Summary

The memory supercycle is real, broad, and simultaneous: every layer of the storage stack — HBM/DRAM, NAND, HDD nearline — is margin-inflecting at once, driven by AI-datacenter demand and a historic pricing shortage (DRAM ASPs rose ~60% in a single quarter). But it is a cycle, not a plateau. The cleanest, most-trustworthy value sits in the un-hyped HDD names; the hype is already priced into HBM; and the $27B+ of capex funding the boom is building the oversupply that eventually ends it. Own the up-cycle with eyes open, and size for the peak.

The broadest coordinated inflection in memory on record

Strip away the individual names and look at the sector as a system. Every layer of the storage hierarchy is moving in the same direction at the same time — and doing it hard.

Micron's Q3 FY2026 (ended May 2026) should settle any debate about whether AI is a real demand signal or a narrative. Revenue came in at $41.5B — up 346% year-over-year, 74% quarter-over-quarter, a record. DRAM alone was $31.3B, 76% of the total, with ASPs up roughly 60% in a single quarter on low-single-digit bit growth. That last detail matters: the revenue isn't coming from shipping more chips, it's coming from pricing power — the most durable form of revenue there is. Data-center revenue exceeded $25B in the quarter (annualizing above $100B), and HBM3E and HBM4 are fully booked through calendar 2027 with demand extending into 2028.

The NAND side is even more structurally telling. SanDisk went from a −111% operating margin in March 2025 — a $1.9B operating loss at the NAND trough — to +69% operating margin a year later, on revenue that went from $1.7B to $6.0B. That is one of the most violent loss-to-profit reversals in recent semiconductor history. The NAND cycle doesn't turn gently; it snaps.

Then there are the names nobody's writing about.

The clean story the hype machine is ignoring: HDDs

The entire narrative has been pointed at HBM — Micron, SK Hynix, AI-server adjacency. Fair. But the most credible, least-crowded inflection in this complex is in the unglamorous nearline hard-drive business.

Seagate's margin expansion — 23% to 32% across four consecutive quarters — is the opposite of a pricing spike. It's slow, steady, compounding. Western Digital shows the same shape (10% to 36%), powered by the same thesis: AI infrastructure doesn't just need compute, it needs somewhere to store the data the compute generates. Inference clusters, model checkpointing, training datasets, regulatory retention — all of it lands on nearline HDD in the data center, because HDD cost-per-terabyte is still an order of magnitude below flash. Nobody writes the "AI needs hard drives" headline. The margin progression is writing it anyway.

That is the tell the desk keeps returning to: the quietest inflection in the complex is also the most believable, precisely because it lacks the pricing-spike discontinuity that makes the DRAM numbers look almost too good.

SK Hynix: the best-in-class name US money couldn't properly own — until now

SK Hynix holds 56.4% HBM market share — the #1 position, and not close. Its Q1 2026 revenue was roughly $35.5B (₩52.6T), up 198% year-over-year, on a ~72% operating margin, with HBM booked about three years out. By every operational metric it is the dominant player in the highest-growth, most-constrained subsegment of memory.

And yet US investors have largely been forced to own Micron as the proxy. That changes around July 10, 2026, when SK Hynix prices its Nasdaq ADR listing — projected to be the largest-ever Korean US listing at ~$29.4B, with proceeds funding new Korean fabs and additional ASML EUV tooling. When the actual market leader becomes directly accessible in a US brokerage account, the proxy trade gets arbitraged. Part of the gap between Micron's premium multiple and SK Hynix's cheaper relative valuation is simple access friction; the ADR removes it. If you're going to own HBM, the case is to own the best-in-class at the cheaper multiple before the liquidity constraint disappears.

The honest risk: you are buying cycle-peak margins

Everything above is true. The following is also true, and the comfortable narrative papers over it.

Micron's non-GAAP EPS of $25.11 in a single quarter — against $1.91 a year ago — implies the stock trades at roughly 9–10x annualized run-rate earnings. Cheap on P/E. Except that's cheap on peak earnings, which is the classic cyclical value trap. SanDisk's 69% operating margin is not a run-rate for a commodity NAND producer; SK Hynix's 72% margin on ASPs that roughly doubled is a number the sector prints at the top of oversupply-to-undersupply swings. When pricing momentum rolls, it rolls fast.

The deeper risk is structural. Micron raised FY2026 capex to ~$27B; SK Hynix is raising $29.4B partly to fund new fab capacity. The investment required to meet 2027–2028 demand is the same investment that creates 2029–2030 oversupply. Memory has run this cycle repeatedly, and nothing about AI demand breaks it — AI pulls the cycle forward and raises its amplitude; it doesn't repeal the supply response. The capex clock is running.

The tell to watch is DRAM ASP momentum. A +60% QoQ ASP move is the signal; when that number turns negative — even modestly — the margin story changes faster than the revenue story, because the revenue line still has volume growth carrying it. Watch ASPs, not revenue.

The durable read

The supercycle is real, broad, and still in its middle innings on the demand side. The move isn't to short it; it's to own it with eyes open and sizing calibrated for a cycle.

• Own the best-in-class or the cleanest inflection. SK Hynix is best-in-class, and the ADR is the entry catalyst that closes the proxy gap. STX and WDC are the cleanest inflections — steady margin compounding in a segment nobody hypes, on a demand driver (AI data storage) that is real and durable. SanDisk's swing is spectacular but represents a NAND cycle peak, not a new baseline.
• Size for the peak. The P/E screens look cheap because earnings are at highs — not the same as cheap on mid-cycle earnings, which is the number that matters when the cycle mean-reverts. If a 20–30 point margin reversion would be ruinous to the position, the position is too large.
• Watch the capex-glut clock. The next oversupply is being funded right now by the same earnings this cycle is generating. ASP momentum is the leading indicator; when it turns, reduce first and ask questions later.

Desk Call

Sources

• Micron Q3 FY2026: SEC 10-Q (mu-20260528, filed 2026-06-25) + Micron's Q3 FY2026 earnings release/call (revenue $41.5B, +346% YoY; DRAM $31.3B; non-GAAP EPS $25.11; data-center >$25B; FY2026 capex ~$27B). FY2025 baseline from the 10-K (mu-20250828): revenue $37.38B, operating income $9.77B.
• SK Hynix Q1 2026: SK hynix official 1Q26 results (₩52.6T, +198% YoY, ~72% operating margin, 56.4% HBM share); $29.4B Nasdaq ADR per the workspace world-event 2026-06-24-sk-hynix-files-29-4b-nasdaq-adr-listing.
• SanDisk / Seagate / Western Digital: quarterly income statements via provider fundamentals; operating-margin progressions computed from reported revenue and operating income.