I saw a stream where Vedal was training a model and was wondering if he uses or has considered using AWS to train larger Neuro language models or run inference faster.
#AWS for training?
fresh rose
stiff ridge
idt AWS would not be the cheapest option but cloud training in general is always on the table
like lambda labs like to claim that they're the cheapest
blazing basin
Training on stream is related to game AI work and it's fine on his current hardware (he could use more RAM, but if he's upgrading he probably would gain a lot from buying an entirely new PC and swapping the 4090 into it for running intensive stuff while the old one could handle running the stream portions and 
model to divide load).
But he has indicated on stream and it's hinted in the donate part of the Twitch about page that there are other processing costs involved with Neuro, so it's pretty likely he's already using cloud services for the LLM training.