A copy of a service or data, the building block of horizontal scaling, high availability, and disaster recovery.
A replica is a copy of a workload, an application replica behind a load balancer, a database read replica, a cross-region replica for DR. Replication is how distributed systems achieve both scale (more replicas absorb more load) and availability (one replica can fail without taking the system down). The hard part is consistency between replicas: synchronous replication is consistent but slow, asynchronous is fast but eventually consistent, and the choice has profound operational implications.
Replica counts are the universal first lever for scaling and availability, more replicas, more capacity, more redundancy. But replicas have hidden costs: replication lag, configuration drift between them, cost-per-replica, and the operational complexity of keeping them in sync. Most 'we need to scale' problems are actually 'we don't understand our replica lag' problems in disguise.
See the part of the platform that handles replica in production.