How do data caps help with that though? People are still going to use most of their internet during peek hours, because...well, those are the times people use the internet the most, by definition.
It's like trying to combat rush hour traffic problems by limiting the number of times people can use a highway. People are still going to use it during rush hour, they just are going to avoid using the highway during non-rush hour periods, which accomplishes nothing.
In Australia the bandwidth caps used to be specifically set up with a different peak and off-peak cap to help with this even more.
In the UK most ISPs will instead perform traffic shaping where they will slow down your connection during peak times after a certain small amount of data transfer. This still seems to be disliked by consumers just the same.
I was so sad when I had to move further north where there's only Cox.
You send 1gbit across the link. Does the power consumption increase appreciably?
The Internet at scale is just that but larger. Perhaps there are 50 switches/routers between you and your destination.
The issue comes about when you have a 10gbit bottleneck and customers want to send 15.
Not sure if at a noticeable level, but I keep wondering: aren't there costs that are directly proportional to the amount of data transferred? Flipping bits costs energy, both at the level of fundamental physics and (more importantly) at the level of engineering - all the electric and optical signals aren't free, and packet processing isn't either. E.g. I'd expect my power bill to be larger if my Raspberry Pi saturated the network by streaming /dev/urandom than it would be if it was sitting idle.
But aside that, I think that beyond the "we can charge you for data because you can", the per-data-transferred costs come from the overprovisioning ISPs do. The fraction of the advertised to actually supported bandwidth per user they can get away with is dependent on average amount of traffic users generate.