The Memory Supercycle Is Real, But Not Every Name Deserves the Same Multiple

I have been holding memory names since late last year, and the last month has been a stress test. So let me lay out how I am thinking about the storage supercycle right now, because the headline "DRAM and NAND prices doubled" hides a lot of nuance.

Why this cycle is different

Previous memory cycles were driven by phones and PCs, which are cyclical and price sensitive. This one is driven by AI data centre buildouts, and the demand is contracted years out. Micron has effectively sold out its HBM supply through 2026, with purchase orders stretching into 2027 and 2028. That is not a spot market you can crash overnight. Once a customer signs a multi year binding contract, the pricing floor gets a lot more stable than the bulls of 2017 ever had.

The three names I watch

$Micron Tech(MU.US) is the one I hold most. It pulled back roughly 30% from its high, and 45 analysts still carry a strong buy with an average target near USD 1,486. Even if you haircut that heavily, the risk reward from here looks reasonable to me given the HBM order book.

SanDisk is the higher beta play. Goldman doubled its target to USD 2,200 in early July on NAND tightness they think the Street has not fully priced in. I like the thesis but the swings are brutal, so I keep it a smaller position.

SK Hynix is the cautionary tale. It had its worst single day on the Seoul exchange on record after DRAM ASP growth got revised down and HBM4 mass production slipped to Q3. Same supercycle, but the market punished a growth rate that merely slowed.

How I am positioned

I am not adding aggressively at these levels. I hold MU as the core, keep SanDisk small, and I am watching whether the Q3 sequential price increases actually hold, because that is the real tell. The cycle is intact. The mistake would be to assume every memory name deserves the same multiple when the earnings quality is clearly different.

Not financial advice, just how one long term holder is reading it.