Hodgman, it's not actually compression like going from Bitmap to Jpeg, where it's obviously a lossy format change. It's more along the lines of data compaction. Euclideon uses a proprietary format called .UDS. It's a zip file where they compact all points into a readable format. These compaction ratios are roughly 17% of original file size, which they've actually stated. And depending on the actual size of the asset, compaction could reduce that number down to only 5% of it's original size.
Also, what you were describing is more along the lines of "Unlimited Assets". Not "Unlimited Detail" in regards to those assets. They never stated they had Unlimited Storage, but art assets themselves will have ZERO limitation.
I'll explain it like this...
..Open Maya and load a primitive cube. Then load a second primitive cube, but this time, subdivide the cube until it's 10M Polygons. Euclideon has stated years ago, that their conversion technology converts at a rate of 64 Points Per Cubic MM.
Those two primitives are of IDENTICAL SIZE in actual 3D Space. So regardless of Polygon Count, when either asset is converted, they will convert to the IDENTICAL AMOUNT of Point Cloud. Euclideon is converting based on a measurement. A physical number associated to the proportions and dimensions of the asset. Not Each Polygon Individually.
The amount of polygons has no bearing on file size, nor does the size of the texture. A 256x256 texture would wrap around UVs the same as a 4096x4096 texture. When Euclideon converts, each point is given color attributes, which some have said is "vertex color" built for their own shader system. So Polygon Counts and Textures are irrelevant since it's technically an offline workflow for asset creation.
So again, the term "Unlimited Detail" in regards to art assets is actually correct and in no way an exaggeration.
Also, Unlimited Power does not mean Unlimited Hardware. There is no amount of geometry on screen that would slow the system down. If the technology renders 1 Point for Every Pixel, then regardless of how many assets are on screen, it would always render.
There wouldn't be a draw call count, because it's only searching for 2.0736M Points if you're rendering for 1080p. If you zoom in on an object, it may take up more pixels, but Euclideon's tech would still be searching for exactly 2.0736M pixels. Whether you're looking at one massive mountain taking up the whole screen or 1,000 trees in a forest, the exact same number of points will always be displayed.
The ability to bypass constraints due to polygon counts, texture counts, draw call counts..etc.. is not an exaggeration. It's the same reason they can "technically" load a 2TB file set in less than a second. When the data-set loads, they are only looking for those initial 2.0736M Points that are based on your cameras current position. It's not trying to load the entire file, it's only looking for those points that need to be displayed at that exact frame when the application launched.
Sorry for typing so much, couldn't find an easier way to be more concise about the topics I talked about.