People and companies are using Elasticsearch for that sort of scale, and some people use it as a primary datastore. We can't say if you should use it as your primary or not, it depends on many factors and your comfort with the resilience trade-offs that Elasticsearch exposes.
You can find more info about our resilience issues (both solved and open) here: Elasticsearch Resiliency Status | Elastic
It should be noted that while Elasticsearch can handle the scale you mention, it is not a relational database. Expecting Elasticsearch to handle relational aspects (joins, etc) will prove frustrating. There are some relational features (parent/child, nested, the new SQL plugin) but they are operating on a fundamentally different datastore than a relational database, so you should expect full relational functionality.
There are many ways to work around not being a relational database... but often it involves denormalizing, etc. Just something to keep in mind 