GPU lifecycle is 1-3 years because GPUs are becoming obsolete for cutting-edge work (especially re: power use) in that timeframe. This is good news if you'd like to see expanded use of AI. Production walls at fabs will matter little since future silicon dies will be capable of far more per unit area than current ones, so there will be plenty of incentive to upgrade.
I'm just stating facts. I don't care whether it's a good or bad thing. You can theorycraft about future utopias w/ computronium as much as you want but the facts as they stand today are what I stated.