"PRIME favors designs that have larger on-chip memory and, as a result, less processing power."
Curiously, a year or so ago some other researchers at Google found a critical threshold in system scale, above which performance improved significantly.
Expect the new round of AI-optimised chips to nail these scale effects good and proper.
Still no nearer to General Intelligence though. One for third-gen AI-optimisation, perhaps?
The self-improving AI cycle has nevertheless begun; make sure the OFF switch is implemented purely in hardware!