This only works while tokens are artificially cheap. The gravy train could come to an end at some point.

The open models that are chasing the frontier labs will stop being open once things slow down and there is less incentive to undercut the front runners. Time will tell if GPU compute gets cheap enough to run stuff locally.