Fermi guess: 60 wpm, that's one 5 character word per second, it probably doesn't make sense to give a new prediction more often than once per character -- that's 200ms to chew on autocomplete and linting and fancy animations. Meanwhile, please render the specific character out of your lookup table in 16ms, thank you.

Do modern graphics drivers for X or Wayland hand off font rendering to the GPU? They probably should -- it's maybe 100 or so small textures per font-selection, 150KB or less prerendered with 3-bit alpha, and maybe 100 loaded up into graphics RAM at a time -- 10MB is nothing, really.

they certainly could. Games can do 16ms frames, while calculating full physics simulations.

It's just that text editors of the modern day are programmed by people who prefer to not write it that way - mainly because it's quite hard, and the modern OS doesn't usually fit well into this framework of rendering for multi-tasking. And it takes more effort too.

Much easier to rely on a UI framework which adds overhead. The expectation is that the user probably won't care, and prefer that the software be more feature rich.