Current hardware supports an evolution of the kind of graphics that appeared in the 16 bit days - arguably even the 8 bit days. A completely different approach may not be a bad idea, but there already are alternative approaches - ray-tracing and voxel-based, at least.
"Unlimited detail" sounds to me like the old fractal-compression fallacy. Fractal compression can represent any image, but with compression ratios not so different from JPEG.
At one point, there were claims of massive compression ratios - but the logic for that was the same as for "vector graphics compression". A single rectangle is a very simple shape to represent in vector graphics, giving an impressive compression ratio, but it's a special case that isn't useful for encoding a typical photo. The same applies to the old fractal compression fallacy. A single simple fractal may have a tiny encoding using fractal compression, but that's hardly a miracle of compression.
Another answer has already associated point-clouds with voxels. The only way I can image "unlimited detail" being justified is if there's some repeating-patterns and/or fractal aspect to those point-clouds, and some of those example images seem to suggest that too. Otherwise you'd need infinite data to represent that unlimited detail.
A fractal arguably gives an infinite level of detail. Theoretically, that is - discrete rendering automatically implies a finite cutoff to that detail, and that applies to voxels as well as pixels. However, you can't easily represent anything you want in that fractal form - it may be possible to define arbitrary point-clouds, but you'll need complex point-cloud descriptions to get the forms that aren't so naturally described as a single simple fractal.
None of this means the idea isn't interesting or useful. It's just a kind of "beware of the marketing claims" thing. And just because something looks good now, doesn't mean it won't be "just another tediously obvious point-cloud effect" in five years time.