Doesn't have to be catastrophic failure. Most people just assume that such services are architected properly to be HA, but they usually aren't. They also aren't operated as HA usually, so things like a config change are rolled out globally to everything at once, and then everything breaks and they can't roll back, and you have total outage. Very common.

...and of course, they might not be down at all, it might just be (example) a route from the west coast got fucked in a BGP table so the west coast can't access it, and if that's where the majority of their users are that's all the comments you'll see.