I think you got me wrong, the SubDs themselves are mathetacially as clean, simple and smooth as it gets. It is really hard to mess up SubDs. It is the CONVERSION to NURBS that really f**ks things up. Converted SubDs are simply NASTY. Not only are they really heavy NURBS, also they are not able to maintain the perfect curvature of the poly mesh. Converted SubDs tend to “pinch” on starpoints. Even if the polysurface is absolutely pristine in Blender or Modo or whatever, once you convert it the curvature will do funny things. T-Splines are good to get you to where you want to be by 90% to visually ok data that is good for some prototyping. When you get into the nitty gritty surface developement it simply doesn’t offer the necessary control over surface curvature and produces way way overdefined NURBS. And when I say Class A I mean single span. And converted meshes usually have a thousand spans. Think of it of converting a pixel image (that’s the mesh) to a vector graphic (that’s NURBS). The output might look ok, but it is usually overdefined beyond usability.
And yes, car companies use it for concept cars and prototyping because poly modelling is friggin fast. But first they usually stay exclusively in polys and don’t convert, so they avoid all the curvature problems and second for an actual production car the data is useless and they will rebuild it single span… because the final precise control over curvature isn’t there.