If you own your own hardware, but you can provision a leased dedicated server from many different providers in an hour or three, and still pay far less than for comparable hardware from AWS.

That sounds like a good deal, what providers offer this?

OVH, Hetzner, Scaleway. Lots of smaller providers. Most people who offer dedicated servers do this.

Which is an awesome capability, if you need it.

I suspect that if you broke projects on AWS down by the numbers, the vast majority don't needed it.

There are other benefits to using AWS (and drawbacks) bit "easy scaling" isn't just premature optimisation because if you build something to do something it's never going to do that's not optimisation it's simply waste.

They need it at the beginning to get started quickly, then they don't but don't bother moving out.

Not too different from how many other lines of business get their clients in the door.

But on Hetzner, you can usually get a dedicated server installed and ready tomorrow.

Hetzner is oversold. It's not appropriate for production in the same sense that EC2 obviously is. It's fine for staging though.

Out of curiosity, how is a dedicated server oversold?

I'm not familiar with Hetzner personally, but maybe they mean the uplink? I've found that with some smaller providers, advertising 10Gbit, but you rarely get close to that speed in reality.

Even sooner than that. I bought a dedicated server on auction and it was ready 2h after the payment.

What really needs thousands of hosts nowadays? Even if you have millions of users. Computers are plenty fast now and leveraging that is not any harder if you choose the right stack.

And even if you are building with microservices, most standard servers can handle dozens in a single machine at once. They are all mostly doing network calls with minimal compute. Even better actually if they are in the same host and the network doesn’t get involved.

If you want to, there are simple tools to hook a handful of them as a cluster and/or instantly spawn extra slightly costlier VMs in case of failure or a spike in usage, if a short outage is really a world-ending event, which it isn’t for almost every software system or business. These capabilities have not been exclusive to the major cloud providers for years.

Of course we are generalizing a lot by this point, I’d be happy to discuss specific cases.