I still think the memory name all will 2-3x from here comfortably.
Models are rapidly commoditizing, which will only increase demand. Inference isn't really a compute issue it's a memory issue. The timeline to bring on more capacity is years. The fact that so many lab companies are getting to opus 4.6+ level models so quickly is hugely bullish for demand.
So many people talking about cycle tops without actually looking at where the cycle is going.
Edit: perfect response here and what I needed. I've been inversing reddit for a longgg time. backing up and loading up here