And where would they get that ?

Is this a serious question?

How many propriatary solutions do you think will hand over the code for them to train, thats the point that i'm trying to make.

Opensource has provided training data, closed source code not so much. They will have a bad time learning.

Um, not all open-source code is GPL. Or even MIT/BSD, for that matter. There's plenty of Creative Commons (CC) code out there to train on. And on top of that, there are reams of public documentation pages that are effectively CC. And more recently, LLMs have shown increased ability things that they weren't explicitly trained on, so if, for example, we can't find a ton of Rust codebases that are open for training, it's not really a big deal.