ArangoDB has no built-in limitations to horizontal scalability. The central resilient Agency will easily sustain hundreds of DB-Servers and Coordinators, and the usual database operations work completely decentrally and do not require assistance of the Agency.
Likewise, the supervision process in the Agency can easily deal with lots of servers, since all its activities are not performance critical.
Obviously, an ArangoDB Cluster is limited by the available resources of CPU, memory, disk and network bandwidth and latency.
Moreover, high numbers of databases, collections, and shards come at a cost. An ArangoDB Enterprise Edition cluster can sustain up to a few thousand databases, and a database can sustain up to a thousand collections, but the total number of shards in the cluster should not go beyond 50,000 or so. Beyond these limits, certain regular cleanup and maintenance operations can take too long for a smooth operational experience. The Community Edition comes without guarantees regarding the database, collection, and shard count.