The same thing mainstream developers use it for, except broader.
Lots of games, movies, etc nowadays make physical models of things, then scan them, or do pose/motion capture, rather than doing everything on computers directly. (Actually, a lot of artists still draw on paper, then use a regular old flatbed scanner to digitize it.)

I'm actually more interested in the pose/motion capture part. Think how they did Golem in Lord of the rings, only scaled down.
If there was, say, an accessible character modeling program (this sounds crazy but it is not; just make Soul Callibor's speak the labels, and it's usable, if hard to do things like position accessories. One would only need one sighted person to make a better list describing the less obvious bits, and that only once), then we can both create and animate things. Even if not, though, I could still do things with character models; I'd just be able to cut the sighted assistance I'd need down to "Design the models I'll be controlling, tell me afterward if anything turned out looking weird", which is a huge improvement over "Do everything, even though I can't possibly pay you enough to get excited about it and I am no Billy Mays".

Add scannable and reskinnable legos, and some degree of set design becomes possible (if only on, say, Mindcraft levels of detail).

But if that's too much, there's always scanning clay and having someone else do the painting. Which is still an improvement over "Do everything the least bit visual for me, which more or less makes you the majority of the work on someone else's project".