Call Siemens and ask what they charge for Parasolid. Hint: you have to ask, so...
Computational Geometry is hard. FOSS CG Kernels are not going to spring from the ether. CGAL exists, but its target audience is CG researchers. OpenCASCADE exists, but it's really a loss-leader for proprietary extensions and consulting contracts.
At a complete wild guess I would say something like $50m.
(Obviously it depends what you mean by "usable" - I use it now, so in a sense it is already usable. I guess you mean to get to the point that businesses would rely on it.)
Computational Geometry is hard. FOSS CG Kernels are not going to spring from the ether. CGAL exists, but its target audience is CG researchers. OpenCASCADE exists, but it's really a loss-leader for proprietary extensions and consulting contracts.