The way I think it, it’s possible a really small number of GPUs would be enough to render the framebuffer, you’d just need an army of low-power graphics units to receive the data and render it on screens.
Having a high-power GPU for every screen is definitely a loss unless the render job is distributed really well and there’s also people around to admire the results at the distance where the pixel differences no longer matter. Which is to say, not here.