If it's apart of my service, I want it on my servers.
Very few startup start with their own server rooms, because part of the "failing fast" is failing cheap. And if you are making infrastructure investments before you find product market fit - well you aren't failing cheap.
I do think there is inherent value in controlling your whole stack, and running all things on your own servers, but I think this is a bit of a luxury most young companies can't afford.
Is this product designed specifically for startups? Is it intended to be removed later when you have 'found product market fit'? Isn't it more expensive to be locked in to this platform, which you don't have the source to, and can't modify if it doesn't 100% meet your needs?
That's what meant by failing fast. Find a quick way to validate your product in the market, then fail. Don't spend ages working on a product that will never work.
On digitalocean you can get 4G RAM for 40/month, 8G for 80. Running nginx and your favorite backend of choice, you will be able to handle any traffic your startup is getting. If you can't, you are already so successful that paying more won't matter....
I don't think that you own a digital ocean or ec2 instance - you rent them the same way you are renting the service this thread talking about. Given that chrismonsanto, is taking an even more hard line approach to controlling his stack than I do, I'm assuming he agrees (I know - dangerous).
When I talk about my servers I'm talking about things sitting in my racks (possibly in my server room), that I can rip components out of and upgrade. I'm talking about very expensive things, if not in terms of purchase price, certainly in terms of care and maintenance.
--
And to tie it back up to my other comment, I think there is value in owning your own servers, and in coding your stack from top to bottom, and having no external dependencies. But I think these are both very expensive choices - and the kinds of choices, most startups don't have the time/money for.
