In my experience, there has been an interesting ebb and flow on the use of 'dedicated' hardware and auto-scaling power of AWS/cloud instances. When first starting out, super cost-conscious, you're getting ideal performance. As soon as you start experiencing interesting traffic patterns, auto-scaling makes sense. Then, your bill grows to such a point as it makes economical sense to pull back to more powerful dedicated services, then lean a little into auto-scaling, etc. A pendulum of underlying services and people to support it. A balancing act of finding efficiencies during wild growth periods, and reaching some sense of stability.