This whole ARM craziness is stemming from the simple fact:
Somewhere between late 90s and today, PC platform has drastically expanded as non-technical people started using PCs for online activities.
Thing is, these people never actually needed the whole PC. However, the technology of early 2000s simply did not offer anything better - ARMs at that point were simply not powerful enough to run "home PC" tasks.
Then, everything changed with powerful tablets such as iPad - and masses of "ordinary people" are simply moving to these devices, as they never needed anything else to begin with.
So, now everybody is in the panic 'PC is dead' mode - which is simply not true, PC is just going to scale back to its pre-2000s days as many "ordinary people" found a better gadget for their daily does of YouTube and Facebook.
ARM found a great market in this - since ARM was, from the day one, producing chips with extremely low idle-power requirements.
However, I am yet to see ARM competing in high-end desktop or server market. I will believe it when I see it, as all of this hype today has no actual substance, but it is just based on vapourware.
Intel is also not siting idle, and I am sure by the time of Skylake, there will be no significant difference in idle power draw of PC-class Intel CPU and mobile chips from ARM. However, on the HPC/server field, I am not seeing ARM gaining experience that fast to counter Intel's steady improvement in IPC and adding more cores.
Problem with running gazillions of small cores such as ARM on the server is that what you save in CPU TDP, you lose in having to add interconnection fabric. This might work for few applications where individual compute units do not require significant CPU load, but for anything which can use CPU cycles, having a large cache and lots of cores glued together is simply the best approach due to latencies.
This is why we are not seeing too many Atom-based servers - simply, it is not efficient for tasks that need to be broken down then to many CPUs, due to interconnection penalties. I believe I read somewhere that Google experimented with this setup years ago and found that the latencies are simply too high.
So, unless ARM figures out how to avoid this problem OR starts producing cores comparable to Ivy Bridge or Haswell Xeons, I doubt they will be able to gain significant foothold in the server market, except few specialized areas.