This is such a wild take, I'm not sure where to start. Let's just consider this:

> Barring things like textures and others.

Other than some really expensive stuff, it's not expensive. Who could have guessed?

As a graphics programmer, the most charitable interpretation I have here is that in 2D all the animations are baked into texture atlases. A 3D model would re-use its textures during animation, in 2D it’s a new one.

But I don’t see Factorio as being a VRAM pig, even with all their animations, which you could certainly improve on from the simplistic “every animation frame is a whole new texture” model.