Sandisk Stock Plunged 14% in a Day. Is the AI Memory Boom Cracking?

SanDisk (SNDK) fell about 14% in a single session on July 2 as AI-hardware/memory stocks sold off on fears that AI compute supply is catching up, stoked by reports Meta plans to sell spare AI capacity externally. However, Sandisk’s fundamentals were strong: fiscal Q3 revenue rose to $5.95B (near-doubling QoQ), non-GAAP gross margin jumped to 78.4% from 51.1%, and management pointed to fiscal Q4 revenue of $7.75B–$8.25B plus up to $33 adjusted EPS. The article argues this looks more like a valuation reset than a demand collapse, with ~$42B of minimum contracted revenue providing some downside cushioning.

Analysis

This looks like factor de-grossing, not an inflection in end demand. The first-order loser is the highest-beta memory basket, because when AI-exposure gets questioned, investors sell the most crowded proxies first; MU and WDC should continue to trade as sympathy names even if their own fundamentals lag the move. Second-order, any sign that hyperscalers are normalizing utilization can pressure procurement behavior: OEMs and module buyers tend to slow spot purchases before they cut long-term commitments, so the real earnings risk shows up with a lag, not in the same quarter.

The important distinction is compute vs. storage. Even if GPU scarcity is easing, data creation, checkpointing, and inference logs still increase bits-per-compute, which supports NAND demand longer than the market is implying. That argues SNDK's backlog and margin profile can stay elevated for several quarters; the risk is not near-term revenue collapse but multiple compression once the market starts discounting a slower 2027-2028 growth slope.

Consensus is probably overstating how much one monetization headline means for the whole AI stack. The more plausible bearish path is a rotation out of peak-quality memory names after a huge run, not a collapse in demand. What would falsify the bullish supply thesis: two consecutive months of weaker spot NAND pricing, a downward revision to hyperscaler capex, or SNDK guiding below the current demand run-rate; absent that, the selloff looks more like a valuation reset than a cycle break.

Key Decisions for Investors

• Pair trade: long SNDK / short MU for 1-3 months. Thesis: SNDK has better contracted visibility and data-center mix; MU is the cleaner beta expression of a memory-cycle reset. Risk/reward: target 1.5-2.0x on a modest further de-rating of the complex; stop if memory ASPs re-accelerate or MU issues constructive pricing commentary.
• Do not chase SNDK here; wait for either a further 8-12% drawdown or a reset to a materially lower forward multiple before adding cash equity. The current setup is more suitable for patience than for momentum buying.
• Buy SMH or SOXX 1-3 month put spreads on any rebound in AI hardware names. Use the recent selloff as confirmation that the market is sensitive to utilization news; downside is limited-premium, upside is another factor unwind if hyperscaler commentary softens.
• Set a catalyst alert for the next hyperscaler capex update and next memory pricing data point. If capex stays firm and NAND pricing stabilizes, cover shorts and rotate back into SNDK; if both roll over, the cycle call is becoming real.