Micron stock continues slide, despite massive data center demand

Micron shares have fallen more than 27% since its Mar. 18 blowout Q2 report (EPS $12.20, revenue $23.86B; EPS +682% YoY, revenue +196%) despite guidance for Q3 revenue above consensus. Morgan Stanley's Joseph Moore says there is no evidence demand for memory/storage is declining and argues memory is the best way to play growth from agentic AI, favoring memory stocks over AMD/Intel. The memory shortage is driving higher prices and margin pressure across device makers; Micron, SK Hynix and Samsung remain up 263%, 323% and 189% over the past 12 months respectively.

Analysis

Memory demand is being pulled in two opposing directions: algorithmic compression reduces bytes/inference while agentic, multi-step workloads expand the effective working set per task and increase concurrency. The net effect is not linear — modest per-inference efficiency gains can be overwhelmed by usage growth if agentic workflows raise average context lengths by 2x–5x or increase session counts across latency-sensitive services. This creates a structural regime where utilization per server rises even if per-inference footprint falls, favoring vendors able to supply dense, low-latency DRAM/HBM at scale.

On the supply side, DRAM/HBM capacity is a slow-moving variable (multi-quarter to multi-year lead times), so short-term elasticity is limited and inventories will remain a key price lever. OEMs that cannot pass through rising memory bills face margin pressure, while cloud providers with pricing power or ability to redesign software stacks capture most surplus. Equipment and specialty material suppliers to memory fabs are second-order beneficiaries; conversely, smartphone and PC OEMs are the most exposed to margin squeeze if procurement windows tighten.

Key catalysts to watch: breadth and speed of TurboQuant-like adoption across major model frameworks (3–12 months), reported HBM spot prices and supplier inventory days (weekly/monthly checks), and capex signals from the three dominant memory suppliers (timelines measured in quarters). Tail risks include a breakthrough on persistent-memory or on-chip compression that meaningfully reduces external memory needs over 12–36 months, which would flip the current leverage profile for pure-play memory names.

Key Decisions for Investors

• Long MU via 3–6 month call spread (buy MU 6-month ITM call, sell MU 6-month OTM call) sized 2–4% portfolio: captures upside from tight memory on positive capacity/capex cadence while limiting premium decay; target 40–60% gross upside, max loss = premium paid.
• Pair trade: long MU (equal-weight) / short AMD or INTC (half-weight) for 3–9 months to isolate DRAM/HBM exposure vs general CPU/accelerator hopes—risk: AMD/INTC rally on AI CPU/accelerator share gains; set stop-loss if pair diverges >20%.
• Event play: buy GOOG/GOOGL calls (3–9 month) as asymmetric hedge to algorithm diffusion — Google’s deeper ecosystem adoption of TurboQuant-style tech accelerates platform monetization and could offset memory-driven downstream weakness; size as 1–2% portfolio optionality.
• Risk management: monitor HBM spot price and supplier inventory days weekly; trim memory exposure by 25% if HBM spot falls >20% or suppliers announce aggressive capacity additions with >30% utilization guidance within 6–12 months.