Micron's AI memory boom is increasingly contracted years in advance

Four days after Micron posted a record quarter, the most important number from its earnings week isn't on the income statement. It's $150 billion — the value of future memory demand that customers have now contractually promised to buy.

On Wednesday's earnings call, Micron disclosed 26 signed strategic customer agreements — multi-year, take-or-pay contracts the company estimates will cover more than 35% of its revenue through 2030. Customers have put up $32 billion in financial commitments behind those deals, the vast majority of it as cash deposits. And remaining performance obligations — Micron's RPO — now stand at about $150 billion.

The subtext: the AI memory boom is increasingly being bought in advance.

The $150 billion, translated

CFO Mark Murphy was careful about what RPO means, and investors should be too. It isn't cash in hand or revenue already booked. It reflects, in his words, "the contract value for only SCAs that have a determined pricing framework, which can be either a fixed price or subject to a pricing floor and ceiling." It's built on committed volumes and minimum pricing — which is why Murphy calls it "inherently conservative."

Two companion points matter. First, three-quarters of the SCA revenue estimate carries a defined pricing framework, most of it with floor-and-ceiling bands; the rest gets priced periodically against the market. Second, Murphy said Micron expects actual revenue to "well exceed" the RPO over the life of the agreements — and that even at floor prices, margins should land "meaningfully above any prior cycle peak margins."

In plain terms: $150 billion is the low end of what customers have signed up for, not the ceiling.

Some coverage has dubbed this a "$150 billion backlog." That's the RPO by another name — but "backlog" implies finished goods waiting to ship, and this is contracted future demand, priced at minimums, for memory much of which hasn't been built yet. The company's term is the precise one.

2027's HBM is mostly spoken for

The tightest disclosure concerns high-bandwidth memory, the AI accelerators' most coveted input. Micron has "completed agreements for the vast majority of our calendar 2027 HBM bit supply," CEO Sanjay Mehrotra said — "with significant price increases year-over-year." On the Q&A, he added that a large part of 2027 HBM volume is already sold out at prices "much higher than 2026 prices."

Zoom out and the picture is broader than HBM. More than 75% of Micron's total 2027 output is already committed, across SCA and non-SCA customers alike, and Mehrotra said the majority of customer discussions have already moved on to 2028. Any new SCA pricing, he added, is being negotiated higher on prevailing market conditions.

Customers are even pushing duration: Micron has signed agreements extending into 2031, plus one-year extensions to 2031 on two existing deals.

Why the fabs can't catch up

None of this would matter if supply could respond quickly. It can't. Micron expects memory and storage supply-demand to be "much tighter in fiscal 2027 and 2028 than they were in 2026," with both DRAM and NAND "supply constrained in both years." Mehrotra was blunt about the horizon: "we do not have line of sight to when supply and demand will return to balance."

The capacity coming online explains why. Singapore's HBM advanced-packaging facility — running ahead of plan — starts output in early 2027. The Idaho ID1 fab begins wafer output in mid-2027. After that the dates stretch out: ID2 in late 2028, a Japan DRAM fab in late 2028, a new Singapore NAND plant in the second half of 2028, and the first New York fab not until 2030.

Micron is spending to pull that forward. Fiscal 2027 capex is rising versus prior plans — about $25 billion in the first half alone, including roughly $11.5 billion in the current quarter, with the second half running higher — and most of the increase is construction capex aimed at accelerating cleanroom availability from late 2028 onward. The company didn't give a full-year capex figure.

A different kind of cycle — maybe

Mehrotra reached for the structural argument: the agreements represent "a fundamental change to the nature of this industry... versus the past, when you would experience more volatility." Murphy went further, saying "the industry is structurally reset."

Neither declared the memory cycle dead, and Mehrotra paired the claim with a hedge — the need to keep "flexibility... particularly given the dynamic nature of our industry." The honest reading: contracted demand changes the shape of the next downturn more than it guarantees there won't be one. Take-or-pay volumes put a floor under Micron's planning; they don't repeal semiconductors.

The other side of the trade

Analysts pressed exactly that point on the call. Bank of America asked whether 2027 might mark a pricing and earnings peak; TD Cowen raised the prospect of a large customer de-speccing HBM. Mehrotra's answer in both cases was the tightness thesis — demand running so far ahead of supply that the peak question is premature.

What's missing is also worth noting. Management didn't address China competition, customer concentration, or how durable AI infrastructure spending proves to be — the risks live in the usual forward-looking-statements boilerplate, not in the call's substance. The contracted book de-risks Micron's side of those questions; it doesn't answer them.

Micron shares have more than tripled this year, Reuters noted on Sept. 30 — the market has already priced in a great deal of this story. What the SCA disclosures add is duration: Wednesday's record quarter was about what Micron sold; the $150 billion is about what's already been bought.

TickerGrove's Sept. 30 earnings recap covered the quarter itself — record $54.23 billion in revenue, adjusted EPS of $33.42, and first-quarter guidance well above consensus. The next checkpoints are execution: Singapore's HBM packaging line in early 2027, ID1 wafers by mid-year, and whether that contracted demand keeps compounding.