That there's still shreds of evidence for transformational improvements to the dreadfully arcane, inefficient, proprietary machines that are currently "the best".

Personally, my favorite is the Deutsch-Josza algorithm which illustrates a task which can be trivially understood, and provides exponential speedup over any classical algorithm that I can fathom to perform the same task.

Consider there was something like a half-century delay between the discovery of the photoelectric effect and the invention of the triode.

What's most striking to me is that these inherently optical phenomena apparently haven't been well-investigated for how they can provide speed-up to graphics processing.