Consider just how much more inference capability one can buy per dollar spent today as compared to a year ago. I can get tasks done quickly for essentially no cost that, early last year, might have cost a frontier lab hundreds or thousands of dollars and would have been unavailable to the general public.
In my mind, the bigger unknown is whether actual profitable use cases will show up quickly enough to justify all the (enormous) investment made in the technology.