At the moment you almost certainly would not want to use them. The production volumes are currently far too small to say whether ARM servers will end up being much cheaper or better in other ways than Intel-based servers.

These particular servers are based on the Cavium ThunderX which has (up to) 48 cores per socket, and usually two sockets per server. Each individual core is rather slow. The servers themselves are not noticably cheaper in price than x86-based servers. As noted above it's very early days.

There are some more interesting servers coming along. In particular Qualcomm have a 48 core ARM server SoC which has great performance per core, and should ship in a few months. (They are already shipping final silicon to partners and a few developers)

What could make the difference would be:

- ARM partners currently ship around 40x cores annually relative to Intel. Of course the vast majority being phone/tablet 32 bit cores, but that kind of volume might mean that future ARM servers become very cheap, or have an incredibly high density of cores.

- Multiple vendors are allowed to design ARM SoCs, so we may find that particular vendors optimize for more niche use cases, and if your use case happens to be that niche, that could be a win for you.

As an example of the last point: Could your application use 16 x ethernet ports and some custom hardware to process SDN dataplane (eg OpenFlow) tables, all on a single System-on-Chip? Good luck getting Intel to design such a chip for you. However an ARM partner would be entirely free to develop such a chip, and maybe hardware like that will form an important part of telecoms/OpenStack cloud deployments in future.