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

Micron projects next-quarter revenue of $33.5B (vs $23.9B this quarter and $13.6B the prior quarter) and its shares have risen nearly 300% over the past year. Management says current capacity can meet only ~50–66% of high-bandwidth memory (HBM) demand, with capacity expansions not expected to materially ease the bottleneck until 2027, while HBM TAM is forecast to grow from $35B in 2025 to $100B by 2028. Google’s TurboQuant cuts memory demand sixfold for part of LLM workflows—a partial headwind—but the article argues it will mainly shift memory usage rather than eliminate long-term demand, supporting a bullish investment thesis on Micron as an AI memory play.

Analysis

The HBM market is governed more by wafer/assembly lead times and packaging throughput than by chip-design cycles; that asymmetry creates a multi-quarter to multi-year window where suppliers with available capacity can extract pricing and share benefits independent of unit demand volatility. Hyperscalers buy lumpy, design-locked capacity and will front-load procurement when they fear bottlenecks — expect backlog-driven revenue recognition spikes for suppliers with confirmed slot capacity and steep inventory destocking/re-stocking dynamics at cloud customers on 2-6 quarter cadence. Algorithmic efficiency improvements change the substrate of demand: they compress memory-per-inference but often enable larger models or wider deployment, so net memory demand can reallocate across HBM generations, on-package SRAM, and system-level DRAM rather than disappear.

Second-order winners include advanced packaging and OSAT suppliers (leadframe, TSV, interposer), cloud integrators who secure guaranteed HBM supply via committed buys, and GPU/server OEMs able to re-price systems when memory is the constraint; losers are commodity DRAM suppliers with excess generic fab capacity and any ODM that cannot lock supply. Geopolitical export controls and local-content requirements will create bifurcated pricing (north-america/europe vs. APAC), opening an arbitrage for vendors able to certify non-China supply chains — this is as much a policy trade as a tech one. Key near-term catalysts: announced capacity slots with customer names, hyperscaler purchase commitments, and competitor fab-start delays; key risks are faster-than-expected algorithmic compression, surprise competitor HBM ramps, or a rapid demand pullback from a single large cloud buyer.

Key Decisions for Investors

• Core asymmetric: Buy MU exposure via Jan-2027 LEAP calls (size ~3% portfolio notional) funded with sale of 6–9 month covered calls (reduce carry). Hedge ~25% notional of the LEAP position with 6–9 month puts to cap drawdown. Timeframe 12–24 months; target ~2.5x payoff if structural HBM tightness persists, stop-loss: cut to half notional on any publicized multi-year capacity oversupply.
• Tactical GPU lever: Buy NVDA 6–12 month call spread (delta-focused; modest premium) sized 1–2% notional to capture upside if HBM bottlenecks force higher GPU ASPs and server refreshes. Risk: premium decay; Reward: asymmetric participation in system-level pricing shock; stop at 40% premium loss.
• Relative-value pair for supply bifurcation: Long MU equity (or LEAP) / short SMH-sized put position to hedge a China-policy selloff — size pair so net exposure to memory-price upside remains ~2% portfolio. This isolates HBM-specific upside while protecting against semiconductor cyclicality triggered by macro/policy.
• Crash protection: Buy 3–6 month SMH or SOXX puts (small allocation) as insurance against sudden hyperscaler demand collapse or rapid algorithmic efficiency wins. This is low-cost tail insurance to preserve gains across the portfolio during headline-driven reversals.