* if the physics is so completely understood that you can confidently predict something will work from your sofa, and give an error-free recipe to build it, you indeed can invent from theory... but how deep can this invention be if the problems of the field are completely solved?
* if you are working in a field at the edge of human understanding, you cannot have the confidence in your ideas without having tested them experimentally; a theoretician makes at most a minor contribution to the actual inventions being realized, because he's producing - most likely somewhat wrong - hypotheses.
This latter kind of "theoretical" inventions are heavily subject to survivorship bias. Fifteen competent theoreticians make different predictions - all according to best, though incomplete, model of the world; a successful experiment validates exactly one of them, and we end up exalting the lucky winner as the "inventor".