The interesting part isn’t just the magnitude of the hike. It’s the speed of the revision. A 30-point upward adjustment in such a short time period suggests demand acceleration that outpaced internal forecasts.
The driver appears to be wafer reallocation toward HBM production for AI accelerators. HBM consumes roughly 3× the wafer capacity per gigabyte compared to standard DRAM, which effectively compresses supply for generic server and mobile memory.
If that structural shift holds, this may not be a cyclical “memory recovery” but a repricing event tied to AI infrastructure buildout.