Alphabet Just Crashed The Memory Trade: Sandisk Looks Like The Winner (Upgrade)

TurboQuant reduces KV-cache needs by at least 6x for HBM/DRAM during AI inference, but does not cut persistent SSD storage demand, leaving SanDisk's NAND thesis intact. Shares fell double-digits recently, attributed to headlines, risk-off trading and memory-sector beta, while Street revenue and EPS estimates for the next four years remain unchanged. Bull case: datacenter revenue is only 15% of sales but grew 64% sequentially and 76% YoY last quarter, indicating AI storage upside is still early.

Analysis

Winners are the capital-light storage stack (NAND fabs, SSD assemblers, and controller/IP vendors) because software-led reductions in hot-memory footprint push architects to trade more expensive HBM/DRAM for higher-capacity persistent layers. Expect hyperscalers to reallocate near-term incremental spend away from capacity-constrained HBM pools into commodity NAND and larger form-factor SSDs; that dynamic amplifies pricing power for high-volume NAND players while compressing incremental demand and realized pricing for HBM/DRAM suppliers over the next 2-8 quarters.

A key second-order effect is component mix and fab utilization: lower HBM intensity reduces OEM order visibility for specialized memory, likely forcing DRAM/HBM manufacturers into aggressive pricing or idle capacity decisions — a catalyst for margin compression and capex deferral that can pressure equity multiples. Conversely, SSD controller vendors and software stack providers that optimize IO patterns capture outsized TAM expansion because they enable the NAND-first architecture hyperscalers prefer; watch controller backlog and ASPs for an early signal (6-12 weeks lead).

Tail risks include rapid, broad adoption of these optimizations (weeks→months) that materially reduces DRAM/HBM spend, or a countervailing move where models shift back to larger memory footprints (e.g., dense multimodal training) that re-tightens DRAM/HBM demand within 6-18 months. Near-term catalysts that will re-rate relative winners are: hyperscaler capex guides, server bill-of-materials disclosures (next 1-2 earnings cycles), and NAND inventory/price trajectory; the 1-6 month window will decide whether the headline-induced selloff is a buying opportunity or start of a memory-cycle rotation.

Key Decisions for Investors

• Long SNDK equity (6–12 months): initiate on intraday weakness / knee-jerk flow with a 15–25% position size of thematic allocation. Target +35–50% if datacenter revenue growth sustains; hard stop -18% (technical or on two consecutive q/q decelerations in datacenter revenue).
• Pair trade — Long SNDK / Short MU (equal notional, 3–6 months): play NAND substitution vs DRAM downside. Expect asymmetric payoff if NAND outgrows and DRAM faces price pressure; size as a market-neutral exposure with stop-loss if spread narrows < -10% from entry.
• Options: buy SNDK 9–12 month call spread (moderately OTM, defined-risk): cost-limited bullish play to capture re-rate if datacenter contribution continues >50% y/y in the next two quarters. Max loss = premium, upside target 2.5x–4x premium if newsflow sustains acceleration.
• Event/flow hedge (days–weeks): if headline-driven volatility continues, hedge long exposure with short put protection on SNDK or a small short position in memory-heavy ETFs (e.g., DRAM names) to limit downside from further risk-off moves; remove hedge once hyperscaler guides confirm NAND demand persistence.