A lot will depend on how fast/hard the hardware will commoditize and improve. 2nm H100 descendants available at like $400 would enable a lot of smaller scale AI fauna. If it never costs down, we’ll get a large dinosaur ecology dominated by states and corps

Retraining of open models also drops the cost of small fauna pretty dramatically, orthogonally to hardware changes, so further innovations there may drive a lot too. There's also the benevolent bot-farm approach (ala SETI@home) that might help parts of the fringe keep up.