I've been SRE at Google. I want to gather a couple of points.
1) Reliability requires engineering. (See what I did there?) If you don't have in-house expertise on reliability, things are going to crash, it doesn't matter where the things are.
2) Redundancy does not provide reliability. Redundancy is a component of reliability. Application architecture MUST be set up to turn redundancy into reliability.
3) Redundancy is a many-faceted gem. You need regional redundancy. But you also need non-overlapping maintenance schedules. I argue that you need to think about weather patterns. Think about hurricane tracks. Think about arctic blasts. Think about the ring of fire on the Pacific rim. Think about a Iceland volcano dropping ash over 35% of Europe. Think about servers for emergency services being hosted someplace other than where the emergency being services happens. Think about raptors kissing across power lines in electrical substations. (Yes, it happens.)
4) The majic number is two. You need two extras. One to take down on purpose (for maintenance) and one to take down because the moon went out of phase. If two DCs have overlapping maintenance schedules, they can NOT both be counted. If two DCs are in the same region, they can NOT both be counted. Same thing for individual servers in a DC. You need guarantees regarding server maintenance.
5) If you don't have the budget to run an ops team in three different States, and your business can not tolerate major outages, then hire a reliability expert and go into the cloud. Once the numbers are run, you probably want to be in at least six DCs (again, in six States) before it makes sense to go independent.
6) Monitor, monitor, monitor. Lots and LOTS of ways that things can go wrong in the world. If your customers ever are the ones that are telling you that there is a problem, you have a much bigger problem than the one your customers are telling you about.
7) Keep an exit strategy. Actually, keep two. First, be sure that you have an executable plan to change cloud providers. Second, have a plan to build your own. Review annually. If you are an early stage startup, review quarterly.
Yes, there is no free lunch. Probably not even a cheap one. But equipment leasing would not be a thing if it did not make business sense in particular circumstances. Some of those circumstances include IT.