MU
Memory: the unglamorous stuff making AI expensive
If you’ve been wondering why Micron keeps acting like the market’s favorite overachiever, here’s the plot twist: the boring bits of AI infrastructure are getting very expensive, very fast. TrendForce says DRAM and NAND could gobble up 68% of spending by the nine biggest cloud companies in 2027, up from 47% this year.
That matters because memory isn’t some side quest anymore. It’s becoming the big-ticket item in the AI buildout — the part where hyperscalers look at the bill and go, “Wait, this much for RAM?”
The math is doing backflips
TrendForce’s forecast is the kind of number that makes your spreadsheet sweat:
- combined capex by those nine cloud giants could hit $1.383 trillion in 2027
- at 68% to memory, that’s roughly $940 billion aimed at DRAM and NAND
- server DRAM contract prices are already climbing hard, and HBM is getting pricier too
So yes, the AI boom is still alive. But the profit pool may be shifting away from the obvious names and toward the suppliers feeding the beast.
Why Micron keeps popping up
Micron is the cleanest public-market way to play the theme. The company is already riding a brutal rerating, and this article is basically saying the runway may not be done yet if memory stays tight and hyperscalers keep buying like there’s no tomorrow.
There’s a catch, of course. If memory gets too expensive, customers can get creative — smaller configurations, different chip architectures, or just squeezing spending elsewhere. That’s the kind of push-pull that can keep the trade spicy instead of one-way.
Big picture
This is the rare market setup where the bottleneck is also the business model. If AI keeps demanding more memory, Micron doesn’t need magic — it just needs buyers who can’t build without it.