My point is that the marginal cost of a hobbyist user, who receives no support, is approximately zero. At no point did I say developing Fusion must have been cheap or easy! What I was trying to say is that those costs are unrelated to the number of hobbyists using the software- Autodesk does not need to hire extra engineers to let me use Fusion. (Support is the big catch here- but I don't expect support from a free tier!).
This is distinct from, for example, a mass-produced bicycle- there, yes, you still have NRE up front (design, hopefully testing, molds/tooling...), but a significant part of the cost of each bicycle is related to the cost of the input materials plus labor. That's not the case with software! With software, practically everything (again, support aside) is a sort of NRE- it's a fixed cost, no matter how many copies you make. Thus, if you somehow definitively knew that a particular customer will not pay for your software, and that their usage of it has benefits for you (greater adoption, the chance to convert them (or their employer) to a paid user later, etc) it would make sense to allow them to use it without paying- the cost to you is ~0 (mostly just bandwidth), but the benefit is non-negligible.
This is not true for a bicycle. If somebody cannot pay for a bicycle, and I am manufacturing bicycles, it generally doesn't make sense to give them a free bicycle unless I'm getting something else out of it that's worth at least as much as the cost to build one additional bicycle - i.e. the cost of parts and labor. But note that the cost to design the bicycle is not part of the calculation there- because it doesn't depend on how many bicycles I build from that design.
The distinction between rival and non-rival goods is very interesting, and I don't blame you at all for not understanding it. It can be counterintuitive!
Does that make more sense? Again, sorry if I was unclear!