Yields Are the Whole ProductWide moat

In commodity DRAM a yield advantage is worth a few points of margin. In HBM it decides whether you have a business.

The arithmetic is unforgiving. A twelve-high stack contains twelve dies that each passed test individually; if the bonding fails, all twelve are scrapped together with the packaging work. A producer running a few points behind on stack yield is not a few points less profitable — it can be losing money at a price where the leader is comfortable, because its cost per shipped stack includes all the stacks that did not ship.

That is the mechanism behind SK hynix's margins — gross margin reached 79.3% in the first quarter of 2026, against 60.4% for 2025¹ — and it is more durable than a design advantage because it is not documented anywhere. It lives in process recipes, equipment tuning and the accumulated experience of having stacked more of these than anyone else.

It is also the reason competitors' HBM announcements have consistently preceded their HBM volumes by longer than expected. Building one is an engineering exercise; building a million profitably is not.

The offsetting truth is that yields improve with volume for everybody, and a competitor shipping in quantity is a competitor learning at the same rate SK hynix once did.

The number that tests this is HBM gross margin against the rest of the DRAM book. If the premium compresses while HBM prices hold, somebody else's yields have caught up.