Why Shares of Sandisk Fell This Week

Sandisk shares fell 14.1% this week after Google unveiled an LLM compression algorithm that reportedly reduces memory requirements by more than 6x, triggering fears of weaker memory demand. Investors are also citing macro risks from the war in Iran and an OECD projection that U.S. inflation could reach 4.2% in 2026; the piece recommends a wait-and-see posture for Sandisk as the technology impact remains uncertain.

Analysis

Algorithmic compression is a demand-deflation threat for raw memory bytes but not a terminal event for the AI-capex complex — the immediate mechanism is price-per-usable-parameter, not absolute compute growth. Lower memory per model can free rack-level capacity and reduce per-model storage needs, which often gets redeployed into greater model breadth, more replicas for latency-sensitive inference, or additional training iterations; any of those paths increases demand for accelerators, HBM/DRAM bandwidth, interconnect, and high-performance NVMe rather than commodity NAND alone.

The clearest second-order winners are suppliers of high-bandwidth memory and system-level integration (HBM, PCIe 5/6 controllers, NVMe enterprise SSD controllers) and hyperscalers that capture displacement savings as budget to buy more GPUs/TPUs. Losers cluster where unit economics are most exposed to density-driven price erosion: commodity client/embedded flash and any vendor sitting on multi-quarter NAND inventory bought at peak pricing. Separately, a geopolitical/inflation shock compresses cloud capex on a 0–12 month horizon, which could turn a nuanced structural shift into a cyclical drawdown for memory vendors.

Key near-term catalysts: cloud capex guidance from MSFT/AMZN/META over the next 1–3 quarters, public benchmarks showing compression throughput/latency trade-offs over 0–6 months, and NAND channel inventory reports and pricing trends over 3–9 months. A durable reversal comes if compression enables a step-change in model proliferation (more models × more inference) within 6–18 months or if benchmarks reveal hidden compute overheads that push designs back toward higher-memory architectures.

Key Decisions for Investors

• Tactical short SNDK via 3–6 month puts sized 1–2% NAV: payoff asymmetric if NAND repricing and inventory markdowns materialize (target 30–50% downside in equity, pain-limited to premium paid). Monitor sell-side inventory reports and set stop at 20% loss on premium.
• Pair trade: short SNDK / long MSFT (equal notional) for 6–12 months — expresses rotation from commodity storage to cloud compute. Reward: capture reallocation of capex if hyperscalers increase accelerator spend; risk: macro-driven capex cuts that hurt both legs (trim if cloud guidance falls >5% QoQ).
• Long NVDA 9–18 month calls (or LEAPS) to play higher inference volume and accelerator scarcity; size 2–4% NAV. Upside >2x if model proliferation drives continued GPU demand; hedge with a cheap out-of-the-money put spread to protect vs. sharp macro drawdown.
• If geopolitical/inflation risk increases, switch part of short-SNDK exposure to a hedge: buy a protective put on NVDA or add 2–3% NAV in short-duration Treasuries/credit protection for 3–9 months to lower portfolio duration risk.