It’s interesting stuff for sure.
What this will essentially allow is for a highly integrated system that can be easily scaled. Not only that, but it can also mean multi-die CPU and GPU models in the same system as well.
How this is different from Infinity Fabric is that it takes the concept of multi-die even further and would break the CPU and the GPU down into even more elementary components. Another argument for this is to allow for dramatically higher yields as the fab processes get smaller and as consumers demand ever more powerful hardware.
Exciting, or is it not needed yet?