Comment by isawczuk

It's artificial scarcity. LLM inference will soon be commodity as cloud.

There is a 2-3years still before ASIC LLM inferences will catch up.