There’s a generational buildout going on right now, and big tech is buying memory chips the way people panic-bought toilet paper in 2020.
When demand spikes that fast, you get the same thing you got at the supermarket back then: empty shelves.
And when memory goes out of stock, prices shoot up. So much so that Apple just raised prices on its Macs, iPads, and home products, pointing to the “extraordinary surge” in memory and storage costs from AI data centers. Their words: “We have never seen a component price increase this much, this quickly.”
Right now, money is flooding out of some of the most durable, recurring-revenue businesses around and into the chipmakers, who happen to be holding the right commodity at the right time.
So the chipmakers’ shares spike, while the companies actually buying all that compute drift the other way, because nobody’s entirely convinced they’ll earn a decent return on the fortune they’re spending.
Will the demand hold?
Will it become a race to the bottom?
Can they even pass these costs on?
Fair questions. But they skip past the bigger one.
Memory chips are notoriously, violently cyclical.
So cyclical that analysts compare the memory market to the “hog cycle.” When pork prices are high, every farmer breeds more pigs. Two years later the pigs all hit the market at once and prices crash below the cost of production.
Memory does the exact same thing.
We’ve seen this movie before, and not that long ago.
Rewind to 2017 and 2018. The first big cloud buildout was in full swing, and for the first time the hyperscalers became a dominant force in the memory market.
Terrified of getting caught short, they over-ordered, and prices took off with margins fattening right alongside them. By the back half of 2018, the research desks had talked themselves into a comforting story. With the field finally narrowed to three rational players, the argument went, nobody would be reckless enough to overbuild and wreck their own pricing. The cycle had finally grown up.
Then came early 2019. The hyperscalers looked at their warehouses, saw six to nine months of memory they didn’t need, and effectively went on a buyer’s strike.
DRAM prices fell about 60% over the next four quarters. Micron’s revenue dropped from $30.4 billion in fiscal 2018 to $23.4 billion in fiscal 2019, then kept sliding to $21.4 billion by fiscal 2020. And the margins went with it. Micron’s gross margin fell from 59% in fiscal 2018 to 31% in fiscal 2020.
The buyers who crashed that cycle are the same ones inflating this one.
Step back and look at what these businesses really are. They’re commodities. A commodity producer is a price taker, riding demand it doesn’t control. When demand runs ahead of supply, prices climb and margins get fat. Those fat margins pull in fresh capacity, the capacity shows up all at once, and the cycle swings hard in the other direction.
This is the opposite of a steady compounder.
You can see it in Micron’s numbers right now. The company is earning a gross margin of around 84%, the kind of figure you’d normally expect from a business with a serious moat and real pricing power.
The catch is that Micron’s margin has nothing to do with either. It’s running this hot purely because memory is scarce, and scarcity in this industry has a shelf life. New supply always shows up, and when it does, those same margins compress just as quickly as they expanded.
The tide will eventually turn. It always does.
So where does that leave me?
I want to own wonderful businesses for years at a time and let them do the heavy lifting of compounding while I stay out of the way. A commodity business at the peak of its cycle is the opposite of what I’m looking for. Its best quarter tells you almost nothing about where it’ll be in a decade.
That doesn’t mean there’s no money to be made here. Plenty of sharp people are convinced AI has rewired this cycle for good, and they may turn out to be right. If you have a genuine edge in calling the top of a memory cycle and the discipline to walk out before it turns, then this is your moment. I just know that I don’t have that edge, and this isn’t the game I would play.
For the rest of us, the more valuable skill is learning to sit still while everyone else is scrambling for the same shelf.
Because the shelf always gets restocked. Either supply catches up with demand, or the buyers come to their senses and realize they never needed a garage full of toilet paper all at once.
Memory is no different.