Micron's stock has done something violent in both directions this year. It ran up more than 700% to a record near $1,089 and a market value above $1 trillion, then got caught in a sector-wide semiconductor selloff that wiped out weeks of gains in a single session. The round trip captures the argument the market is having about the stock, and the argument turns on one specific claim: that Micron has escaped the memory cycle by selling its entire 2026 output of high-bandwidth memory in advance, on binding, fixed-price contracts.

That claim is the whole thesis, and it is worth examining on its own terms, because it is more concrete and more testable than the usual "this time is different" debate about memory chips.

Why the contract structure is the actual argument

Memory has always been the most brutally cyclical corner of the chip industry, because DRAM and NAND are commodities priced on the spot market, where a glut sends prices off a cliff and a shortage sends them soaring, and the swings gut earnings on the way down. That volatility is why the market historically refused to pay a high multiple for memory makers no matter how good a given year looked. A great year was assumed to contain the seeds of the bad one to follow.

Micron's high-bandwidth memory business is being sold as a break from that pattern, and the mechanism is not vague optimism. It is contract structure. CEO Sanjay Mehrotra said the company completed agreements on price and volume for its entire calendar 2026 HBM supply, which means the HBM revenue for the year is locked before the chips ship. Demand is reportedly outstripping supply by an estimated 50 to 67%, and the margins the structure produces are extraordinary: gross margin around 75%, roughly triple the 30-to-45% that characterized memory in prior cycles.

The bull case follows directly. If the price and volume are contracted in advance, the spot-market volatility that defined memory does not reach this revenue. A commodity sold on a binding annual contract behaves less like a commodity and more like an annuity, predictable cash flow rather than a bet on where prices settle. That is why the market re-rated Micron toward a valuation it never gave a memory maker before, and it is a coherent, evidence-backed argument rather than hype.

What the contract genuinely fixes, and what it only postpones

Here is the distinction that the sold-out headline blurs, and it is the key to the whole question. A one-year contract removes price volatility for one year. It does not remove it forever. It moves the risk to the renewal.

Think about what a fixed-price annual contract actually does. Through 2026, Micron knows its HBM price and volume, so the cyclicality is genuinely suppressed for that window, and the revenue visibility is real, not an illusion. But contracts expire. When the 2026 agreements come up for renewal, they will be renegotiated at whatever the supply-demand balance is at that moment. If HBM is still scarce, the new contracts are struck at high prices and the good times extend. If capacity has caught up with demand by then, the renewal happens at lower prices, and the cyclicality that the contract postponed arrives on schedule, just a year later than the spot market would have delivered it.

So the contracts do not repeal the cycle. They time-shift it. They convert immediate spot volatility into a periodic renewal risk, which is a real improvement in visibility and a real reason to pay a somewhat higher multiple, but it is not the escape from cyclicality that "sold out through 2026" can be read to imply. The question the bull case has to answer is not whether 2026 is locked in, it plainly is, but what the supply-demand balance looks like when the contracts roll, and that depends on the one variable no contract controls.

The variable that decides it: how fast the rivals build

That variable is capacity, and it is where the memory cycle has always reasserted itself. The reason memory booms end is not that demand collapses. It is that high prices and fat margins are an irresistible signal to every producer to build more, and the new capacity arrives together and floods the market.

That process is underway now. Micron is one of only three companies that make essentially all the world's DRAM, alongside Samsung and SK Hynix, and all three are racing to expand HBM production as fast as they can. HBM is harder to manufacture than commodity DRAM, which slows the flood and is part of why this cycle has run hotter and longer than past ones. But "harder" is not "impossible," and 75% margins are precisely the incentive that, historically, pulls enough new supply into the market to break the pricing eventually. The contracts protect Micron through 2026. Whether the renewals are struck at high or low prices depends on how much HBM capacity Micron's two rivals have brought online by then, which is the classic memory dynamic operating on a one-year delay.

This is the same structural tension that runs through the whole HBM trade, and it does not resolve in Micron's favor just because its current year is contracted. It resolves based on the collective capacity decisions of three companies all responding to the same enormous margins, which is exactly the setup that has ended every prior memory cycle.

The demand side carries its own dependency

There is a second contingency the contracts do not remove, and it connects Micron to the broader AI-spending question rather than insulating it. The reason HBM demand is outstripping supply is that hyperscalers are spending enormous sums on AI infrastructure, projected above $600 billion in 2026, and HBM is essential to the AI accelerators that spending buys.

Micron's sold-out order book is only as durable as that demand. The 2026 contracts are binding, so near-term revenue is protected even if sentiment shifts. But the renewals, again, depend on hyperscalers continuing to build at the current pace, which depends on their belief that AI will generate returns justifying it. If that spending decelerates, HBM demand softens, and the renewal contracts are negotiated in a weaker market. So Micron's escape from the cycle is doubly conditional: it holds through 2026 by contract, and beyond that it depends on both rival capacity staying constrained and hyperscaler demand staying strong. Neither is something Micron controls, and the June selloff, triggered partly by a peer failing to lift its AI outlook, was the market briefly repricing exactly that dependency.

How to read it

The balanced reading is that Micron has genuinely improved its position without genuinely escaping its nature. The sold-out, fixed-price 2026 book is real, the margins are real, the revenue visibility is real, and the shift from spot-priced commodity toward contracted supply is a legitimate reason the stock deserves a higher multiple than memory makers used to get. Investors dismissing the whole thing as just another memory bubble are underweighting a real structural change in how the highest-value part of the business is sold.

But the stock's violence in both directions reflects a market that has not settled whether the change is permanent or temporary, and the honest answer is that the contracts make it temporary-but-longer rather than permanent. They suppress the cycle for a year and move the risk to the renewal, where it will be decided by rival capacity and hyperscaler demand, the two classic forces that have always governed memory, neither repealed by a contract. The useful question for anyone watching is not whether 2026 is locked in. It is what the HBM supply-demand balance looks like when Micron sits down to negotiate 2027, because that is the moment the market is really pricing when it argues about this stock. A sold-out year is a strong hand. It is not a new deck.

Further reading