Horizontal vs vertical scaling: when is vertical scaling actually the right choice?

early on, at around 5k requests per second, we had a classic horizontal vs vertical scaling debate for one of our services. the horizontal path involved spinning up more ec2 instances, say going from 2 to 8 medium-sized machines. the vertical path was upgrading our existing 2 instances to much larger, more powerful ones. we ended up going vertical for a while, just upgrading our existing instances to larger sizes. it was simpler, faster to implement, and often cheaper in the short term because we didn't have to rethink our service discovery or load balancing much. i often hear 'always scale horizontally', but when is vertical scaling actually the pragmatic, even optimal, choice? are there specific workload patterns or stages of growth where it makes more sense to just throw more cpu/memory at a few machines instead of spreading the load across many?