I mean on PS4
launch the game Resogun, used GPU compute.
How Resogun uses the PS4's architecture to create 'chaos'
Now imagine a game engine on next gen that was counting on 20% of the GPU power to do GPU compute to simulate more advanced enemies, better building destruction, etc. Great, Scarlett had the power to easily accommodate their demands.
Now, we look at Lockhart. Maybe they have 20% of the GPU power left, after rendering the same graphics at a lower resolution. Well, 20% GPU power of a lower spec part doesn't give them enough "juice" to simulate as many enemies, and the destruction can't be as detailed. Either they simplify the enemies and destruction, or they design levels that can only use the lesser amount of power. Now extend that to whatever additional feature you want (not graphics).
Now as the generation goes further, developers can extract more power from the hardware, but there will always be a hard limit between what's available to do the "other" non-rendering on the screen at some resolution graphics part and the part the happens under the hood that controls what we see on the screen.
This is why this is not good news to me.