> The whole idea of the internet (and even some of our infra, like suburbs or highways/rail) is that there's no one single point of failure. Like designed-to-survive-nuclear-war redundant.

Sure, the routing algorithms can quickly adapt to changes in network topology, but they assume infinite bandwidth, which hasn't been the case since a long time now.

In other words, if a couple of important pipes disappear between tier1 peers, alternate routes will certainly have trouble handling all the new traffic, which would make everything grind to a halt, and will only be solved by pissed network admins null-routing that additional load.