We're running on five EC2 instances, each instance is running Elasticsearch configured to use 25GB of RAM. With the current data set we might be able to get by with less RAM, we're still in the process of figuring out what works best for us.