Adding more replicas of a service to handle more load, the cloud-native scaling model versus making a single replica bigger.
Horizontal scaling (scaling out) is the practice of handling more load by adding more instances of a service rather than making one instance bigger (vertical scaling, scaling up). Cloud-native architectures favor horizontal scaling because it's automatable (Kubernetes HPA, AWS Auto Scaling Groups), fault-tolerant (one replica can die without taking the service down), and unbounded by single-machine limits. The tradeoff is that the application has to be stateless or share state externally, sticky sessions break, in-memory caches duplicate.
Vertical scaling has a ceiling: the biggest instance type your cloud sells. Horizontal scaling doesn't, you just keep adding replicas. Designing services to scale horizontally from day one (stateless workers, externalized state, idempotent handlers) is dramatically cheaper than retrofitting it during a capacity emergency.
See the part of the platform that handles horizontal scaling in production.