But their virtual offerings are much less reliable than a standalone system by a lot (they guarantee to refund you the 25cents for your instance if it goes down, not the value of the service interruption or its cost. lol! Read that TOS)

What's the solution to their inherent unreliability? Redundancy at more cost.

Well, hey, you can rent two colocation facilities if you really need redundancy across geographic regions. And maybe you can just use your colo as a source for a CDN that is geographically diverse (for latency, not hurricanes).

Geographic diversity and HA is beyond most people? If you're the kind of business that needs that, you can hire the exact same people that Amazon Pip'd and fired because they didn't hit some arbitrary ticket metrics, to scale your business.

e.g. https://www.forbes.com/sites/lucianapaulise/2022/10/27/amazo...