Opinion: Has Micron Finally Broken the Memory Cycle?

Memory companies usually make their best money shortly before investors start worrying about the next glut. Micron is testing whether AI has changed that pattern.

The numbers certainly look like a peak. In Q3 FY2026, Micron reported $41.5 billion of revenue and an 84.6% gross margin. Adjusted free cash flow was $18.3 billion in a single quarter. DRAM prices rose by more than 60% sequentially and NAND prices by roughly 85%.

What makes this cycle unusual is what Micron has done with the shortage.

By June, the company had signed 16 strategic customer agreements, generally running through 2030. They contain take-or-pay commitments and price floors. Around 20% of DRAM volume and one-third of NAND volume were covered, with management aiming eventually to put more than half of revenue under these agreements. Customers have made about $22 billion of financial commitments, including roughly $18 billion in cash deposits.

For a business famous for building expensive fabs into uncertain future demand, that is a meaningful change. Micron now has customers committing capital alongside it.

The contracts should make the next downturn less brutal for the volumes they cover. They also have limits. Micron does not disclose the profit mix inside those agreements, and most of the business is still exposed to market pricing. Some contracts contain price ceilings as well as floors. Customers can default. Enforcement can end up in court.

HBM gives the argument for a longer cycle more weight. It consumes more wafer and cleanroom capacity per bit than ordinary DRAM and requires stacking, advanced packaging and customer qualification. Micron was already producing HBM4 for Nvidia's Vera Rubin platform in 2026, with HBM4E expected in 2027.

That makes supply slower to respond than it was in the old PC memory cycle. It does not make Micron irreplaceable. SK hynix is shipping HBM4, Samsung is scaling it, and both are signing customers of their own.

There is also a nasty feedback loop hiding in Micron's own filings. HBM uses so much capacity that weaker HBM demand could free wafers for conventional DRAM. More conventional supply would then hit the market just as demand was slowing. The technology that helps tighten memory today could worsen an ordinary DRAM glut later.

That matters because a lot more capacity is coming. Micron expects output from Idaho beginning in mid-2027, with additional production and packaging capacity arriving through 2028. Capital spending is rising with it.

The stock leaves little room for getting the duration wrong. At the September 24 price used in the research, Micron's enterprise value was about $1.22 trillion. Annualising Q3 free cash flow gives roughly $73 billion. A reverse DCF using a 10% discount rate and 3% terminal growth requires that already exceptional cash flow to keep growing by about 6% a year for five years. Starting from a more normalised $60 billion pushes the required growth above 10%.

So the argument around Micron comes down to duration.

If AI infrastructure keeps absorbing memory faster than the industry can build capacity, today's earnings may last much longer than previous cycles suggest. The contracts would make that period easier to finance and give Micron more protection when pricing eventually softens.

If AI spending slows before the new fabs fill, the old economics return quickly. HBM capacity can migrate back into DRAM, uncovered volumes lose pricing power, and expensive new capacity still has to earn a return.

There are useful tells to watch. Two quarters of steep DRAM or NAND price declines would be one. Inventory moving above 150 days would be another. So would stalled growth in contracted revenue, a major HBM qualification loss, or rising capex while customers begin cutting AI infrastructure budgets.

Micron has probably made the memory cycle less violent for part of its business. Whether it has made the cycle longer is the bet investors are now paying more than $1 trillion to make.