IF you're referring to the Capital One incident, that had nothing to do with AWS. Their systems behaved as intended. It was a error in the implementation of Capital One's systems.
Your misinterpretation of their comment is the source of your confusion: note that they said “a lot” and “more secure”, not perfect. There are many more breaches of on-premise systems but we don’t say that those are too risky to use — it all comes down to cost. One big advantage that cloud environments have is that you can assume everything is API-driven and there are off the shelf tools to look for common problems like the Capitol One WAF setup. You certainly can do that on-premise but you have more work to do and the bespoke nature of the environment makes misunderstandings easier.
Having trouble finding a reference since the search terms aren't too friendly (lots of targeted ads though), was there one where it wasn't an account configuration issue?
Some of that is on AWS for initially making the defaults too open, but at the end of the day, S3 was doing what it was told.
Is there some case where S3 was locked down, and the data still leaked?