The Memory Shortage Is Not Over: More Upside for Micron and Sandisk?

DRAM and NAND prices have roughly tripled in 2026 as AI-driven HBM demand collides with constrained capacity, with meaningful new memory-fab output not expected until mid-to-late 2027 or later. Micron reported fiscal Q3 revenue of $41.5B versus $9.3B a year earlier, 85% gross margin and $25.11 adjusted EPS, while guiding fiscal Q4 to about $50B revenue, 86% margin and roughly $31 EPS. Sandisk posted fiscal Q4 revenue of $8.97B, up 51% sequentially, and guided Q1 revenue to $10.3B-$10.8B with $44-$46 adjusted EPS; the article favors Micron as the core DRAM exposure and views Sandisk as a higher-beta NAND trade, while noting NAND could loosen in H2 2027.

Analysis

The investable question is not headline memory pricing but whether suppliers can convert scarcity into contracted, durable gross margin without restarting the capex cycle. MU is structurally preferable to SNDK: HBM allocation raises switching costs and supports customer qualification moats, while NAND remains more exposed to commoditization and a faster inventory correction. A tight memory market can also become a shipment constraint for NVDA and custom-ASIC platforms; it preserves accelerator pricing power but may defer system revenue where memory availability, rather than compute demand, is the bottleneck.

The near-term catalyst is MU's September 30 earnings, but the article's unusually large financial figures and contract assertions require independent verification against filings, management commentary, TrendForce contract data, and hyperscaler capex disclosures. Consensus may be underestimating earnings persistence through 2027, yet the stocks can still de-rate before fundamentals turn because peak-cycle EPS is not a reliable valuation anchor. The decisive leading indicator is the HBM/commodity DRAM spot-to-contract spread: narrowing alongside rising distributor inventories would signal that buyers have shifted from securing supply to managing working capital.

Second-order beneficiaries are memory-test and packaging suppliers such as TER and ACLS only if HBM mix drives utilization rather than merely price; equipment names LRCX and AMAT are a later-2027 capacity-addition trade, not a clean near-term shortage proxy. The contrarian risk is that hyperscalers respond to memory inflation by optimizing model architectures, extending server lives, or redirecting AI budgets toward power and networking, reducing memory content growth before new fabs arrive. A meaningful reduction in AI capex guidance from AMZN or GOOG would matter more than another quarter of supplier upside.

Key Decisions for Investors

• Establish a 3-6 month long MU / short SNDK pair, sized 1:0.7 beta-adjusted, ahead of MU earnings: this expresses durable DRAM/HBM scarcity while hedging broad memory-price beta. Reassess if MU does not extend qualified HBM supply visibility into calendar 2027 or if NAND contract pricing reaccelerates relative to DRAM.
• For directional exposure, buy MU only on a post-results pullback of 8-12% if guidance and verified contract-pricing data support stable or higher next-quarter margins; target 20-30% upside over 6-12 months, with a stop/reduction on a two-month decline in DRAM contract prices or a material inventory build.
• Avoid adding SNDK outright at current momentum until enterprise-SSD volume, not pricing alone, is independently confirmed. A tactical long becomes attractive only after a 15%+ reset with forward NAND pricing still firm; cap position size below MU because NAND downside is sharper once buyer inventories normalize.
• Set alerts on AMZN and GOOG capex guidance, Korean memory spot prices, and semiconductor distributor inventory data. Any combination of AI-capex cuts, falling spot premiums, and inventory accumulation should trigger de-risking of both memory longs before late-2027 supply additions become the consensus concern.