There are, of cost, open source licenses that don't allow for commercial use without a license.
Also, there are lots of industries that need guarantees about the software, and even CC0 open source software doesn't come with those guarantees; those come from a commercial use and support contact.
Re: your last paragraph:
I think the future is likely going to be more task-specific, targeted models. I don't have the research handy, but small, targeted LLMs can outperform massive LLMs at a tiny fraction of the compute costs to both train and run the model, and can be run on much more modest hardware to boot.
Like, an LLM that is targeted only at:
The more specific the model, the smaller the LLM can be that can do the targeted task (s) "well".