Semiconductors / AI Infrastructure

🔥 Micron Made $54 Billion in 90 Days — and Still Can't Make Enough Memory

Micron just booked $54.23 billion in a single quarter, up 379% year over year, as AI's hunger for high-bandwidth memory turns a commodity-chip maker into one of the most profitable hardware businesses on Earth.

By Tech Arcade··4 min read

Micron Made $54 Billion in 90 Days — and Still Can't Make Enough Memory — Tech Arcade
Photo: Brecht Corbeel / Unsplash

Micron just had the kind of quarter that breaks the spreadsheet. The memory-chip maker pulled in $54.23 billion in revenue in the three months ending September 3 — up 379% from the $11.32 billion it booked a year earlier, and well past the roughly $51 billion Wall Street had penciled in (Micron).

That is not a typo, and it is not a full year. It is a single quarter. Twelve months ago Micron was a cyclical commodity-chip company that investors treated like a weather system — sunny one year, flooded the next. This quarter it posted $37.7 billion in net income on an 86.8% gross margin, the sort of figures you expect from a software monopoly, not a factory that bakes silicon (Unite.AI).

The short version: the AI boom everyone counts in GPUs actually runs on memory — and the company that feeds the machines just turned into one of the most profitable hardware businesses on the planet.

🧠 Why This Matters

Every AI model you touched this year — the one drafting your emails, reading a scan, steering a warehouse robot — lives and dies on how fast it can shuttle data in and out of memory. The GPU gets the glory. High-bandwidth memory (HBM) is the part that keeps that GPU fed. Starve a $40,000 accelerator of memory bandwidth and it idles like a sports car stuck behind a school bus.

Micron is the only major HBM maker headquartered in the United States, and its chips sit shoulder-to-shoulder with Nvidia’s newest accelerators (BigGo Finance). So when hyperscalers pour hundreds of billions into AI data centers, a slice of every dollar lands on Micron’s loading dock. This quarter is simply what that looks like once it reaches the income statement.

📊 Deep Dive

Pull the quarter apart and the AI fingerprints are on everything. The growth is not coming from phones or laptops — it is coming from the racks.

  • Revenue: $54.23 billion in Q4, versus $11.32 billion a year ago (+379%) and $41.46 billion the prior quarter (+31% sequentially).
  • DRAM: $39.8 billion — 73% of all revenue, up 343% year over year.
  • NAND: $14.1 billion — 26% of revenue, up a frankly absurd 526%.
  • Data center: roughly $18 billion in the quarter, about 11 times the year-ago figure.
  • Full year: FY2026 revenue of $133.19 billion, up 256% from $37.38 billion, with $84.97 billion of net income and $89.68 billion of operating cash flow (Unite.AI).

An 86.8% gross margin is the number to sit with. In the commodity-memory era, Micron routinely ran margins in the 20s and 30s and sometimes lost money outright. When supply is this tight and demand this desperate, pricing power stops looking like a chip business and starts looking like rent.

“AI is becoming Super Intelligence (SI), and memory enhances this intelligence and the competitiveness of our customers’ platforms.”
— Sanjay Mehrotra, Micron CEO (Micron)

⚠️ The Catch

Before you mortgage the house, remember what Micron is. Memory has been the most violently cyclical corner of semiconductors for forty years: every boom has ended in a glut, and every glut has gutted margins. Nothing here repeals that physics.

The company is also spending to keep up. Net capital expenditure hit $27.37 billion in FY2026, with roughly $25 billion earmarked for just the first half of FY2027 (Unite.AI). Fabs take years to build and cannot be switched off when demand cools. That is exactly why, despite a blowout on every headline line, Micron’s stock barely moved — the heightened capex forecast rattled investors as much as the beat impressed them (Yahoo Finance).

And the demand is concentrated. A handful of hyperscalers are writing most of the checks. If even one trims its AI build-out, the order book thins fast.

🎯 What Happens Next

Micron is guiding to $61.5 billion (give or take $1.5 billion) for the current quarter — above the roughly $57 billion analysts expected — with non-GAAP earnings of about $38.15 a share (BigGo Finance). In plain terms: management thinks this quarter’s record is next quarter’s floor.

The real question is supply. HBM capacity is effectively sold out well into next year, which is why Micron is pouring concrete as fast as it can. Watch its Korean rivals Samsung and SK Hynix, who are racing to add capacity of their own — the moment the three of them collectively over-build, the pricing magic fades.

“Micron delivered record fiscal 2026 results, and we expect an even stronger fiscal 2027.”
— Sanjay Mehrotra, Micron CEO (Unite.AI)

🧩 Bigger Picture

For a decade the story of computing was processing power. The Micron quarter is a reminder that the binding constraint has quietly moved. You can have all the compute in the world, but if you cannot feed it fast enough, the silicon sits there drawing power and doing nothing.

Memory has become the choke point of the AI economy — the narrow pipe every model has to squeeze through. That makes the people who widen the pipe extraordinarily valuable right now, and extraordinarily exposed the day the rush slows. Micron is riding both edges of that blade at once.

The chips that remember are suddenly worth more than the chips that think. Right up until everyone remembers how this cycle usually ends.

Sources