From a modelers point of view, this type of technology would be perfect, essentially rendering redundant most of the current workflow, from battling with poly counts to playing tricks with displacement maps and baked occlusion. The model swapping he mentioned is pretty outdated, most engines use dynamic LOD which automatically reduces a models polycount depending on your view distance, also removing parts of a model that are under a certain size.

I'm not sold yet, it's making some incredibly bold claims. I'm aware of point cloud data models from back when it was included in the 3D Mark '01 benchmark (the rotating horse statue, the test was called point sprites http://www.ixbt.com/video/images/3dmark2001/gf3-sprites.jpg ), and found it an oddity at the time, not all that visually appealing and certainly not as a believable modelling method. It's not raytracing, although it does bare a lot of similarities.

I'll be keeping a sceptical eye and an open mind on this one.

Anyway, here's their website http://unlimiteddetailtechnology.com/

Edit: Thinking about it, it's biggest downfall will occur when physics are involved. For a static scene it's ideal, but to get the branches of the trees to blow in the wind, or to achieve any kind of real time environmental change beyond lighting would likely cause some serious difficulties.