Scaling from 0 to 1M users: what architecture changes were necessary?

i'm curious to hear real stories about the architectural pivots made when scaling systems from zero to a million users. everyone talks about it, but what were the actual, concrete changes? for us, the first big bottleneck was definitely the single postgres instance; caching helped for a bit, but eventually, we had to shard. what were your initial caching strategies and when did they start to fall apart? also, what were some of the early architectural decisions that came back to haunt you? maybe something that seemed fine at 10k users but completely broke at 100k? things like relying too heavily on a synchronous request-response flow, or a monolithic service that became impossible to scale independently. i'm particularly interested in the point where you decided to move beyond a single database, and what that journey looked like.