Micron Technology, Inc. MU

Key data

The business

Micron makes memory chips, the two kinds that hold data while a computer works. DRAM is the fast working memory in every server, phone and PC. NAND is the slower flash that stores files. Both are commodities: interchangeable, priced by the market, made by a handful of giants. In DRAM only three are left standing, Samsung, SK Hynix and Micron, and that scarcity of suppliers is the whole reason the business can make money at all.

The product that matters now is high-bandwidth memory, a stack of DRAM wired directly beside an AI accelerator. Every Nvidia and AMD training chip needs a ring of it, and there is not enough. Qualifying a supplier takes years, so the three incumbents share a market they cannot be quickly pushed out of. This is the chokepoint the whole AI build-out runs through, and for the first time in Micron's history it is the most profitable line rather than a cyclical afterthought.

The moat is real but narrow: a three-player structure, the tens of billions of capital a fab demands, and a process-node lead measured in nanometers. None of that stops the cycle. Memory has always been feast or famine, and Micron has lost money at the bottom of every cycle it has run. What you are buying is the best possible seat in a business that has never once stopped being cyclical.

The numbers

The sequence below is not a growth story, it is a detonation.

Revenue quadrupled year over year and net income rose fifteen-fold, but the last cell is the story: $24.67 of earnings in one quarter, more than Micron earned per share in the three prior fiscal years combined. Each quarter this year landed well above the consensus estimate, the June quarter by about 18%. This is what the top of a memory cycle looks like when it collides with a genuine shortage.

Step back and the cyclicality is unmissable. FY2023 was a $5.8 billion loss. Two years later the company earned more in nine months than in its best five years put together, and 9M revenue of $79.0 billion is already more than double all of FY2025.

At $1,016 the stock costs about 23 times the past year's earnings and 11 times book, both above where Micron has been priced for most of the last twelve years. Run it forward and the math inverts: analysts model $170 of earnings for fiscal 2028, putting the stock near 6 times. A quality grower earning $170 durably would fetch $2,500 to $3,000. At 6 times, the market flatly refuses to believe FY2028 resembles that number. That disbelief is the entire question. My read: the truth sits between, HBM's three-player structure holds the next trough higher than any before it, but higher is not gone, and the print that settles it is the direction of gross margin over the next two quarters.

Read the trough row. In FY2023 gross margin went negative while Micron still spent $7.7 billion on plant and threw off almost no cash. That is the shape of the risk: this business must spend tens of billions through the bottom whatever prices do. Margin ran 77% in 9M FY2026 against negative 9% two years earlier, an 86-point swing no amount of AI narrative makes permanent.

Management

The insider record runs one way: zero purchases in twelve months against 95 sales worth $138 million. The largest single sale was $14.0 million by Sumit Sadana on August 18. Plan status is not disclosed, so whether these are automatic 10b5-1 sales or discretionary calls cannot be told from the filings, and that distinction matters at a top. Capital allocation, by contrast, turned conservative in the right way. Buybacks were zero in FY2025. The windfall went to the balance sheet: cash rose from $9.6 billion to $25.0 billion, long-term debt was cut from $11.5 billion to $8.8 billion by November 2025, and a stack of senior notes was repaid ahead of schedule. The survivability question that hung over prior downturns has been answered, Micron can weather the next one comfortably.

How it fails or surprises you

The cycle turns. Memory has crashed every three to four years for forty years. At 85% gross margin pricing has one direction, and when DRAM and HBM selling prices roll over the fall through a fixed cost base is violent, a 40-point margin swing inside a year is ordinary here. The first tell is sequential gross margin: two down quarters and the peak is in.

HBM stays sold out (right tail). If AI demand keeps HBM capacity fully booked through FY2027, the three suppliers hold pricing and trough earnings reset permanently higher. At 6 times a $170 forward number, the market pays nothing for that durability. Multi-year HBM supply agreements at firm prices would be the print that forces a re-rating.

Insiders know the price curve. The people closest to the selling-price data are selling, all of them, at the exact top. If durability were as strong as the bulls claim, you would expect at least one open-market buy. Continued discretionary selling into the next quarter would say management sees the top ahead.

Closing thoughts

The market is pricing an uncertainty that two quarters of gross margin and pricing data will resolve. If margins hold or climb, the durability case is live and 6 times forward is a mispricing. If they roll, the cycle has turned and the $170 number evaporates. The left tail is fatter in probability, a normal memory crash is the base rate, but it is not fatal: with $25.0 billion of cash and debt cut, a downturn is a drawdown in the stock, not a threat to the company. The right tail, a structural HBM shortage that resets earnings permanently higher, is less likely and pays multiples if it lands.

The bet is still that AI data centers keep buying every chip Micron can make and keep paying up for the memory stacked beside their accelerators, and that the memory business does not crash before you get out. What breaks it is one pair of numbers: gross margin and average selling price turning down together. Watch those two and little else. Everything the stock is worth, in either direction, comes down to whether this cycle behaves like all the others or finally does not.