When is vertical scaling the better answer, and what makes horizontal inevitable eventually?

What is being tested

Whether you recognise that vertical scaling is under-used and that availability, not capacity, is what eventually forces horizontal.

When vertical is better

When it is simply enough. A larger machine is a configuration change; horizontal scaling of a stateful component is an architecture change with cross-node coordination, partitioning and consistency consequences.

Modern instances are very large, and a great many workloads that "need to scale out" fit comfortably on one machine with room to grow.

When state is expensive to partition. A workload with a rich relational model and cross-entity transactions pays enormously to distribute, and the price is paid in every future feature.

When licensing inverts the economics. A per-core database licence can make horizontal scaling more expensive than vertical, which is a commercial constraint producing a technical decision — and it should be recognised as such rather than rationalised.

When the team is small. Operating one large instance is dramatically simpler than operating a distributed system.

What makes horizontal inevitable

Availability, before capacity. A single machine is a single failure domain. However large it is, it can fail, be rebooted for patching, or sit in a zone that goes down.

That is what forces horizontal even for workloads whose capacity fits vertically — and it means the first horizontal step is usually about redundancy rather than throughput, which changes what you optimise for.

Then, eventually: the largest available machine is a ceiling.

The order that works

Scale vertically first, further than fashion suggests. Then add redundancy horizontally for availability. Then partition only when a measured constraint demands it — and choose the partition key as though you cannot change it, because in practice you cannot.

What a strong answer adds

That the two are not exclusive. The common mature shape is a small number of large instances: horizontal enough for availability and rolling deployment, vertical enough to avoid the coordination cost of a large fleet.