AI Memory Stocks Micron and Sandisk Are Up 200% in the Last 3 Months. History Says This Will Happen Next.

Micron (MU) shares are up 203% over three months and Sandisk (SNDK) up 217% on a severe NAND/DRAM supply shortage tied to AI infrastructure demand, with NAND and DRAM prices up ~200% and ~300% YoY. Despite recent strength, Wall Street is flagged as concerned that AI-driven memory demand may peak in 2028, with projected adjusted earnings declines of 27% (Micron fiscal 2029) and 54% (Sandisk fiscal 2029) and historical cycle risk implying a potential 50%+ drawdown. Near-term guidance remains aggressive (Micron sales +340% and adjusted net income +900% YoY in the current quarter; Sandisk sales +320% and adjusted net income +10,700% YoY), but the article frames forward risk as the supply/demand cycle turns.

Analysis

The key market mechanism is not "AI demand" itself but margin convexity: once a commodity memory market is tight, incremental price gains flow almost entirely to EBITDA because fab utilization, depreciation, and inventory gains are already embedded. That makes both names behave like leveraged exposure to supply discipline, but the risk asymmetry is very different: MU has a diversified DRAM/HBM mix and can likely defend returns longer, while SNDK is more exposed to a NAND reversion where pricing power disappears faster and valuation can compress before earnings fully roll over.

The near-term setup is still constructive as long as hyperscaler capex remains front-loaded and manufacturers keep output growth below demand growth. The first reversal signal will be not a headline about AI slowing, but a turn in spot contract pricing, longer inventory days, or capex announcements that imply fabs are being rebuilt faster than end-market consumption. Over a 1-3 month horizon, any pause in pricing could hit the higher-multiple name hardest; over 6-18 months, history suggests the market will re-rate these equities before reported earnings peak, not after.

The contrarian point is that investors may be treating all memory as one trade when the cycle is already bifurcating into differentiated DRAM/HBM versus more fungible NAND. If that split persists, the "memory bubble" narrative will be wrong on MU for longer than expected and right on SNDK sooner. The real falsifier for a bearish call is continued upward revision to memory ASPs alongside disciplined industry capex; if those two stay aligned, the current move can extend well beyond the standard cyclical playbook.

Key Decisions for Investors

• Relative-value: long MU / short SNDK on any strength, 3-6 month horizon. Thesis is MU's mix and scale should preserve earnings power better than SNDK's more commodity-like exposure; risk/reward is better than a naked short because it isolates the valuation and mix gap.
• If initiating a bearish expression, prefer SNDK put spreads 6-9 months out rather than outright short stock. The stock is already discounting a lot of growth, so the cleaner catalyst is multiple compression once NAND pricing rolls over; use the trade only if spot pricing stops accelerating.
• Add MU only on pullbacks, not after parabolic momentum. The better entry is after the first guidance miss or commentary that inventory normalization is underway; until then, upside is mostly multiple-supported and vulnerable to a sharp de-rating if the cycle turns.
• Set a watch item on industry memory ASPs and vendor capex commentary. If DRAM or NAND pricing flattens for two consecutive months, that is the first actionable signal to reduce long exposure and increase hedges.
• Avoid chasing NVDA downside off this story alone. Memory tightness is a cost headwind, but it is more likely to compress OEM margins than to impair near-term GPU demand; the cleaner trade is in the memory suppliers, not the compute leader.