First of all I'm not saying that the whole infrastructure could be replicated, only critical parts, and parts that can be easily hosted locally.

Second, pointing to a new machine is as simple as updating IP in your hosts file, or dns server.

Third, you can use vmware or any other virtualization stack to replicate your infrastructure locally. In fact that's the best way to build things - create virtual network, use it for testing, troubleshooting and debugging, and deploy only when everything is working.

All I'm saying is that if you're company is making any kind of money, and your development environment depends 100% on online services, you're doing it wrong.

Not that I'm saying that you're completely wrong, but you're oversimplifying the problem and the solution to it. Your last statement is not necessarily true; you have to balance the cost and how much of a PITA it is to set up and maintain vs how much money would be lost by disgruntled customers for a single rare outage. Granted, it depends on the kind of business you are running, but not every business is so fragile. In fact, I'd say that most(as in >50%) are not that fragile. Even better when you can simply put the blame on someone else, which everyone can do in our case right now.

I'm under the impression that you say self-hosted ~= dev-machine-hosted ?

If that's the case, i think you're misguided : imho, the internet as it was designed was conceived so that everyone has its little self-hosted thing, with dev-machine just for the purpose of, well, test and dev, with the latter goal of it being self-hosted.

Just look how email is technically designed and how it was meant to work, and we use it now, relying mostly on Gmail or Outlook, or worse, using Facebook for emails : we put all our eggs in the same basket.