Micron's Crash Is Priced for a Memory Bust the Booked Numbers Don't Show

In memory chips, every few years the same movie plays: profits explode, suppliers pour billions into new factories, supply floods the market, prices collapse, everyone bleeds, and the cycle starts over. Micron has lived that plot in basically every decade it has been public — which is why a stock that crossed $1 trillion for the first time in May and became the AI boom's biggest winner can still spend the summer falling. It now trades near $1,000, roughly 20% below its record high, after an earlier correction of about 30% from the June peak.

The market's reflex on any memory name that has tripled is to assume the bust is next. The reason to slow down is that the company has just put up a quarter that does not fit that script.

For the fiscal third quarter ended in late May, Micron reported revenue of $41.5 billion, up 346% from a year earlier and a fifth straight record. Non-GAAP earnings per share of $25.11 were up 1,215% year over year on a record gross margin near 85%, against 39% the same quarter a year ago. It guided the current quarter to roughly $50 billion in revenue and gross margin around 86%. The balance sheet is net cash, and free cash flow over the trailing year runs near $26 billion. So the opening fact under the crash is a disconnect: the price has fallen while the business accelerated.

What is actually different this time

The center of the bull case — and it is genuinely new for memory — is that Micron has started to sell chips the way Nvidia sells GPUs rather than the way commodity parts used to be sold. It has signed 16 multi-year, take-or-pay "strategic customer agreements" worth roughly $100 billion in minimum committed revenue through calendar 2030, covering about a fifth of DRAM volume and a third of NAND, with a little under half of revenue carrying fixed or ceiling prices. Customers have put up about $18 billion in cash deposits plus $4 billion in letters of credit. The fastest-growing product, high-bandwidth memory, is fully booked through 2027 with demand running into 2028.

Historically, memory was priced on the spot market, which meant a glut destroyed revenue almost overnight. Contracted, pre-priced, take-or-pay revenue is designed to smooth exactly that. If it holds up, it breaks the boom-bust logic that has always justified punishing a memory stock for peak earnings. It is the strongest anti-cycle evidence the industry has ever produced.

The bear fact the market has already priced

But this is still an industry where one supplier's record profit is, in effect, the seed of the next oversupply. Today's margins are funding a synchronized build-out across Samsung, SK Hynix, and Micron — Micron alone is spending more than $25 billion this fiscal year and has committed over $100 billion to new factories. Korean brokerages are openly asking whether 2027 will be the peak, and a fourth DRAM producer, China's CXMT, is now a listed, expanding supplier even if it cannot yet touch high-end HBM. Sell-side targets range widely, from roughly $361 to $2,200 — a dispersion that is really a disagreement about whether the cycle has ended, not a disagreement about this quarter.

Here is the part that has to be acknowledged honestly: the market is not ignoring this risk, it is pricing it. At around $1,000 the trailing multiple of about 22x looks almost absurd against 346% growth. But the forward multiple the data services compute sits near 158x, which only reconciles with today's price if you assume next year's earnings collapse to a small fraction of the current run-rate. The market's base case already is a memory bust. The crash is not Wall Street failing to notice the good numbers; it is Wall Street refusing to believe they will last.

That reframes the opportunity. The bull case is not that investors are too slow to appreciate Micron. It is that the market has baked a full downcycle into the forward price while the evidence on hand — sold-out capacity through 2027, contracted revenue through 2030, net cash, and an industry that is deliberately holding supply back — does not yet justify that much doom.

The honest read

I am not going to argue for chasing this. The durability question is real, and it is a 2027–2028 question that today's reports cannot answer. The bust, if it comes, arrives after the new capacity ships — not now. And price has not de-risked: relative strength sits near 58, well short of oversold, and the stock is still holding above its 50- and 200-day averages. That is a pullback in an uptrend, not sellers exhausted. Buying "another leg higher" before either the price washes out or the capacity reports keep confirming scarcity is a timing bet the market is not giving you yet.

What the crash has actually produced is a test anyone can watch resolve. If booked capacity stays booked through 2027 and margins hold near the guided 80s, then the forward-multiple doom pricing was wrong and the next leg resolves up as the oversupply the bears fear fails to materialize. If the synchronized build-out cracks pricing instead — or hyperscaler AI spending slows and the take-or-pay contracts turn out to be less binding than advertised — then the market was right, and this was memory doing what memory always does.

The invalidation is concrete: watch the 2027 bookings and the margin line. The moment capacity stops being pre-sold, or gross margin rolls over from the 80s, the "crash equals opportunity" frame dies. Until then, the gap between what Micron is actually booking and what its forward valuation assumes is the largest disconnect in this trade — and the burden of proof sits with the bears, who need a bust the booked numbers do not yet show.