That's a reason, but it's far from the only one. The way I've seen DI used over the years is far more about separating out concerns in the code via interfaces, or making things testable without side effects, than it is making those things dynamically pluggable / controllable via config, not code. The majority of interfaces passed as dependencies into constructors have a single real implementation (outside of unit tests).
I lately err on the side of explicit manual wiring as the cost saving of doing it all automatically does not usually offset the costs of extra magic and lack of compile-time safety that it incurs. Exceptions are where there is a lot of AoP interceptor stuff to wire in, or if I want something to help with lifecycle scoping (e.g. per-request lifetime for a set of related things), at which point a DI framework will likely start paying for itself.