I've tested Aerospike  and other nosql databases when we were looking for a viable alternative to what we had as we were reaching its limit. What I liked in Aerospike is that you can easily confirm the exactitude of the performance claims, which is not the case for most commercial products. Also, in contrast to usual databases, there is no need to define sharding (i.e. how data will be fragmented on the different nodes of the database) or replication parameter (i.e. the number of node on which your data will be replicated so that you don't loose anything in case of node failure). These parameters are essential but as a developer you don't have to care much about them as Aerospike will do.
Also, Aerospike gives you the choice between in-memory storage (e.g. to cache user session data) or SSD disk storage (for durability) without any compromise in performance.
It has a powerful query language, and enable users to define custom aggregation functions based on the Lua programming language  which is very flexible especially when it comes to define business related aggregation algorithms.
Above all that, it is an Open Source solution.
We really liked Aerospike, but as our use case was about Aanalytics that needs running heavy read-only analysis workload on the database. In addition, our use case required one big write workload directly from Apache Spark RDDs (Resilient Distributed Dataset) that may consists of billions of rows and hundreds of properties. These use cases were sadly not the appropriate cases where Aerospike can be used as it's a key-value database and not columnar oriented.
If what you need is a very low-latency access to single data, then you must consider trying Aerospike. If you are on AWS and look for an easily deployable key-value store, then you have to consider Aerospike as it has an officile AMI (Amazon Machine Images) that you can use directly from the marketplace. In contrast, if you're looking for a nosql solution to use as backend for you analytic workloads then Aerospike may not be the appropriate solution, but you can still evaluate it as this may evolve over time.