I can assure you that without the facilities of AWS, which saved us a ton of time and development overhead early on, there would have been no Everpix. Period. Actually, the very first Everpix was on Google App Engine, but that's a different story.
As an early stage startup, whose core business is not building infrastructure, do you really want to be in the business of managing for instance your own S3 type storage with the same performance and reliability at 200+ TB scale? We never lost a single user photo or account. Never lost a database server either. When your users trust you with their life photo collection, that does matter.
Then of course, if your company starts gaining traction, your bills increase so things change. As a matter of fact, post Series-A, our plan was to switch out of AWS, concurrently to our infrastructure's redesign to go from managing 100s of million of photos to billions.
Doing the switch earlier would have been premature IMO: there was no one the team with such experience, even less bandwidth. The opportunity cost would have been significant on all the other aspect of the company. Let's not forget the required upfront capital which takes some time to recoup so savings would not kick in from day 1. Finally there's also the fact our infrastructure was still rapidly evolving (type of EC2 machines used, database architecture, etc...), so even buying reserved instances was very difficult as there was a good chance of buying machines we would not need anymore 6 months down the road ultimately wasting money (it did happen).
So what we did instead what focusing on driving continuously driving down AWS costs through various optimizations - if you were to look at our dashboard graphs, you would notice AWS costs were growing slower than users / photos, which reflected that work.
Long story short, the infrastructure was paying for itself through subscription revenues. The real "killer" was payroll. I use quotes because in this type of business, it's expected to have payroll your largest expense - there was no surprise. BTW note that the team was already 1 person = 1 entire product component like iOS or infrastructure so quite lean and highly productive. For a VC backed consumer business, you pretty much need the VC money to cover these fixed costs until you reach profitability.