Not Nvidia, Not Palantir. This Might Be September's Most Important AI Infrastructure Stock.

Nvidia reported $96.2 billion of revenue, with data-center sales up 117% year over year to $89 billion, while Palantir posted 93% revenue growth to $1.9 billion and raised full-year guidance to more than $8.2 billion. The article argues these results reinforce a tightening AI-memory market that benefits Micron, one of only three scaled HBM suppliers, as DRAM pricing rises and Nvidia expands supply and capacity commitments through fiscal 2032 by $160 billion to $279 billion. Micron is scheduled to report fiscal Q4 2026 results on Sept. 30, with consensus calling for $50.8 billion in revenue and $31.28 EPS; continued HBM and DRAM strength could support an earnings beat and forward P/E multiple expansion.

Analysis

The investable issue is not simply HBM demand but allocation economics: MU’s upside depends on HBM4 qualification yields, mix, and contracted pricing holding ahead of broader DRAM supply additions. SK Hynix remains the incumbent HBM leader and Samsung’s qualification progress is the key competitive swing factor; a successful Samsung ramp would cap MU’s pricing power even if accelerator demand stays strong. Conversely, constrained HBM capacity diverts wafer supply from conventional DRAM, potentially extending tightness across server and mobile memory and improving MU’s blended gross margin over the next 2-4 quarters.

The published consensus figures appear implausible relative to MU’s historical revenue and EPS base, so they should not be used for positioning. The relevant earnings catalyst is Sept. 30 guidance: HBM bit-growth, gross-margin trajectory, customer prepayments/long-term agreements, and capex intensity. A nominal earnings beat without upward pricing or margin commentary is vulnerable to a sell-the-news reaction because the AI-memory thesis is increasingly consensus; a guide implying conventional DRAM normalization in the December or March quarter would likely compress the multiple quickly.

Contrarian view: the market may be underestimating memory’s strategic value but overestimating the durability of spot-price scarcity. Memory producers have historically responded to supernormal returns with capacity investment, and HBM manufacturing complexity delays—but does not eliminate—that cycle. The best 1-3 month setup is therefore a qualified long around guidance rather than blind pre-earnings beta: confirmation of constrained supply into calendar 2027 supports rerating, while evidence of Samsung supply gains or aggressive MU capex argues for reducing exposure. Over 6-18 months, equipment suppliers exposed to HBM capacity expansion may offer cleaner volume upside than memory makers’ inherently cyclical pricing exposure.

Key Decisions for Investors

• Maintain a tactical long MU into Sept. 30 only at reduced event size; add materially only if management guides gross margin higher sequentially and confirms HBM demand visibility beyond the next two quarters. Target a 10-15% post-print upside on a credible rerating; cut if forward gross-margin guidance falls or conventional DRAM pricing is described as normalizing.
• Express relative AI-memory tightness via long MU / short SOXX in a 1-3 month pair, rather than outright semiconductor beta. The thesis is MU-specific mix and pricing leverage; unwind if Samsung announces broad HBM4 qualification or MU raises capex enough to imply materially faster supply growth.
• Monitor SK Hynix and Samsung HBM qualification disclosures as the highest-value falsification signals. Do not buy MU calls until implied volatility versus the expected post-earnings move is checked; elevated pre-event volatility would make a post-results entry or call spread structurally preferable.
• For a 6-18 month capacity-build theme, screen long AMAT, LRCX, and KLAC versus memory producers after validating incremental HBM wafer-fab-equipment orders; this is a watch item, not a recommendation, until capex guidance identifies tool intensity and timing.