Micron's AI Memory Stampede Could Last Years, Analysts Say

Two major analysts raised Micron price targets sharply (TD Cowen to $500 from $450; RBC to $525 from $425) and materially lifted EPS estimates, driving shares up ~5.5% to $449.43 near a 52-week high. TD Cowen models $10.40 EPS for the Feb quarter (Street $8.82) and $13.50 guide for May (Street $10.92), with long-term models of $65 EPS for 2026 and $90 for 2027 (book value $120 end-2026, $215 end-2027); RBC forecasts $13.52 EPS on $25.5B revenue for Feb and $17.19 EPS on $31.3B for May, and full-year EPS of $54.30 (2026) and $75.44 (2027) with gross margins above 75%. Analysts cite AI/data-center-driven DRAM and HBM demand and long-term agreements as re-rating catalysts but flag inventory buildup risk that could trigger a correction.

Analysis

The bullish thesis rests less on a single earnings beat than on establishing durable pricing and LTAs that convert cyclically volatile margin streams into predictable cash flows — that’s the only mechanism that justifies a multi-turn re‑rating from a memory multiple to something nearer HDD-style multiples. If LTAs begin to appear in customer disclosures or procurement schedules over the next 3–9 months, expect a step-function compression in implied volatility and a rapid multiple expansion; conversely, any language that points to inventory digestion will produce outsized negative gamma given current positioning.

Key catalysts are time-staggered: days — the upcoming print and management commentary will set immediate direction; months — customer inventory builds/shipments and early LTA evidence will determine whether pricing is durable; years — capacity decisions and LTA renewals will lock-in a re-rating. Tail risks concentrate around demand concentration (AI capex swings), order pull-forwards that reverse, and an HBM pricing trajectory that could disappoint as HBM3E/HBM4 transitions normalize ASPs.

Second-order winners include memory-equipment suppliers and OSAT/test players who get clearer multi-year capex visibility (AMAT, LRCX, KLA, ASE-style contractors), and hyperscale OEMs that secure fixed-price supply for optimized TCO on AI systems. On the flip side, NAND- and HDD-centric OEMs (and any vertically exposed consumer NAND suppliers) are at risk of relative underperformance if DRAM/HBM outpaces NAND, and any broad-based cloud capex slowdown would compress all cohort valuations.

Contrarian alert: the consensus assumes a steep, sustained pricing trajectory into 2027; that is a binary bet on supply discipline + LTA adoption. Positioning is crowded; absent clear LTA rollouts, the path to >$100 EPS is heavily mark-to-market sensitive. Hedge entry exposure where possible and treat any post-print pop as an opportunity to reduce gamma risk rather than add naked exposure.

Key Decisions for Investors

• Tactical earnings trade (30–90 days): Buy a 3–6 month MU call spread (e.g., near-term $450/$650 structure) sized to 0.5–1.0% of portfolio notional. Rationale: captures upside from beat + positive guide while capping premium outlay; target 2.5–3.0x return if market re-rates, stop-loss on premium at 50%.
• Core re-rating play (12–24 months): Allocate 1–2% portfolio to MU LEAPs (calendar 2028 calls at strikes around current implied re-rating level) or buy shares and hedge with a 12‑month put. Rationale: asymmetry to multi-year LTA realization — target +40–80% if LTAs establish margin floors; max loss limited to premium/cost of hedge.
• Pair/relative-value (12–18 months): Long MU (1% weight) vs short WDC or a NAND-heavy name (1% weight) to isolate DRAM/HBM outperformance versus NAND/HDD exposure. Rationale: isolates earnings multiple expansion tied to DRAM tightness; target spread convergence of 30–40%, risk is broad memory cycle reversal.
• Risk control: Given crowding and inventory risk, buy protective puts (3–6 month) equal to ~25–40% of directional exposure or size initial entries small and scale up only after post-earnings volatility compresses. Rationale: prevents a 2021–22 style correction from inflicting disproportionate portfolio drawdown.