Why Shares of Sandisk Fell This Week

Google unveiled a new LLM compression algorithm that reportedly cuts memory requirements by more than 6x, and Sandisk shares dropped 14.1% this week on investor concerns. Large tech firms are collectively planning roughly $650B in AI-related capex this year, leaving open the question whether compression reduces total memory demand or simply enables larger, more memory-hungry models. Geopolitical risk from the war in Iran and an OECD projection of U.S. inflation rising to 4.2% in 2026 add recession and macro risk, so a wait‑and‑see approach on Sandisk is advised.

Analysis

The market reaction is priced like a near-term structural demand shock for commodity flash, but two multi-year levers work in opposite directions: (1) efficiency gains at the model level can compress raw capacity need per instance, and (2) cheaper per-instance cost can dramatically increase the number of instances deployed (more models, wider inference fleets). Expect this to produce substantial dispersion across memory product types — high-bandwidth, low-latency components and controllers should hold value even if bulk NAND demand softens.

Inventory and capex timing are the critical mediators. Server/storage vendor inventory cycles typically play out over 2–4 quarters; a coordinated hyperscaler pause can knock supplier revenues for a season, but balance-sheet-driven buybacks or price cuts can restore order flow within 6–12 months. Watch sequential billings and spot ASP moves (>=10% swings) as leading indicators — these will move faster than quarterly commentary.

Geopolitics and inflation are credible tail risks that could convert a technical correction into a structural pullback if cloud providers materially re-assign capex to energy/defense or slow deployments. That said, the current repricing looks overstretched relative to plausible downside scenarios; if hyperscalers validate efficiency in production, the net effect could be neutral-to-positive for total memory dollars over 12–36 months as compute density and inference scale accelerate.

Key Decisions for Investors

• Contrarian LEAP (SNDK): Buy SNDK Jan-2028 LEAP calls (1–2 year horizon). Thesis: patience on inventory digestion + secular storage growth. Risk: premium loss if demand permanently compresses; Target: 2x–3x if normalization/reshoring drives procurement within 12–24 months.
• Momentum compute play (NVDA): Buy NVDA Jan-2027 call spread (debit spread to cap risk) to express continued accelerator demand. Timeframe 6–18 months. Risk: capped to premium; Reward: 1.5x–3x if hyperscaler GPU orders sustain.
• Relative-value pair: Long NVDA Jan-2027 1x call spread vs short SNDK cash or buy SNDK puts (6–9 months). Purpose: capture rotation to high-margin compute over commodity flash. Risk: SNDK rebounds (cut losses at 30–40% adverse move); Reward: asymmetry if NVDA outperforms and SNDK remains re-rated.
• Tactical hedge for portfolios: If net-long tech exposure, buy 3–6 month SNDK puts or reduce SNDK exposure ahead of next hyperscaler capex prints. Use this as a low-cost tail hedge to protect against a multi-quarter capex pullback driven by macro/geopolitics.