As I understand it, unfortunately most of the memory that has already started manufacturing or been sold will not be usable, since it is specifically packaged HBM for AI servers/systems.

So once the crash happens, that is presumably the beginning of a longer re-supply ramp-up.