It takes more than a good product to take market share, especially from Intel in the Server segment.
And unlike certain claims in the thread, from purely performance side Intel has architecture advantages in Server segment. 28 core monolithic chip without NUMA craziness is better in all workloads. INT IPC, vector FP throughput, inter core comms, you name it, Intel has advantages everywhere.
Unlike Opteron days, where Athlons had IMC and great architecture, while Intel was stuck with FSB and throwing cache at problems, Intel has advantages in performance and architecture overall.
The real worry for Intel is margins and gradual loss of their hegemonic position in lucrative server market. Since Ryzen is MCM, obviously there is huge advatange in yields, AMD is content with selling 32 cores worth 768mm^2 of silicon , while Intel has to build ~694mm^2 massive chip with 28 cores.
So 14nm vs 14nm, obvious choice is Intel, unless you can't get good price out of them*. Or maybe Your workload fits EPYC like glove ( like smallish VM, that can be bound to CCX, or some niche I/O or high mem needs ).
We can fully expect that to change on 10nm. For AMD obviously they will be able to build a chip with larger module/chip and provide more cores per socket at uptuned performance, that will get nearer. For Intel? I have no clue what they will do, but obviously there is nothing to stop them from building MCM chips themselves, it's not like Infinity Fabric is magic dust, Intel can take 4 shrunken HCCs core (or 2 XCC), slap them on chip with HMB and sell those ~56-72 cores ~1000**mm of silicon on 14nm, who knows how much silicon @ 10nm.
In fact they might already be doing that, with rumors surfacing about Intel Cascade Lake-AP “Advanced Processor Family” (BGA 5903). so many pins can surely fit a ton of I/O and power delivery to feed 2 or 4 chip MCM.
* server people are not paying list prices
** before you nitpick, please consider reductions from not needing so many IMC tiles and other savings