Micron Took $22B in Binding Deposits. The Memory Cycle Just Changed.

The Number That Wasn't on the Income Statement

Micron's fiscal Q3 2026 results cleared the bar by a wide margin: revenue of $41.46 billion, up 346% year over year, non-GAAP EPS of $25.11 against roughly $20.28 expected, and a record gross margin near 85%. GAAP net income landed at $28.24 billion. Management then guided Q4 to about $50 billion in revenue with EPS near $31 — well above the ~$43.6 billion the street was modeling.

Those are the headline numbers, and they're why the stock is up 20.64% to roughly $1,208. But the most important detail in this print isn't a margin or a guide. It's a balance-sheet item: the cash hyperscalers have already wired to Micron before a single additional chip ships.

Take-or-Pay Rewrites the Risk

Micron disclosed 16 binding, take-or-pay Strategic Customer Agreements totaling more than $22 billion in commitments — roughly $18 billion arriving as cash deposits and about $4 billion as letters of credit. These aren't the quarter-to-quarter handshake pricing that has defined memory for decades. They're multiyear contracts with set price bands and prepayments, which means the customer eats the cost whether or not it takes delivery.

That changes what MU actually is. CEO Sanjay Mehrotra said Micron can currently fill only half to two-thirds of HBM demand, with the entire 2026 HBM supply already sold out. When buyers are prepaying billions to reserve capacity that doesn't exist yet, the seller is holding pricing power the memory industry has almost never had.

The Glut Bears Were Right About Supply, Wrong About Power

Two days before this print, MU had given back roughly 24% on a Korea-led memory glut scare — the read that SK Hynix slowing its HBM4 ramp signaled cracking demand. The deposits invert that thesis. Demand was never the question; the binding contracts and the sold-out 2026 book prove buyers are scrambling for supply, not walking away from it.

Micron now expects HBM TAM to cross $100 billion in 2027 and lifted FY26 capex to around $27 billion to chase it. In an oligopoly of three — Micron, SK Hynix, and Samsung — with demand outrunning supply through and beyond 2027, the structural argument is that the classic memory boom-bust cycle is, for now, broken. That's the rerate the market is pricing, not just one good quarter.

What Could Still Bite

None of this makes MU a one-way trade. The stock is up more than 700% over the past year and trades near a record valuation around a $1.2 trillion market cap, which means a lot of the cycle-break thesis is already in the price. Post-earnings profit-taking and rotation out of crowded semiconductor names are the obvious near-term risk, and any sign that AI capex growth is moderating would hit memory first and hardest.

The MU perp ran roughly $900 million in 24-hour volume into this move, and a 20%-plus candle on a leveraged book cuts both ways — the same thin liquidity that powered the squeeze can reverse it fast. The structural story is strong; the entry is where the risk lives.