Comment by simiones

If you invest 100B dollars into compute while asking for 1T, but in reality would need 100T dollars to meaningfully move the needle, you're still significantly underestimating the amount of compute needed, despite not getting as much as you wanted.