Here's a sentence I never thought I'd write: Micron Technology is about to become the third most profitable company in America. Yes, that Micron — the Boise chipmaker that spent decades selling DRAM chips like they were potatoes. The same company that, five years ago, was scraping by on razor-thin margins, eating Intel's dust, and begging for a seat at the grown-ups' table.
Look at the numbers. Wall Street expects Micron to post gross margins north of 62% next quarter. That's higher than Apple. Higher than Microsoft. Higher than almost any company you can name that isn't Nvidia or Google. The only two U.S. firms projected to be more profitable are the AI chip king and the search engine that's been buying up every H100 it can find.
How did we get here? One word: memory. But not the kind you stick in your laptop. We're talking high-bandwidth memory — HBM — the stuff that sits next to Nvidia's AI accelerators and feeds them data at the speed of light. For years, HBM was a niche product, too expensive and too complicated for anything outside supercomputers. Then generative AI hit, and suddenly every hyperscaler in the world needed it. Like, yesterday.
Micron blinked, then pounced.
The AI memory gold rush
Big Tech's appetite for AI compute is bottomless. Meta wants to run its recommendation algorithms on custom silicon. Microsoft is building out Azure's AI infrastructure as fast as it can string fiber. Amazon's AWS is churning out Trainium chips. And every single one of these systems needs high-bandwidth memory to function. Without HBM, even the most powerful GPU is just a very expensive paperweight.
Micron is one of only three companies on earth that can make the stuff, alongside Samsung and SK Hynix. But Micron has a secret weapon: its HBM3e, the latest generation, which is already inside Nvidia's H200 and B100 GPUs. The company locked in supply deals with the hyperscalers early, and now it's reaping the rewards. In the last quarter alone, Micron's revenue jumped 78% year-over-year to $8.2 billion. More than half of that came from HBM and data center SSDs.
There's a catch, though. This isn't a sustainable moat — it's a window. Samsung and SK Hynix are ramping their own HBM3e production as fast as they can. By next year, supply will likely catch up with demand, and margins will compress. But that's next year's problem. Right now, Micron has pricing power it has never seen before. The company is selling HBM for three to five times the cost of standard DRAM, and customers are still buying every wafer it can produce.
Why this time is different
The cynic in me wants to point out that memory has always been a boom-and-bust business. I covered Micron in the mid-2010s, when the company was stuck in a death spiral with DRAM prices falling off a cliff. The stock was a value trap. Investors who bought at the peak of the last cycle watched their money get cut in half. So why should anyone believe this time is different?
Because AI isn't a cyclical product cycle. It's a structural shift. Cloud providers aren't building servers for next quarter's traffic; they're building for a decade of exponential compute demand. The hyperscalers have committed hundreds of billions of dollars to AI infrastructure, and memory is the bottleneck. Standard DRAM is commodity; HBM is a specialty product with higher margins and stickier customer relationships. Once you're qualified into a hyperscaler's supply chain, you're not easily replaced.
Moreover, Micron has cleaned up its own house. Under CEO Sanjay Mehrotra, the company has shifted its product mix toward high-value memory, cut costs, and closed older fabs. Operating margins have gone from negative to over 30% in two years. The balance sheet is solid, with $12 billion in cash and short-term investments. The turnaround is real.
The risks nobody's talking about
Let's not get carried away. There are three big threats that could pop this balloon.
First, geopolitics. Micron's reliance on China for manufacturing is a risk — almost all of its DRAM is made in Taiwan and Japan. Any disruption in East Asia, whether from a Chinese invasion of Taiwan or a natural disaster, could shut down production for months. The company is building a new fab in Idaho, but that won't come online until 2028 at the earliest.
Second, technology change. HBM is essentially a stack of DRAM dies connected by through-silicon vias. That's a mature technology. What if Nvidia or its competitors decide to integrate memory directly onto the GPU package, bypassing HBM altogether? Micron would lose its golden goose. The company is investing in new memory architectures, but nothing has been proven at scale.
Third, competition. Samsung is the 800-pound gorilla of memory, with deeper pockets, better brand recognition, and a longer history of vertical integration. It's already shipping its own HBM3e and has announced a next-generation version. SK Hynix is no slouch either. Micron's lead in HBM is real, but it's measured in quarters, not years.
“Micron's lead in HBM is real, but it's measured in quarters, not years.”
For now, the market is pricing in a best-case scenario. Micron's stock has tripled in the last 18 months, giving it a forward P/E of over 30. That's rich for a memory company, even a profitable one. If AI hype cools or demand disappoints, the stock could fall hard.
The bottom line
Micron is experiencing a once-in-a-generation windfall. The AI revolution has turned a middle-tier chipmaker into a profit machine. For the next six to twelve months, the company will mint money faster than almost any other U.S. corporation. But the window won't stay open forever. Investors who buy in now are betting that Micron can extend its lead, manage its risks, and turn this cyclical boom into a structural advantage.
Maybe it can. Maybe it can't. What's certain is that the old Micron — the one that was always a step behind, always fighting for scraps — is dead. The new Micron is rich, arrogant, and pounding the table. And for the moment, they have every right to.
