Combined revenue at the five largest NAND suppliers rose 77% quarter on quarter, to $68.87 billion. Samsung kept the top spot, ahead of SK Hynix and Micron.
In the same stretch, Micron shares fell 4.23% in a session and SK Hynix sank alongside them.
A 77% revenue increase in three months is not a normal number. It is what happens when a commodity is genuinely short.
The market
| Metric | Value |
|---|---|
| Top-five NAND revenue, quarter | $68.87B |
| Growth, quarter on quarter | +77% |
| Leader | Samsung Electronics |
| Second, third | SK Hynix, Micron |
| Micron shares, session | −4.23% |
Why good numbers get sold
Memory is not a technology story, it is a spread. Suppliers earn the gap between what a bit costs to make and what someone will pay for it, and that gap is set entirely by how tight supply is at that moment.
Which means the market does not value memory companies on trailing revenue. It values them on where the cycle is going. And a 77% quarterly jump, to most cycle-aware investors, reads as evidence that the peak is nearer — not that the trend continues.
This is the opposite of how a software business is valued, and it is why memory shares regularly fall on their best quarters and rally on their worst. Anyone applying growth-stock logic to Micron will be wrong at both ends.
What is actually short
Two demands are hitting supply at once.
Training and inference infrastructure needs high-bandwidth memory in volumes that were not planned for three years ago, and the capacity that makes it competes with the capacity that makes everything else.
Enterprise storage is absorbing the rest — the data that models are trained on has to sit somewhere, and it is not sitting on spinning disks.
Neither is a fad. Both are also fully capable of being over-served by 2028, because that is what happens every time.
The variable that decides the next eighteen months
Capital discipline among the top three.
The historical pattern is unambiguous: prices spike, all three suppliers announce expansion within a few months of each other, and the new capacity lands simultaneously about eighteen months later, killing the price. It has happened enough times that it is close to a law.
If Samsung, SK Hynix and Micron hold capacity growth below demand growth this cycle, the spread persists longer than anyone expects and the current share prices are wrong. If they do what they have always done, the shares are right and 2028 arrives with a glut.
Their capex announcements are the leading indicator, and they matter more than any quarterly revenue figure.
How I read it
I would rather own memory than accelerators here, and the reason is narrow: memory has a visible price and a visible shortage, and its shares are already down. Accelerator names are carrying both a capex-scepticism discount and a discount-rate hit while their pricing power remains an assertion.
That is a preference, not a conviction. Memory is the most brutally cyclical business in the sector, and the entry is only good while supply stays behind. The moment the top three announce expansion together, the trade is over — regardless of what revenue is doing that quarter.