor cycle case
- Strengths
- Reduced channel inventories by managing shipments, preventing a further market collapse.
- Weaknesses
- Experienced multi-quarter revenue contraction during the inventory correction period.
- Comparison
- Its demand profile is closer to terminal-output constraints and does not have the sustained expansion profile of AI storage demand.
- Risks
- Inventory cycles, end-market demand deceleration, and customer de-stocking.
- Hemlock SemiconductorLegacy take-or-pay contract failure case
- Strengths
- Contract legal terms were relatively strong, ultimately yielding judgments or settlements in some cases.
- Weaknesses
- Protection depended on litigation and buyer repayment ability, with actual cash recovery fragmented and diluted.
- Comparison
- Compared with new storage LTAs, Hemlock lacked upfront collateral that could be immediately drawn and support from high-quality counterparties.
- Risks
- Commodity price crash, customer bankruptcy, trade frictions, and commercially unenforceable contracts.
Key data
- Microchip Technology sales declinemore than 42%In Q3 FY25 versus Q3 FY24, during the peak of inventory normalization.
- Polysilicon spot price peakover $400/kgPrice level during the severe mid-2000s shortage.
- Hemlock contract pricearound $40-$60/kgFixed-price range in long-term agreements.
- Polysilicon spot price after crashbelow $15/kgMarket prices fell sharply after Chinese supply expansion and trade friction.
- SolarWorld Sachsen judgment amountaround $800 millionHemlock won in court but recovery may be insufficient because the customer was insolvent.
- Kyocera settlement amount$450 million, or about 51.1 billion yenOut-of-court settlement paid in 2018 to extinguish remaining LTA commitments.
- SanDisk remaining performance obligations and guaranteesabout $69 billion in obligations, with over $11 billion in guaranteesThe report says most guarantees sit on third-party balance sheets in a custodial-like form.
- Micron financial commitments$18 billion cash deposit plus $4 billion letter of creditUsed to support its related obligations.
- SNDK target price$3000The report discloses valuation at 11x FY28 EPS or 14x FY26-30 average EPS.
Impact & implications
If the report's thesis is correct, the market may be underestimating the role of new storage LTAs in the earnings sustainability and cyclical downside protection of storage suppliers, especially for companies like SanDisk and Micron that benefit from long-term storage-demand lock-in. The report also stresses that protection is not unlimited, and persistent NAND cyclic weakness, disclosure and investor communication issues, and potential structural NAND fatigue may still pressure valuations.
Risks
- New storage LTAs provide meaningful but not unlimited downside protection, and extreme price collapses or demand reversals can still weaken that protection.
- Near-term SNDK performance may be elevated and subject to cyclical downside.
- SNDK disclosures and investor communication appear disordered, which may affect higher-quality investor participation.
- If NAND weakness persists and becomes structural, SanDisk's DCF value and asset damage value could be materially below expectations.
- If AI capex or hyperscaler demand comes in below expectations, the sustainability assumptions of storage LTAs will be challenged.
- Even with legal clauses, recovery may still be impaired if counterparty credit quality deteriorates.
What to watch
- SanDisk and Micron disclosures on LTA collateral, cash deposits, letters of credit, and third-party guarantees.
- The degree of HBM, DRAM, and NAND supply-demand tightness to 2026 and whether premium memory remains sold out.
- Hyperscaler AI capex plans, data-center buildout pace, and long-term storage procurement commitments.
- Whether NAND price weakness is only a short-term drawdown or evolving into structural fatigue.
- Whether customers begin delaying performance, renegotiating contracts, or seeking to exit LTAs.
- Whether investors re-rate memory suppliers’ earnings sustainability and valuation multiples.