To be fair, AWS downtime always make the news because they affect a lot of majors websites, but that doesn't mean an average sysadmin (or devops, whatever) would do better in term of uptime with his own bay and his toys.
But this is part of the problem: we have multiple web properties, and the fact that AWS issues can affect all of them at once is a huge downside. Certainly, if we ran on metal, we would have hardware fail, but failures would be likely to be better-isolated than at Amazon.
1. Calculate the odds that a company with the resources of Amazon will be able to provide you better overall uptime and fault tolerance than you yourself could.
2. Calculate the cost of moving to the Oregon AWS datacenter.
3. Reassure your investors that outsourcing non-core competencies is still the way to go.
> we can actually get into the datacenter to fix it.
But better and faster than Amazon?
I'd rather spend three hours at home saying "Shit. Well, we'll just wait for Amazon to fix that", than dropping my dinner, driving to the datacenter, and spend three hours setting up a new instance and restoring from backup.
Also, What is happen in cloud is stay in cloud because nobody can able reproduce outside of cloud.
(And many other relevant quotes.)