Are they? Or are they just realizing that a cluster of redundant, possibly virtualized, machines is just as reliable even if each single machine is not? Two linux boxes with 99% uptime each running the same service redundantly is equivalent to one machine with 99.99% uptime but I bet the linux boxes are cheaper.
Exactly. Hardware and software architectures have changed a lot since 1973. Redundancy that used to be done in one piece of hardware -- "the server" or "the mainframe" -- is now handled by "the cluster". We still have expensive hardware when you look at the servers, network infrastructure, storage infrastructure, clustering and/or virtualization software and monitoring systems. But individually, we can take our pick of vendors for each of these components and that competition is what keeps the costs down.
Our vendors know that they cannot screw us (as, for example, Sun/Oracle does my previous employer) because they will very quickly find themselves with one less customer. There is healthy competition in the marketplace. And we work to avoid vendor lock-in.
We can also identify bottlenecks and selectively upgrade the pieces as needed. The cluster is organic in that regards. Our software runs on the same cluster it did years ago -- but all of the components have been upgraded numerous times, just like the cells in our bodies.