I think the main issue is that to replace SQL, you'll need to build a database which you then need some method of interfacing with.
You could expose the underlying structures and operations with some API of course, but so much work has gone into optimizing the high-level CRUD/join operations that you'll force your users to reinvent the wheel.
You're not wrong. Just note that in almost every other area of the stack language and technology transitions occur all the time. SQL is one of the few parts of the stack that has remained steadfast. Many argue for good reason, I would agree, but I would also agree that there is much room for improvement.
What's strange to me is why developers welcome change anywhere else on the stack but vehemently dislike suggestions for improvement in a relational database api.
You could expose the underlying structures and operations with some API of course, but so much work has gone into optimizing the high-level CRUD/join operations that you'll force your users to reinvent the wheel.