Even without your own rack or colo, The math with AWS stops working as soon as you no longer fit in the free tier, since providers like Hetzner are 40% cheaper.
S3 is designed for 99.999999999% durability. Hetzner's Volume storage is just replication between 3 different physical servers.
In terms of durability that's a universe apart.
S3 is beyond impressive, but how many workloads truly need that? I’ve never had a single instance of data loss on a NetApp or Pure array.
Truly need, I don't know. But customers will request (and pay for) the 9's.
I suppose it’s like how someone who’s already made up their mind to buy a Lamborghini never questions whether they really need a 800HP engine.
On the other hand you have transient failures in the cloud (at least on Azure - this behavior is even documented) so does that count towards the 99.99999%?