The engineering effort required to split up a mammoth code base where everything has been written based on those assumptions about the rest of the codebase is making such a split extremely unlikely to succeed. We're not talking about a few millions lines of code, we're talking about billions[1] of lines of code. It would take more than a decade of having thousands of engineers work only on this and not do anything else (like, you know, actually develop features to keep being competitive). I don't see how it would work.
What I could see working tho would be to focus on exporting some of the very core internal libraries[2] and then go on a project by project basis and do the work required to be made to work externally. The latter would be a never ending effort but the end result would be a number of averagely successful projects that spin off Google.
[1] https://www.wired.com/2015/09/google-2-billion-lines-codeand... 2 billion lines of code in 2015, I'd imagine there are at least 3 by now, considering the increase in employee count.