Which is better? Having a day of downtime each year, or not launching at all?
Can you specifically give me examples of why using a cloud provider is better for a startup than, for example, using a couple desktops in your garage?
You can't say it's because of backups because the cloud doesn't provide a backup (unless you purchase an extra data backup solution with your cloud provider?). And correct me if i'm wrong, but you still have to set up your development environment on your local computer to write the code, install libraries to test with, etc.
What exactly are the steps involved in "deploying" that you couldn't do on your laptop, or a VPS?
For us (a small three-man team), the full portfolio of AWS services lets us shove responsibility for some of our infrastructure off to Amazon, which saves us loads of time (EC2, ELB, S3, CloudFront, SQS, Route53, SES, and some light DynamoDB in our case). Even with their repeat EBS issues, we've engineered around the common failure points, and do so with a tiny number of self-managed VMs relative to our traffic. Even if we were to go down every few months, our hosted services do better than a one-man devops team could ever do on his own in his garage, or even in a co-lo. Though AWS is not expensive, we gladly pay up in order to focus on our own software. Sure beats hiring another person, in our case.
The common counter-argument is that if we managed it ourselves, that we'd have the ability to resolve outages on our own, because it'd be our responsibility. That is just the thing, we don't want it to be our responsibility, we've got one ops guy (me), and we need to be iterating on our product fast at this point. We'll instead plan for failure, design our architecture to continue operating under some degree of failure, and keep shipping.
Not to say that the "cloud" is a silver bullet. It's not. However, especially from the developer point of view, it lets our entire (tiny) team stay more focused on product development.
Addendum: I've managed servers in my basement, in co-lo, and in the "cloud". Each of these routes is more (or less) appropriate in various cases, but AWS has been a boon to our particular usage case.
For a product prototype, the initial primary goal is "get it online so that we can start validating our assumptions". System administration skills and in-house server administration teams are valuable but not necessary.
Even given that I have a good bit of sysadmin skills, I am needed more as a software developer right now in the early goings. I expect, as you've pointed out, that priorities will change with time and growth. We may even move to bare metal eventually, if we find ourselves needing and able to do so.
Excellent analogy.
> You want a rapid development platform
No, we want low-maintenance infrastructure.
> that doubles as a production system
It is a production system. It does successfully serve many thousands of users for us every day. We've yet to have an outage that wasn't our own fault.
> and costs nothing to maintain
What? I specifically said we're willing to pay more not to have to spend as much time on infrastructure.
It's OK if you're too set in your ways to even attempt to level with alternative points of view, but at least try to read a little more thoroughly. And maybe admit that you're not willing to budge, so nobody wastes time trying to explain an alternative point of view.
Obviously this doesn't happen because it's hard, but also companies have a vested interest in piping all sorts of data through them for analytics purposes. This is not in the interest of the users at all.