EMIB shows promise as a superior "chiplet" methodology compared to the passive, or active, interconnect that AMD is currently using. The EMIB bridge between a Xeon + Alteria FPGA is a major demonstration of its practicality.
They're definitely onto something. But Intel CPUs don't really have anything fashionable to connect to yet. Xeon + FPGA is too niche. The GPU project is key: it needs to be useful enough that people actually care when it comes out. Intel probably should split its cores into AMD-style chiplets for better yields, and use their EMIB technology to connect them together into bigger arrays that no one else can do. I dunno how many years Intel is away from doing this (or even if they're attempting to build something like this).