Same argument can be somewhat applied to CPUs circa 90s. The growth did stop/stalled in the end.
One difference is that the more compute you have, the more work you can do. You can basically run tokens 24/7 building apps or trying to do research or running a massive model in thinking mode.