For nearly forty years, memory chips followed the most reliably destructive cycle in technology. The mechanism was mechanical. Prices climbed, every manufacturer rushed to build new cleanrooms at once, and two years later that tsunami of fresh silicon arrived just as demand cooled. Prices cratered, balance sheets bled, and everyone slashed spending until the shortage returned. In 2026, that machine finally broke.
Micron Technology makes the digital memory that keeps computing alive. The industry runs on two distinct formats. DRAM is volatile working memory—the desk a processor works on, which clears the instant power cuts out. NAND flash is permanent storage.
Toshiba invented it in 1987, and you live with it daily inside your phone, holding your photo library and offline Spotify playlists without battery drain. When packaged with a controller into enterprise solid-state drives, NAND becomes the storage layer for every cloud provider on Earth.
As the only major memory maker headquartered in America, Micron has lived on the sharp end of this whip. In fiscal 2023, after personal computer and phone sales slumped, Micron's revenue plummeted by half to $15.54 billion. It lost $5.83 billion and printed negative gross margins. By fiscal 2026, the picture inverted completely.
Driven by an insatiable server market, Micron booked $133.19 billion in revenue and $84.97 billion in net income, posting an 80.7% gross margin. In its final quarter, gross margin reached 86.8%.
What changed was the nature of artificial intelligence. Training large language models was mostly an accelerator and DRAM story. But deployment—answering queries and running autonomous agents—requires massive, persistent storage. Enterprise SSD bit demand surged over 80% in 2026.
TrendForce noted that SSDs are no longer merely passive data lockers; they now support active computational workloads. When reasoning models generate massive conversation caches, that data spills from scarce, expensive DRAM straight onto fast flash storage. Micron designed drives specifically to catch this spillover.
Meanwhile, the traditional escape valves jammed. When flash memory prices climbed in previous cycles, hyperscalers pivoted to cheaper mechanical hard drives. That door is shut. Hard drive maker Seagate saw its nearline drives allocated into calendar 2028, locking up capacity.
At the same time, memory producers added virtually no new NAND wafer capacity in 2026. After losing billions in 2023, suppliers directed cleanroom space and equipment budgets almost exclusively to high-bandwidth memory for AI chips. NAND ran a 4% to 5% supply deficit all year.
This shift explains why Micron's equity looks so unusual. Shares trade at roughly five times forward earnings despite software-tier margins. That compressed multiple means the market assumes this windfall is an unsustainable spike. Yet Micron has rewritten its commercial plumbing.
Management locked in multi-year take-or-pay customer contracts featuring fixed price bands, backed by $12.75 billion in customer deposits. In its annual filing, Micron stated that even at contractual floor prices, these deals yield margins meaningfully above its peak quarterly margins in past cycles. If those floors hold, the business ceases to be a wild commodity play and begins to trade like contracted infrastructure.
A short-seller would counter that every memory supercycle ends the exact same way. The supply response is already funded and massive. Kioxia and SanDisk announced joint investments exceeding $31 billion in Japan through 2032 to expand production. Micron is ramping its own capital spending toward roughly $25 billion in just the first half of fiscal 2027.
China's YMTC is already shipping roughly 14% of global NAND units, mass-producing 267-layer chips with state backing. TrendForce projects global supply will outpace demand in 2027, pushing the market into surplus by the second half. Hyperscalers running negative free cash flow could easily stretch deliveries. If they do, there is no consumer safety net; global smartphone production is forecast down 15% to 20% in 2026.
The test lands in the second half of 2027, when TrendForce expects the supply balance to flip positive. If enterprise SSD contract prices hold above historical peaks, the cycle's forty-year rules are officially dead. If take-or-pay clauses get quietly renegotiated under an avalanche of fresh Japanese and Chinese silicon, the old machine will claim another victim.
Disclosure: The Signal holds no position in MU. Positions may change. This is not financial advice.
NAND spent four decades as the most predictable crash in technology: every shortage carried the seed of the glut that ended it. What is different this time is not that demand is strong, it is that there is no substitute at the bulk tier, because nearline hard drives are allocated into 2028, and the producers spent three years refusing to build. That lends Micron extraordinary pricing power now. It does not repeal the machine. The Japanese and Chinese fabs now under construction arrive in 2028, and TrendForce already expects the supply balance to flip in the second half of 2027. So the one number that settles the argument is not the record margin, it is whether contract floor prices still clear the last cycle's peak when the new bits land.
