The technology is too useful to be a bubble. It's only going to get more efficient and when we can run very large LLMs from our own hardware at a reasonable cost? It will be used everywhere.