If you spend 1000W on 1 gigaflops, but could have gotten 1 teraflops instead on newer hardware, you are mostly just throwing money away. Unless fab capacity is severely limited in the future or energy becomes too cheap to meter, the opportunity cost is just too great for your statement to be true.