It's clear that the easy proverbial cream on the top of electronics has been scooped and extracting it more has lead us to skim it down to 5nm but there are many problems that are clearly required for the modern world that we currently solve with electronics despite an inherant inefficiency. Considering that some of the original AI research was done using similar methods it's suprising that it has taken us this long to end up back at analog computing, though the novel approach of photonics is exciting.

The slightly distant world of the future will probably contain large A.I. processing systems that are almost entirely analog and photonic...but even your phone, desktop, and laptop will more quickly have a co-processor on board to quickly handle A.I operations that will be the R&D component of those future systems which even in their infancy will process complex transforms and analog data orders of magnitude faster than their metallic coworkers at equivalent or hopefully fractional power draw.