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

Micron stock has risen ~300% over the past year but recently pulled back; management says current capacity can meet only 50–66% of HBM demand while forecasting the HBM market to grow from $35B in 2025 to $100B by 2028. Micron guides ~$33.5B revenue next quarter versus $23.9B this quarter and $13.6B the prior quarter, signaling rapid sequential growth; however, capacity expansion not expected to alleviate the bottleneck until 2027. Google’s TurboQuant could cut memory needs six-fold for part of LLMs, tempering near-term demand growth but the article argues long-term memory demand and Micron’s upside remain intact.

Analysis

Memory scarcity is not a binary shortage — it’s an intensity-of-use problem with steep price elasticity, and that distinction creates asymmetric outcomes across the stack. When hyperscalers face constrained HBM, they shift procurement behavior (longer lead purchases, consignment, design wins prioritized) which amplifies winners with standing supply agreements and execution capability while punishing firms that promise volume but can't deliver. Expect multi-quarter inventory cycles: vendors with limited fab headroom can command sticky higher realized prices early, then face margin compression if supply catches up and customers de-risk by standardizing on cheaper or more memory-efficient architectures.

Algorithmic compression (like quantization/compilation advances) alters the slope of demand rather than the intercept — it can reduce memory per model instance but also lowers the marginal cost of running more models, spawning additional use cases and netting out some of the demand decline. The immediate second-order impact is on product mix: demand will tilt from absolute HBM bytes to higher-performance, lower-latency HBM flavors and co-packaged/near-die memory solutions; that favors firms with advanced packaging partnerships and end-to-end supply agreements.

Geopolitics and capex timing are the wildcard. Export controls or faster-than-expected capex from competitors can flip pricing power inside 6–18 months. From a risk calendar perspective, expect outsized volatility around quarterly bookings and any capital-spend updates tied to 2026–2028 capacity ramps — these are the real de-risking events for valuation.

Key Decisions for Investors

• Core long (MU): Buy a 9–15 month MU call spread sized to 1–2% NAV (e.g., debit call spread to limit premium). Thesis: capture upside from pricing power and execution without paying full premium. Risk: max loss = premium; target 40–80% return if ASPs remain elevated through the next two inventory cycles.
• Tail hedge (MU protection): If owning MU stock, buy 3–6 month 15–25% OTM puts sized to 50% of equity exposure to protect vs rapid adoption of memory-saving algorithms or sudden demand destocking. Cost ~1–3% NAV; this limits drawdown from algorithmic or order-cancellation shocks.
• Pair trade (execution vs expectation): Long MU equity or calls + short SK hynix (or another DRAM peer) equity sized 1% NAV each to isolate execution risk. Rationale: capture spread if Micron’s standing contracts and fab roadmap outperform rivals; unwind at 30–50% realized spread capture or on confirmed capacity announcements.
• Hedge the stack (NVDA): Buy 6–12 month NVDA calls (smaller size, 0.5–1% NAV) as a hedge to AI compute upside that is less HBM-sensitive (e.g., architectures favoring more chips vs more memory). If compute demand expands while HBM tightness normalizes, NVDA should appreciate; risk = option premium, reward asymmetric if AI tailwinds persist.