美国IT硬件存储：新型长期协议之二——微芯科技等半导体长期协议案例研究，此次有何不同？

Memory: New Memory LTAs part 2 - Case studies from Microchipand other Semiconductor LTAs - What's different this time? Last week we published an analysis onNew Memory LTAs(Long-term agreements), whythey are different and what this means for sustainability of earnings. Today’s part 2 divesdeeper intoCase Studies from the Semiconductor industryto compare and addressquestions and feedback from investors. Mark C. Newman+1 212 845 7822mark.newman@bernsteinsg.com April Li+1 917 344 8339april.li@bernsteinsg.com Historically, long-term agreements in semiconductors and memory functionedprimarily as a buyer's tool: Customers used LTAs to secure guaranteed volume,while sellers had little structural leverage of their own. Even when suppliers held somenegotiating power, contracts were largely limited to two structures: take-or-pay, where abuyer promises to either accept a minimum volume at a pre-agreed price or pay a penaltyfor any shortfall; and so-called "non-cancelable" orders, a buyer's promise not to canceldelivery that, as later cycles would prove, buyers could and did cancel in practice.Bothstructures offered sellers little real downside protection, since neither carried anupfront financial commitmentto compensate the supplier if a buyer walked away; theonly remedy was to sue for damages, a process that was slow, expensive, and often hollowif the buyer was no longer solvent by the time judgment was reached. Phoebe Sun+1 917 344 8481phoebe.sun@bernsteinsg.com Two case studies illustrate exactly how previous LTA structures failed in practice: 1. Microchip / Analog Semiconductors; and 2. Hemlock / Polysilicon. Out of the two casestudies, the Hemlock case bears more resemblance to what we see in the New MemoryLTAs as there were some legal teeth to the agreement but very much relied on suing itscustomers to get the money owed to it. The Microchip case study, which is often referencedby Semiconductor investors, was a much looser agreement without any notable legal teeththat customers had little disincentive to walk away from. We identify three structural differences between legacy semiconductor LTAs andthe New Memory LTAs, each addressing a distinct weakness that undermined priorcontract structures:1. Upfront financial commitment:cash deposits, letters of credit,or third-party guarantees, specificallyback-end weighted(where the coverage ratio ofremaining financial commitment as a portion of remaining purchase obligations increasessubstantially over time, hence increasing downside protection in the latter part of thecontract when it is more needed);2. Better counterparty quality: financially sound,diversified buyers replace the thinly capitalized or diffuse customer bases that underminedHemlock and Microchip;3. Structural demand:sustained AI-driven buildout replaces thetemporary pull-in and double-ordering that ultimately undermined Microchip. The New Memory LTAs provide meaningful but not infinite downside protection.Wefind that these new memory LTAs are structured in a way which provides significantly higherdisincentives to customers to walk away - particularly in the latter part of the contract term.In our previous note (New Memory LTAs),we showed a range of scenarios of future pricecollapse and how those would impact earnings with and without LTAs. In conclusion, thereissignificantly more downside protectionin these agreements than the previous caseswe analyzed, but not infinite downside protection. BERNSTEIN TICKER TABLE Table Of Contents Part I: Blast from the Past - Case Studies in earlier Semiconductor LTAs...........................................................................................................3Case Study 1: Microchip / Analog - non-cancelable/non-reschedulable supply program led to inventory glut............................ 3Case Study 2: Hemlock / Polysilicon - take-or-pay contract failure.............................................................................................................. 4Part II: Key Differences: Old Semiconductor LTAs vs. the New Memory LTAs....................................................................................................6Upfront Financial Commitment: Real Teeth, Not Just a Promise.....................................................................................................................6Fundamentally better Counterparty quality.............................................................................................................................................................8Limited Demand vs. Structural Scaling.....................................................................................................................................................................8 DETAILS PART I: BLAST FROM THE PAST - CASE STUDIES IN EARLIER SEMICONDUCTOR LTAS Historically, long-term agreements in semiconductors and memory functioned primarily as a buyer's tool: customersused LTAs to secure guaranteed volume, while selle