alt Hacker NewsRailroads, like airliners, form a network. There are multiple paths to the same destination. There are also competitors in the form of waterways, air corridors, and highways.
Before the government began regulating the railroads, this is exactly what happened (though no air corridors then!).
Power lines also form a grid, and can route around failures. The power to my house got a lot more reliable when the other side of the neighborhood got connected to the grid.
When the power does go down, so does the cable internet, but the internet still works because the cell towers take over the last mile traffic. There's also Starlink.
One would think it's the same, but it's not. Each leg of a route is unique. Travel time, distance, and stops on the way. They are not interchangeable.
Practical example: a former customer of mine had a specific route carved out for transporting fruit and vegetables from the West Coast to the East Coast. They had exactly one time slot in the day they could take, which brought the travel time down to about 3-5 days. If they missed that time slot, it was up over 10 days. If one of the route segments was out of service, it could have been two weeks or more. Yeah, you could get from point A to point B, but the time made the alternative routes unacceptable.
The power grid is not sufficiently redundant to protect against many forms of damage. Find power substations and count how many 10 kVA lines feed each substation. Then ask yourself, what happens when that substation goes down because a good old boy decided to take potshots at the transformers? When there's a natural disaster like a winter storm here in Boston, there are a lot of power outages, indicating that the grid is not a grid but a lot of branches.
As for the last mile, no, cell towers don't take over for my FiOS fiber when it fails, nor does Starlink. Cell, fiber, cable, and Starlink are not expanding the market for internet access. The cable market, the internet end-user market, is mostly saturated. Users switch but do not duplicate the service.
FWIW, this is another example of you never eliminate single points of failure, you only move them.