Beyond Nvidia: This Under-the-Radar AI Stock Is Up Nearly 300% Over the Past Year

Micron expects the HBM market to expand from $35B in 2025 to $100B by 2028, but management says current capacity will only meet ~50–67% of demand, creating a production bottleneck until additional capacity comes online (targeted in 2027). The company forecasts next-quarter revenue of $33.5B (vs $23.9B this quarter and $13.6B the prior quarter), indicating rapid top-line growth. Google’s TurboQuant could cut memory demand by ~6x for one large-language-model component, but the author argues this will not eliminate broader HBM demand and views the recent pullback as a buying opportunity.

Analysis

Micron sits at the intersection of two asymmetric forces: structural HBM demand growth through 2028 and near-term capacity scarcity that compresses supply elasticity until new fabs ramp in 2027. That scarcity creates pricing power and margin expansion opportunities that can compound FCF materially even if unit growth moderates; treat revenue growth as convex to capacity constraints rather than linear.

Algorithmic efficiency (e.g., TurboQuant) is a genuine shock to marginal memory intensity, but it is a demand reshaper, not a destroyer: efficiency frees compute budget that will be redeployed into larger models, higher batch sizes, or more complex inference topologies, which in many cases increases aggregate memory footprint or shifts demand to different HBM form factors. The key second-order risk is a regime shift to architectures that materially reduce HBM reliance (on-package SRAM, CXL pooling, or near-memory compute) — these transitions are multi-year and capital-intensive, giving incumbent memory vendors a multi-year runway.

Competitor and supply-chain dynamics matter: any incremental HBM capacity outside of Micron (Samsung, SK Hynix) arriving before 2027 will cap pricing; conversely, trade-policy barriers that limit cross-border capacity expansion effectively raise the value of U.S.-based production and create asymmetric upside for Micron relative to non-U.S. peers. Short-term market moves will be sensitive to quarterly guidance and algorithmic announcements from hyperscalers; read those as demand-reshaping catalysts rather than binary hits to the secular thesis.

Key Decisions for Investors

• Core long MU (buy on dips over next 3 months): accumulate to a target position representing 2-3% of portfolio by end-2026. Plan: scale into 3 tranches on 10% pullbacks; target +60% total return by end-2027 while using a protective stop at -25% to limit downside if architecture risk accelerates. Rationale: capture pricing power from HBM tightness pre-2027 ramp; risk/reward ≈ 2.4:1.
• Directional options calendar (MU): buy MU Jan-2028 LEAPS calls and sell shorter-dated Jan-2027 calls (diagonal) to finance premium. Timeframe: 12–18 months. Objective: retain upside to the 2027 capacity rerate while monetizing near-term theta; downside is capped by sold calls, breakeven requires sustained HBM tightness or accelerating revenues — asymmetric payoff if scarcity persists.
• Relative-value pair (long MU / short INTC) equal notional, 12-month horizon: go long MU to capture memory pricing upside and short INTC to hedge general semiconductor cycle exposure and foundry/capex execution risk. Target: +30% relative outperformance; stop/monitor if Intel reports a durable foundry win or if broad AI demand lifts both names — risk of bilateral upside in a runaway AI capex cycle.
• Event hedge / convex kicker (NVDA call): buy NVDA calls (3–6 month tenor) as a convex hedge to capture a broader AI upside scenario that drives both chip and memory demand higher. Use as a small portfolio kicker (0.5–1% notional) — if NVDA gaps higher, take profits and reallocate into MU to capture extended HBM pricing power.