Completely doable, but very much non-standard so if you have no inhouse experience to start with you'll have to spend a few years just building the initial team unless you find a really good headhunter.
In CAD you have to combine at least two non-trivial[0] discipline, computer graphics and and computational geometry. And they are completely unlike 'normal' software fields where you can just plug in various third party components and call it a day. You have to write your stuff to a great extent. And you have to design it - there are not that many public examples of how to put such a thing together.
It's not magic, it's not intrinsically very deep, but it's complex in the sense that the systems that have to interact are far and wide. It needs a lot of work.
And there are lots of things that are not in books, you just have to understand how they go together and which ways to do things are better than others.
[0] Non-trivial as in "I'll just figure it out" vs. "I'll have to pour through a few decades of academia and industry research and technical publications plus scour obscure websites" to get a clue on what the state of the art is and where it could go on from here
I don't really know about the architectural tools. I've used Inventor, Fusion 360, and Maya. Autodesk became a leader in 3D animation software because the animation software companies kept going bust. Silicon Graphics bought Alias and Wavefront and came out with Maya. Then SGI got run over when gamer graphics cards became good enough to make SGI workstations obsolete. (Also, a badly timed sale and leaseback deal with Goldman Sachs for the property that is now the Google campus in Mountain View.) When SGI tanked, Autodesk picked up Maya.
Softimage was also picked up. They had a good 3D animation system, and then they came out with an early pro-grade film and video editing program. Avid, which used to sell overpriced furniture with built-in computers as editing stations, bought Softimage to get control of the editing software. They had no idea what to do with the animation software and sold that to Autodesk.
Then there was Lightscape, which had the first good radiosity renderer, but by itself a renderer wasn't a viable product. So Autodesk bought that.
Autodesk managed to get all those pieces to play well together, and now they're all part of Maya. Autodesk ended up owning the Hollywood-grade animation business.
A side effect of owning the animation business was that the architectural tools had really good rendering. Good enough that you could put all the light fixtures where they would go on the real building, and see how the finished building would look when lit. This was a big win when you needed to sell your design to the customer for the building. That's not so exotic today, but it was ahead of its time 20 years ago.
On the engineering side, getting computational solid geometry to work right is hard. There's free software for that. Try threading a rod and then applying a chamfer on the end. It works in Inventor and Fusion. FreeCAD, not so much. This is an area where the distance between "works for the easy cases" and "just works" is considerable. There's a reason that Autodesk employed a "staff geometer", a geometry theorist. Or play with the 2D constraint system in the Fusion 360 sketch module, where you designate A is tangent to B, C is parallel with D, circle E is tangent to the three circles F, G, and H (that required a new solution to the Problem of Apollonius) and watch it make the drawing consistent.
(Disclaimer: I'm an Autodesk stockholder, but haven't been involved with the company for decades.)
The business model of large AEC companies is a bit different than that of hollywood VFX subcontractors. The projects for one are already expensive and the data itself is a critical part of the deliverables (instead of just a rendering of the data).
The domain knowledge + software engineer skills + computer graphics skills needed are extremely rare.
"It is in this context that a number of practices, who represent a revenue stream for Autodesk of over $22m over the last 5 years"
A $22m over 5 years is not a large deal for the large CAD companies such as Dassault, PTC, Siemens, Nemetschek.
Revit as a software package started as Reflex and then acquired by PTC, PTC decided they couldn't make enough money off it, and basically let former employees leave and start Charles River Software with the Reflex code base. The amount of work required to create Revit, It would not be good return on investment for any company other than Autodesk to sell it. The only reason it makes sense for Autodesk is it keeps the AutoCAD DWG monopoly going.
There is nothing short or easy about writing Revit/BIM software and it couldn't be done with the 5 million dollars that these firms had been spending on ADSK products.