Sandisk's $94 Billion Bet Against The NAND Cycle

Q4 revenue surged 51% sequentially to $8.97B, while gross margin expanded to 84.6%. Datacenter revenue jumped 103% sequentially to about $3B, tied to AI inference-driven NAND demand. Visibility also improved: eight NBM agreements lock in $93.9B minimum expected revenue, covering over half of fiscal 2027 bit supply.

Analysis

The market’s bigger mistake is treating this as a one-off revenue beat instead of a shift in NAND’s end-market mix. AI inference is a better demand engine than training for storage vendors because it is persistent, distributed, and read/write heavy; that should raise enterprise SSD attach rates and reduce the depth of the next NAND downcycle. The clearest beneficiaries are MU and, secondarily, WDC on mix; the less obvious spillover is to AMAT/LRCX/KLAC if customers move from “proof of concept” to capacity expansion.

The contract visibility is bullish, but it is also a reminder that revenue can be pre-sold before economics are fully locked in. If these commitments are mainly volume guarantees, they may cap spot upside while improving utilization and smoothing earnings; that is good for valuation stability, not necessarily for peak-cycle margin assumptions. The real signal to watch over the next 1-3 months is whether enterprise SSD lead times and pricing firm in channel checks, not the headline agreement value.

Contrarian risk: consensus may be overestimating how fast inference translates into incremental NAND bits. Hyperscalers can defer storage upgrades, multi-source aggressively, and substitute software optimization for hardware spend. If NAND ASPs soften again or inventory days stop falling into the next quarter, the rerating should fade; structurally, the better tell will be FY27 guidance revisions and whether datacenter mix keeps outgrowing total bit supply into calendar 2026.

Key Decisions for Investors

• Add to MU on any post-event weakness; best risk/reward is a 1-3 month trade into the next guide cycle, with upside if NAND ASPs and enterprise SSD commentary tighten.
• Pair trade: long MU / short WDC over 3-6 months. The thesis is that pure-play NAND exposure should capture the mix shift more efficiently, while WDC remains more exposed to lower-multiple HDD cash flows if the market rotation stalls.
• Initiate a small long in AMAT or LRCX as a second-order beneficiary, but only on evidence that memory capex is re-accelerating; otherwise treat this as an alert, not a core position.
• Buy SOXX on a pullback if channel checks confirm datacenter SSD demand is improving; if NAND spot pricing rolls over or inventory builds, exit quickly.
• Set a falsifier: if next-quarter enterprise/datacenter commentary fails to show follow-through in NAND pricing or bit demand, reduce exposure to MU and the broader memory basket.