Memory Is The Real Bottleneck and Micron Seems To Own It For Now

Welcome to the subscriber only version of Investing in AI. Today we look at Micron, an AI rocketship that has recently fallen a bit with the semiconductor sell-off, and we ask what kind of future MU could have. Note that we are AI people not financial people so, please do your research before investing. Disclosure: We own a position in MU and have for over a year.

Micron Technology is up sharply over the past year. Its market cap sits north of $1 trillion. Its shares recently touched all-time highs near $1,200. By any normal reading, this is a stock that has already had its moment.

So here is the number that should stop you: Micron trades at roughly 6 to 10 times forward earnings. For context, the semiconductor industry median forward multiple is about 35x.

Micron’s PEG ratio (price-to-earnings relative to growth) sits near 0.04, one of the lowest of any large-cap stock in the market. Wall Street’s average price target of roughly $1,486 implies over 50% upside from here, and that’s the consensus, not the bull case.

That is a strange combination, no matter how we look at it.

A stock at all-time highs, at the center of the single hottest theme in markets, priced like the market expects its earnings to fall off a cliff. Either the market is right and Micron’s current profitability is a mirage, OR the market is still pricing this company as the same cyclical, boom-bust commodity business it has been for forty years, and hasn’t updated for what actually changed on its June earnings call.

This week’s thesis: the second explanation is closer to correct, and the newly disclosed contract structure (combined with how long independent forecasters expect memory to stay tight( gives a rough answer to the only question that matters for the multiple: how long does Micron need to keep proving this before the Street believes it?

The easy version of the Micron story is “picks and shovels” — a supplier riding the AI capex wave, the same framing applied to every chip name since 2023. That framing understates what’s actually happening. Micron isn’t riding the AI buildout. It’s gating it.

On the fiscal Q3 2026 call, Chief Business Officer Sumit Sadana made a striking admission: bit shipment growth is no longer really a function of demand. It’s a function of supply. Demand for memory — DRAM, HBM, NAND — is now so far above what the industry can produce that the growth rate of AI infrastructure itself is effectively capped by how many chips Micron, Samsung, and SK Hynix can physically manufacture.

“Demand continues to be well above our supply... this is all we can do in this time frame.” — Sumit Sadana, Micron Chief Business Officer, describing customer requests for HBM capacity through 2028

The numbers behind that statement are worth sitting with. CFO Mark Murphy disclosed that Micron now expects the total HBM market to cross $100 billion in calendar 2027 — a milestone the company had previously guided to 2028. Every AI accelerator shipped today needs HBM stacked directly onto it; there is no version of the AI buildout that doesn’t run through memory capacity. When Nvidia, AMD, or any other AI chip designer sells a GPU, the ceiling on how many they can ship is partly set by how much HBM Micron and its two competitors can produce — not by demand for the chip itself. That makes Micron less a rider of the AI cycle and more a structural constraint on its speed.