Here's a nice problem to have: you're selling something that everyone wants, and you literally can't make enough of it. That's the enviable position Micron Technology (MU) finds itself in, according to CEO Sanjay Mehrotra. On Thursday, he laid out the stark reality of the AI memory market—demand is so far ahead of supply that Micron is only fulfilling a fraction of what customers are asking for.
"We are only able to supply, for our key customers in the midterm, about 50% to two-thirds of their requirements," Mehrotra told CNBC. Think about that for a second. If you walked into a car dealership and they said they could only give you half a car, or maybe two-thirds of one, you'd probably go somewhere else. But in the world of high-bandwidth memory (HBM) for artificial intelligence, there is no "somewhere else." The cupboard is bare industry-wide, and Micron is one of the few companies with a key to the pantry.
This isn't a short-term blip. Mehrotra told analysts that tight conditions for both DRAM and NAND memory are expected to persist beyond 2026. In the age of AI, where every tech giant is racing to build and train larger models, memory isn't just a component—it's what Mehrotra called a "strategic asset." It's the fuel for the AI engine, and right now, the tank is only half full.
When Demand Outstrips Supply, the Numbers Get Silly
This supply crunch translates directly to the financial statements, and the numbers from Micron's fiscal second quarter are the kind that make accountants do a double-take. The company reported revenue of $23.86 billion. Wall Street analysts, who are paid to guess these things, were expecting about $19.94 billion. So Micron beat by nearly $4 billion. That's not a beat; that's a different ballgame.
Earnings per share came in at $12.20 on an adjusted basis, cruising past the estimate of $9.21. "Micron set new records across revenue, gross margin, EPS and free cash flow in fiscal Q2," Mehrotra said, attributing it to "a strong demand environment, tight industry supply and our strong execution." He then added the kicker: "and we expect significant records again in fiscal Q3."
The company's guidance for the current quarter backs that up. They're forecasting revenue around $33.5 billion and earnings of about $19.15 per share. Both figures are, unsurprisingly, well above what the street had penciled in. When you can't make enough of your product, pricing power tends to be rather excellent.
Opening the Wallet to Open the Spigot
So, what do you do when you're selling every chip you can make? You try to make more of them. A lot more. CFO Mark Murphy detailed the plan: ramping production, including for the next-generation HBM4, and significantly increasing investment. The company now expects capital expenditures to exceed $25 billion this year, with plans to spend even more in 2027. This is largely about building new factories and production lines to ease the constraints.
"In the AI era, memory has become a strategic asset for our customers," Mehrotra reiterated, "and we are investing in our global manufacturing footprint to support their growing demand." This is the cycle in motion: insane demand leads to insane profits, which are funneled back into insane levels of capital spending to try and meet that demand. It's a virtuous circle if you're a shareholder, and a frustrating wait if you're a data center manager trying to build an AI server.












