The “at a loss” scenario comes from (1) training costs and (2) companies selling tokens below market to get market share. Neither of those imply that people won’t run models in future. Training new frontier-class models could potentially become an issue, but even that seems unlikely given what these models are capable of.
I have access to quite a few models, and I use them here and there. They are sort of useful, sometimes. But I don't pay directly for any of them. Honestly, I wouldn't.
But then, without this huge financial and tech bubble that's driven by these huge companies:
1/ will those models evolve, or new models appear, for a fraction of the cost of building them today?
2/ will GPU (or their replacement) also cost a fraction of what they cost today, so that they are still integrated in end-user processors, so that those model can run efficiently?
These people won't sit still and models will keep getting better as well as cheaper to run.