As well as being able to utilize significant amounts of memory (terabytes rather than gigabytes), the also tick another feature box demanded by server admins right now: virtualization.

ARM chips have a major limitation on the server, however. Quite simply, they're not x86 (or x86 64-bit, which is de rigueur for most server setups nowadays). ARM chips have their own architecture, which is how they can manage to be so power efficient.

A different architecture shouldn't make much difference; can be easily recompiled for ARM architecture, for example, as can many applications, like the popular Apache Web server or Samba file sharing software. However, ARM on the server is unknown territory and, while theoretically simple, various undiscovered spanners could end up in the works. Most admins would rather tread a well-worn path than strike out on their own, and ideally buy everything as part of a supported package.

Microsoft is also working on porting Windows to the ARM architecture but it seems this is more about creating a consistent experience across desktop and mobile technologies. While it's not hard to imagine how in-house ARM expertise could be applied to Microsoft's server products, that would be pure speculation.

While the software infrastructure for ARM isn't yet present, the server hardware scene is also perhaps in an even more dire state. Nobody is yet selling motherboards that can accommodate four, eight, or more ARM chips, for example (although I'll bet they're in the works). We lack the kind of sophisticated chipsets that can support the data throughput server applications require, too.