You still need to do enough to buy time if you do need more scalability, since scalability tends to be architectural. Waiting until you hit a wall in production

You still need to do enough to buy time if you do need more scalability, since scalability tends to be architectural. Waiting until you hit a wall in production is usually months too late to start working on it. As a moving target, I often try to test at 10x the current workload, which is usually enough to deal with load spikes and surfaces scalability issues early enough that customers don't see them.