It is more or less obligatory to do it this way; if you tried to avoid generalizing to an infinite version, you would have to qualify various arguments with something like "assuming sufficient resources...", which amounts to the same thing.
No finite state machine is strictly Turing-complete, but the category of those that would be, but for the resource limitation, is an extremely useful one.
The "generalizing" process is called idealization. Whenever Turing machines are mentioned, it's an idealized version of a machine that is being referred to.
Every finite physical machine is limited in different way, so it doesn't make much sense to try to lump them together. Whereas every Turing-complete machine is equivalent.
No finite state machine is strictly Turing-complete, but the category of those that would be, but for the resource limitation, is an extremely useful one.