If the AI Boom Is So Strong, Why Are Memory Stocks Crashing?

AI infrastructure spending is projected to exceed $700B in 2026, but memory stocks are falling: Micron (MU) is down ~30%, SK hynix (SKHY) is below its July 10 IPO price, and SanDisk (SNDK) is down ~35% after a ~600% earlier-year rally. The article attributes weakness not to collapsing AI demand—HBM remains sold out into future production—but to expectations of easing HBM/DRAM supply and price pressure as manufacturers ramp capacity. It flags a risk/reward shift for investors: pricing may have already peaked, so future earnings could compress even if AI ecosystems continue growing.

Analysis

The market is separating AI infrastructure into two businesses: scarce compute/packaging capacity with durable pricing power, and memory with classic commodity dynamics. That makes NVDA structurally better insulated than MU, SKHY, and SNDK because GPU demand can stay strong even if memory ASPs start to normalize; the risk is that incremental AI capex migrates away from the tightest bottleneck and toward power, networking, and buildout spend, leaving memory as a lower-share wallet item.

Near term, the key catalyst is not AI demand but whether HBM capacity additions and yield improvements translate into visible pricing pressure over the next 1-3 quarters. If inventories rise or contract pricing softens, the rerating in memory can continue even with record shipments. Over 6-18 months, the more important risk is margin mean reversion: a business that was priced for scarcity can de-rate fast once the market believes supply is catching up.

Contrarian view: the selloff may be front-running an inflection that is still too early to confirm. If HBM remains effectively sold out into the next budget cycle, the current weakness becomes a positioning/liquidity event rather than a fundamentals call, and MU in particular could bounce hard on any sign that pricing is stable. Falsifier for the bearish memory view is simple: no sequential ASP deterioration, no inventory build, and continued upward revisions to HBM mix or gross margin guidance.

The second-order winner is NVDA: even if memory prices compress, its relative scarcity and ecosystem lock-in should keep multiple support better than the supply-chain names. The losers are the more levered memory equities where expectations already embed peak profitability; those names can underperform a healthy AI tape if investors rotate toward less cyclical AI exposure.

Key Decisions for Investors

• Long NVDA / short MU as a 1-3 month relative-value pair: expresses AI exposure without betting on memory ASPs; stop if MU guides to stable-to-rising sequential pricing or inventory stays exceptionally tight.
• Avoid chasing SNDK into strength; consider shorting or buying puts on SNDK over the next earnings window if the stock keeps discounting peak-cycle margins. Thesis breaks if NAND/HBM pricing remains firm and management re-accelerates gross margin.
• Set a tactical buy alert on MU only after the next read on HBM pricing and inventory: if sell-through remains tight and there is no evidence of channel build, the current drawdown may offer 2-3x upside versus limited downside on a confirmed trough setup.
• Use SOXX/SMH weakness to add exposure selectively to compute leaders, but stay underweight memory until pricing data turns; the best risk/reward is owning bottleneck beneficiaries, not commodity up-cycle names.
• Watch for a 1-2 quarter lag in capex mix: if hyperscaler spend keeps shifting toward power, cooling, and networking, that is a structural headwind for memory multiples and a tailwind for non-memory AI infrastructure.