There's pretty clearly a law of conserved latency, that users will gladly suffer x latency for y capability, with capability complexifying until things are too slow, requiring then hw/sw stack upgrades to make it all fast again, while overall capability utilization decreases