After a Monster Run, Is Micron Still Worth Owning?

Micron shares have surged 324% over the past year as AI-driven demand for DRAM, NAND and HBM gives the company pricing power, strong revenue growth and expanding gross margins. However, Alphabet's TurboQuant lossless compression has sparked recent selling pressure, introducing a plausible downside scenario where software efficiencies reduce raw memory/storage needs. The author still views Micron as a high-conviction but volatile AI-infrastructure exposure, cautioning that further gains will be harder won and investors must tolerate significant volatility.

Analysis

Micron’s run has shifted the debate from “will memory rebound?” to “who captures the structural share and pricing premium as AI drives nuanced layer demand.” The industry’s two levers are capacity and architecture: suppliers that can deliver HBM at scale and with predictable yields command pricing power, but HBM capacity build-outs are multi-quarter projects that create a sustained window for margin capture — until they are not. Inventory and contract cadence at hyperscalers will therefore determine whether recent gains are durable or a multi-quarter re-pricing ahead of new supply coming online.

Google’s compression milestone is a genuine asymmetric risk because it substitutes silicon with software at the margin; adoption need not be universal to materially dent demand for bulk NAND/DRAM. Expect a bifurcation: latency-sensitive HBM demand (for training/inference) will remain elevated while cold/capacity NAND demand is the most vulnerable to compression and improved data-engineering practices. That divergence increases dispersion across memory sub-segments and makes single-name risk in MU larger than headline growth rates imply.

Timing and triggers matter: market moves over days will be driven by episodic headlines (compression adoption, hyperscaler order beats/misses, quarterly inventory revisions), while structural reversals play out over 6–24 months as capex and fab lead times resolve. The clean contrarian: the stock already prices in sustained share gains and elevated pricing; the next leg requires execution on HBM volume and margin resilience, not just continued AI spend. Position sizing and hedging, not pure conviction, are the dominant portfolio decisions here.

Key Decisions for Investors

• Tactical long MU with capped downside: size 2–4% portfolio; enter on a 15–25% pullback from recent highs or post any sign of hyperscaler re-stocking. Hedge with 3–6 month puts ~15–20% OTM (~cost expected to be 3–6% of notional). Risk/Reward: asymmetric upside if HBM share gains continue (target 40–80%+ 12-month upside) vs limited, pre-funded downside from the put premium.
• Pair to express software-substitution risk: short MU equal notional vs long GOOGL (or buy a GOOGL 6–12 month call spread) size 1–2% portfolio, horizon 3–9 months. Rationale: if TurboQuant-like adoption accelerates, MU underperforms while GOOGL benefits from lower infra intensity; if AI capex remains unabated, pair reduces market beta. Risk/Reward: payoff concentrated around compression-adoption inflection; max loss limited by position sizing and defined in the call-spread leg.
• Volatility/range trade on headline sensitivity: sell near-term (30–45 day) MU calls funded by buying 3–6 month MU calls to create a calendar/diagonal spread; execute when IV term-structure steepens post-big run. Reward: collect short-dated premium if momentum stalls after headlines; Risk: capped by longer-dated long calls, but requires monitoring around earnings and hyperscaler guidance windows.
• Convex exposure to infrastructure winners: buy 9–18 month NVDA and AVGO call spreads (or allocate 1–3% each) to capture continued data-center interconnect and accelerator demand even if memory growth gets lumpy. R/R: these names have more direct leverage to compute layer spending; downside risk is high absolute multiples, so use defined-cost call spreads to limit capital at risk.