4 pointsby beardyw3 hours ago1 comment
  • springtimesun2 hours ago
    One thing I have been increasingly confused about is why the memory makers are not building or planning to build more fabs. A few months ago I thought it was obvious, that they correctly identified the AI bubble, but after going down the rabbit hole of local inference, it seems to me that memory pressure is only going up and I think that’s disconnected from the frontier labs. Big providers may need GPUs to serve many users, but a single user needs memory and the long tail of the AI explosion seems like it’s going to be local AI built into everything.