The problem is, if you cannot afford to maintain it, how could you afford to both build AND maintain your own version of it?
Building and maintainance cost are not linear, especially when you inherit legacy code. The AOSP codebase isn't great, is 4x bigger than the Linux Kernel, and full of "Ship now, patch later" mess.
But I agree that it is a significant endeavor. But the OSS community succeeded in similar projects before, and the current state of the Linux desktop makes me hopeful.
I don't think it's true, but ...
"Google built Android to be impossible to maintain without them."
Could be a very genuine answer to that question. Do you really need all of Android? What if you can build a very similar thing at a fraction of the size.