# HBM demand — X 热门讨论 (2026-09-26 23:23 UTC)
## @EhrmantrautCap_ (Ehrmantraut Capital) · 09-26 11:22 · ♥30 ↻2 💬1 Especially the HBM/DRAM wafer tradeoff notion is interesting, because it hasn't appeared in previous cycles and is very relevant to how the Memory Big 3 can protect their margins.
HBM requires 2.5-3x more wafers per bit than normal DRAM, so when HBM capacity is ramped up, that comes at the expense of DRAM capacity, which remains constrained. The tradeoff will become even worse with HBM4E, which consumes 4x more wafers per bit than DRAM.
So the Memory Big 3 could essentially use the HBM capacity as a valve to control DRAM supply and protect their margins in case demand is plateauing or falling a bit. Obviously, that is not going to happen any time soon, but should it happen in the future, the HBM/DRAM wafer tradeoff may act as an insurance measure.
$MU $SKHY $005930.KS $DRAM > 引用 @EhrmantrautCap_: There's more and more conviction that memory stocks will remain extremely profitable beyond 2028 and perhaps even beyond 2030.
The current valuation still doesn't reflect that consensus. The current valuations of 5-7x forward PEs suggest that after 1-2 years, earnings will collapse atleast 50% and the traditional memory cycle still holds.
But given that we have an unprecedented AI boom, LTAs that are vastly different from past agreements, and the HBM/DRAM wafer tradeoff that leads to DRAM supply remaining constrained if HBM capacity is ramped up... is there any doubt that these forward PEs are just plain wrong?
$MU $SNDK $SKHY $005930.KS $DRAM https://x.com/EhrmantrautCap_/status/2103807523832697167