Memory is supposed to be cyclical. Hynix is supposed to be a beta play on DRAM. Both statements were true ten years ago and are increasingly less true today.
What changed
HBM (high-bandwidth memory) is a structurally different product from commodity DRAM. The qualification cycle with hyperscalers is multi-year, the design wins are sticky, and the capacity is constrained by advanced packaging — not silicon. Hynix has roughly a two-generation lead on the competition and the customer concentration to defend it.
Why the market is mispricing it
Sell-side models still treat HBM revenue as if it carries commodity-DRAM gross margins. Spot checks on order book and ASPs say otherwise. The gap between the right model and the consensus model is, in my view, where the alpha lives.
This is a high-conviction position because the unit economics are knowable and the demand runway is visible through 2028.