It's a question of return on investment
Investing several billion in software development to make hardware redundancy redundant (see what I did there?) makes perfect sense if you're Google. Not so much if you're a smaller web property and much less if you run many traditional applications that are highly interoperable and risk sensitive (banking or air traffic control, for instance).
Because software resilience is progressing at a million miles an hour (along with tremendous growth in the skills to set it up and manage it) and hardware resilience has more or less stagnated, however, I expect the proportion of services that operate the way hyperscale ones do today will grow exponentially over time.