You only have 2^10000 if you have bad boundaries and overcoupling of your subsystems. Thus another benefit of unit testing is validating your architecture: bad architectures are inherently difficult to unit test.
>bad boundaries and overcoupling of your subsystems
And that's the overwhelming majority of ways to make money with software. Think of every video game, desktop application, and mobile app that people purchase. What you call "bad boundary" isn't bad at all. The customers just expect the software to do 2^10000 combinations of possible things across N platforms on M hardware and integration with P external services. Imagine if you emailed a company's tech support saying that the CAD software you bought doesn't work with your Nvidia 1060, and they responded "sorry, that makes our sybsystems overcoupled".
And that's the overwhelming majority of ways to make money with software. Think of every video game, desktop application, and mobile app that people purchase. What you call "bad boundary" isn't bad at all. The customers just expect the software to do 2^10000 combinations of possible things across N platforms on M hardware and integration with P external services. Imagine if you emailed a company's tech support saying that the CAD software you bought doesn't work with your Nvidia 1060, and they responded "sorry, that makes our sybsystems overcoupled".