It's arguably too complicated for hackers to do well in their spare time, but there are plenty of FOSS projects in similarly complicated niche areas. The most successful ones tend to have companies or consortiums backing them, who aren't looking to make money on the softare itself but have it complement or support their other activities. OpenModelica for one example.
I always wonder why a consortium of big manufacturers (say Ford, Caterpillar, Hitachi, etc) don't just throw their weight behind some FOSS CAD initiative and get something standard that they can rely on forever. I don't think it would cost them more than 3rd party software in the long run.
> in the long run
People have taken the initiative and talked to the engineering departments and bean counters at large engineering houses (brands you're not familiar with, that do the engineering for e.g. Ford, etc but not the design which is typically in-house). The reasons are A) short term cost with no benefit, B) having to train staff, and C) compatibility with extant files. Not only will the current staff have to be retrained but nobody can be hired with experience with the tool.
The problem for these businesses is that their internal CAD libraries are essentially proprietary tooling, they don't have an interest in making it open source, and much of the tooling around those libraries and software packages is proprietary or patented.
It's kind of analogous to Qualcomm's stranglehold over cellular communications. If you don't pay them you're also throwing away decades of IP and tooling around it, and lose interoperability.
The only decent FOSS parametric CAD software I've found is SolveSpace, but it is quite limited, e.g. it doesn't even have a bevel tool, and there's no CAM. It's pretty much the Notepad of CAD.
CAD is just one of those things that is too complicated and niche for FOSS to do well.