Micron Locked In $100B of Take-or-Pay Demand. That Breaks the Memory Cycle.

The Contract Book

Micron's fiscal Q3 print was a clean blowout, but the line that actually moved the stock sits in the slide deck, not the income statement. Management disclosed 16 Strategic Customer Agreements representing roughly $100 billion in minimum contracted revenue, structured as take-or-pay deals with binding volume commitments running through calendar 2030 (three-year terms for the automotive customers).

These aren't soft letters of intent. Fourteen of the sixteen already carry a cumulative floor of about $100 billion at minimum price, and Micron expects to collect roughly $22 billion in cash deposits and financial commitments — about $18 billion in cash and $4 billion in letters of credit — to backstop them. Customers are paying, up front, for the right to buy memory.

Why Take-or-Pay Breaks the Memory Cycle

Memory has been the textbook commodity cyclical for four decades: capacity gluts, price collapses, margins to zero, repeat. The reason MU has historically traded at a single-digit multiple is that nobody could underwrite the next trough — you were always one oversupply quarter from watching pricing crater.

Take-or-pay attacks exactly that. Micron now expects fixed prices or ceilings at or near current levels to cover roughly 40% of total revenue once all planned agreements are live. When that much of your book is contractually price-protected, the downside leg of the cycle gets truncated — the floor stops being a market-clearing price and starts being a contract term. That, not the margin headline, is the entire bull case for re-rating a memory maker off cyclical-trough math. The Street is repricing fast: JPMorgan moved its target from $550 to $1,540, and desks across the board reset by hundreds of dollars in a single session.

What the Print Showed

The underlying numbers gave the contracts credibility. Revenue hit $41.46 billion, up 346% year over year, with adjusted gross margin at 84.9% — software-tier economics on a DRAM and NAND business. The Core Data Center unit did $11.5 billion at 87% margins, up from $1.53 billion a year ago — roughly a sevenfold jump as HBM and high-capacity DRAM absorbed AI buildout spend.

Guidance was the exclamation point. Micron guided Q4 to $50 billion ±$1 billion at about 86% gross margin and $31 EPS, against a Street modeling closer to $43 billion. When the forward guide beats consensus by seven billion dollars, the contracted-demand story stops reading as a slide and starts reading as a run-rate.

The Risk Side

None of this makes MU a one-way trade. The stock is up more than 20% to roughly $1,221 and ran hard into the print, so the easy money on the reaction is already gone; thin after-hours and overnight liquidity exaggerate the percentage move on a perp.

The deeper caveats are structural. Take-or-pay floors protect price, not Micron's own volume bets — if AI capex disappoints, customers still pay the minimums, but Micron's capacity expansion could overshoot the real demand. The contract book is a 2026-2030 story, and execution on the HBM4 ramp plus supply discipline from SK Hynix and Samsung will decide whether those floors ever get tested. The thesis is that this cycle is genuinely different. The risk is that "this time is different" has buried memory bulls before — the contracts are what's supposed to make this round the exception, and they're brand new.