[lgbr]

At the current point in time, this is probably limited to ARM-specific workloads (such as if I need to run ARM binaries, or generate ARM binaries without the hassle of cross-compiling).

The long term, however, is where things start to get interesting. I suspect that by the second generation we're going to see ARM servers able to deliver a better price-per-core (and perhaps therefore price for computing performance) than the x86 alternatives. This will be particularly beneficial for applications which require a lot of CPU power and are highly parallelizable. Picture a workload that can take advantage of 64 cores, with ARM you can get this many cores for a lot less cost than x86.

[rbranson]

These instances already deliver better cost efficiency. The a1 instances are within 5% of the performance of similar c5 instances, but are 40% (!!!) cheaper.