Micron Gets the Headlines, But Sandisk Could Offer More Upside

Micron is guiding Q3 FY2026 revenue to $33.5 billion with gross margin around 81%, supported by strong AI-related HBM demand and NAND supply constraints. Sandisk is also highlighted as a stronger near-term AI memory play, guiding Q4 FY2026 revenue to $7.75 billion-$8.25 billion and non-GAAP EPS of $30-$33, while its expected current-year earnings growth of 2096.7% far exceeds Micron’s 626.5%. The piece is largely bullish on AI memory demand and pricing power, but it is framed as comparative analysis rather than new company-specific news.

Analysis

The real signal is not that memory is hot; it is that AI capex is forcing a structural repricing of the memory stack from commodity cyclicality to quasi-infrastructure scarcity. That benefits the names with the tightest allocation discipline and the most contracted exposure, because pricing power now comes less from spot demand and more from multi-quarter visibility into hyperscaler build plans. In that setup, suppliers with cleaner balance sheets and better execution on advanced packaging and HBM mix can sustain elevated margins longer than the market typically allows.

Second-order, the market is likely underestimating how persistent memory tightness can delay competitive normalization. If HBM and NAND constraints hold into mid-2026, customers will prioritize qualification continuity over price, which strengthens incumbent share and makes near-term share shifts sticky. The flip side is that this also incentivizes aggressive capacity additions and process migration, which can set up a harsher unwind in 12-18 months if AI infrastructure spending decelerates or if one large hyperscaler pauses orders.

For MU, the upside is quality of earnings rather than just magnitude: the key question is whether the margin peak becomes a plateau. For SNDK, the market is rewarding operating leverage and contract visibility, but the move has started to price in a very high bar; any sequencing miss or evidence of later-cycle product mix deterioration could trigger an outsized multiple reset. NVDA is indirectly relevant because every incremental GPU deployment raises the memory content per AI server, but the beneficiary list is broader than the compute vendor trade and should include the memory suppliers with the strongest contract structures.

Contrarian read: consensus may be too focused on the scarcity narrative and too light on the duration risk. In memory, peak fundamentals often look most durable right before supply response becomes visible; the first sign will be lead times normalizing, not revenue growth rolling over. That argues for owning the strongest near-term earnings momentum, but only with a plan to fade rallies once the market starts pricing 2027 supply expansion.

Key Decisions for Investors

• Long SNDK on 3-6 month horizon; best risk/reward remains with names where forward guidance is still inflecting upward. Use tight risk controls because the stock is likely discounting a lot of good news, but momentum can persist while contract visibility stays intact.
• Barbell long MU / short a basket of weaker memory peers or generic hardware suppliers over 6-12 months; the thesis is that scarce HBM exposure should command a premium while less differentiated memory names face more cyclicality when supply eventually catches up.
• Add to NVDA only on pullbacks as a second-order beneficiary trade, not as the primary expression; memory content per AI server is rising, but the better alpha is in the constrained suppliers rather than the compute leader everyone already owns.
• Consider selling out-of-the-money calls on SNDK into strength for investors already long; implied upside has likely outrun near-term fundamental surprise capacity, and call overwriting monetizes the elevated optimism while preserving core exposure.
• Watch for evidence of capex acceleration at memory peers over the next 1-2 quarters; if found, rotate from outright longs into pairs or reduce exposure, because the setup could shift from scarcity premium to supply-response risk faster than expected.