Just my own experience here, not sure how much of that is shared with others. We launched with a duct-tape and baling wire version of an idea cobbled together

Just my own experience here, not sure how much of that is shared with others. We launched with a duct-tape and baling wire version of an idea cobbled together from mostly stock components. It blew up overnight and kept on growing exponentially for months. All that time we managed to extend the life of the first version by throwing money at the problem, buying larger servers and tweaking the bits to remove the bottlenecks whenever they popped up. In the meantime we re-wrote the code until it would scale reasonably well, and this included the re-writing of a bunch of those off the shelf components that had kept things afloat. That re-write took 3 times as many developers as we had had until that time. The final version scaled through 4 orders of magnitude growth (in tests, in real life we never got that far) without any major hickups and I suspect it could have grown quite a bit further had we made all the right decisions. Unfortunately we didn't and that's where we got stuck, around 100K uniques / day.