Micron's Record Margin Is the Tell, Not the Proof

Micron is heading into its September 30 earnings report having already told the world the numbers are extraordinary. Its fiscal fourth quarter is guided to roughly $50 billion in revenue at a gross margin near 86% — a profit rate no memory maker has ever printed, on a quarter bigger than any memory maker's entire previous year. And the stock, up something like 900% over the trailing year at its June peak, sits about a fifth below that high. That is the whole puzzle in one screen: record margin, falling stock. The market is not confused. It is doing what markets do with cyclicals — pricing the peak before the print.

The record margin is the tell, not the proof

Memory is the archetypal cyclical business. Three companies — Samsung, SK Hynix, and Micron — control the bulk of DRAM production of a product that trades, to a large degree, like a commodity. So the pattern has run on loop for three decades: prices boom, suppliers over-build, supply floods in, prices crash, margins fall out, everyone bleeds, and it starts over.

The last complete cycle is the clean reminder. In the 2016–2019 run, Micron's gross margin peaked at just under 59% in 2018, and the stock topped out roughly two quarters before the earnings peak — then fell more than half into late 2018 as margins collapsed toward 27%. That precedent is the conditioned reflex doing the selling today. Investors who lived through it have trained their hands to exit before the crash, and a gross margin of 84.9% in fiscal Q3, expanding to ~86% in Q4, looks like the kind of number that historically sits at the top of the curve.

That is why the pullback makes sense mechanically even as everything reported looks spectacular. The stock's late-June record — after Q3 revenue of $41.5 billion, up over 300% year over year — was met within weeks by a 13% single-day drop in July and a drawdown that briefly touched 39% off the highs by the end of that month, with the whole memory complex knocked 20% or more below its recent peaks. The market was not disputing the quarter. It was asking the only question that matters in a cyclical: not whether this print is strong, but where we are in the cycle.

The contracts are what this cycle does differently

The bulls' answer — and it is a data answer, not a vibes answer — is that the mechanisms that used to guarantee the bust have been partly rewired.

Start with the physical reality: high-bandwidth memory is voracious. Each HBM module eats three to four times the DRAM wafer capacity of a standard chip, so as AI demand scales, it compresses the wafer supply available for everything else. That makes HBM a natural governor on total memory supply, not an add-on to it. Micron says its HBM3E and HBM4 capacity is fully booked through calendar 2027, with demand reaching into 2028, and that it has no line of sight to when supply catches up with demand — tightness it expects to persist beyond 2027.

Then there is what is genuinely new: the multi-year take-or-pay contract. Micron has announced 16 strategic customer agreements covering committed minimum volumes through 2030, with a cumulative minimum revenue value around $100 billion and projected customer cash deposits of $22 billion. Roughly 40% of revenue under those agreements carries fixed prices or price ceilings near current market levels, and management expects half or more of total revenue to end up under such contracts. This is a structural departure. Memory used to be renegotiated quarterly on a spot market; now a growing share of the book is locked in years ahead, with floor prices that hold gross margins above any prior peak.

That changes what a downturn, when it comes, looks like. Historically, memory downturns knocked revenue down 25–40% and gross margins from above 50% toward low-20s. With half the revenue contracted at price floors and AI inference demand acting as a much higher demand floor than consumer gadgets ever provided, the plausible trough is shallower — not eliminated, but no longer the cliff it used to be.

Where the risk actually sits now

The pullback is the market discounting a future that the contracts have pushed later, not one the report will resolve. The skepticism is concentrated in the parts of the business that remain genuinely cyclical, and it is worth naming them precisely rather than waving them away.

Roughly 60% of revenue is still exposed to spot pricing — that is the un-contracted, commodity side of the book, and it is the part that will fall when the cycle turns. The physical supply relief arrives on a schedule the market can already see: new fabs in Idaho, Taiwan, and New York ramp through 2027 and 2028, and management itself flags that supply additions come after 2028. The longest-dated risk named by analysts is Chinese memory capacity emerging to pressure pricing outside the U.S. Citi, while staying bullish, cut its price target in early August and forecasts memory prices peaking around the middle of 2027 as the pace of gains slows.

Set next to the numbers those fears are priced against, the tension resolves into a timing argument. The stock entered September at a low single-digit multiple of forward earnings precisely because investors refuse to pay normal-multiple prices for what they assume is a peak-cycle earnings print. Every memory boom before this one ended the same way, and the market's trauma discount is not irrational.

But the evidence that defined those past tops — uncommitted capacity, spot-only pricing, a pure commodity product — has been at least partly replaced by contracted scarcity and a demand base that scales with inference rather than handset upgrades. The honest framing is not that the cycle is dead, only that its shape has changed: demand runs with a technology on an exponential curve, while the leverage that used to guarantee the bust has been pushed roughly a decade out, into 2027–2028 supply and a remaining 60% spot book.

That is the real question the September 30 report will not answer, because the report will be beautiful. The market already knows what the last two quarters and the next one say. What investors are really paying or refusing to pay for is the judgment about that 2027–28 spot book and the capacity scheduled to land alongside it. Record margins were never the reassurance a cyclical stock needs — the reassurance is what happens to the margin after the record, and this time the floor under it is partly contractual. The pullback is the market betting the floor is lower than the company has contracted it to be.