I think it's more likely LLMs become a weird niche hobby. You can roll your own at home with ollama on a beefy system with a good GPU and lots of RAM; it'll have terrible latency and won't be as sophisticated as the commercial models, but an open model is fine for tinkering.
