Many people say: storage has peaked.

But after my research, I found that—

Storage is not a single thing at all.

In the AI era, storage is divided into four layers, and the supply-demand logic of each layer is completely different.

First layer: HBM (GPU's workbench) is directly attached next to the GPU chip, with the fastest speed and highest price. SK Hynix almost has a monopoly, with Samsung and Micron catching up. Current status: severe shortage, AI GPUs have locked up all production capacity.

Second layer: DDR5 (server's preparation area) temporarily stores data before the GPU processes it. Micron is the strongest, with SK Hynix and Samsung competing. Current status: the supply-demand gap is widening, and prices continue to rise.

Third layer: NAND (AI's long-term memory) is where things are actually stored, with large capacity. Samsung, SK Group, Kioxia, Micron, and SNDK are competing. Current status: diverged. The consumer side is weakening, but the AI data center side is still rising.

Fourth layer: Enterprise SSD (the most important form of NAND in the AI era) is the large warehouse for AI data centers, the fastest-growing segment in NAND demand. Kioxia has publicly stated that its full-year 2026 NAND production capacity has been sold out. Current status: AI inference demand continues to lock in supply, leading to a shortage.

Why do some people shout "peak," while others say there's still opportunity?

Because one thing is happening: major storage manufacturers are shifting limited wafer production capacity from consumer electronics to AI data centers.

This brings two types of price increases:

AI side—demand is too strong, can't get it, real shortage. Consumer side—supply is being diverted, leading to passive price increases, but it doesn't mean demand is recovering.

Once AI-side demand slows down and manufacturers shift capacity back, consumer-side prices may fall rapidly.

Those who say storage has peaked are looking at the consumer side. Those who say storage still has opportunity are looking at the AI side.

Both are not wrong, they are just talking about different things.

This is also why when I research storage, I don't just look at SNDK alone,

but need to figure out: among these four layers, which layer has the most certain supply-demand, and which company has the best position in which layer.

The research continues.

Which layer is the storage company you follow mainly in? Let's chat in the comments.

Real-time observation, not investment advice.