You're quite right - spending significant effort on scaling when you're building an MVP is probably wrong. The point is that we got our first 10,000 users without any scaling effort.

When writing this post we talked a lot about this, and the wording is quite carefully chosen: be ready to scale. Having a contingency plan in case you need to scale, however, is a good idea if it doesn't cost too much (in either money or distraction).

The observation we make in the post is that cloud hosting like Heroku or Google App Engine makes it much more likely that you can be ready to scale without any substantial effort early on.

Obviously the "cloud scales easily" line only gets you so far. For example, while the web tier scales easily, if you built on a relational database that will impose a limit on your magical cloud scaling powers. Once you hit that limit, you still have to face up to the fact that scaling takes care and effort. But cloud hosting lets you delay that effort until you're at a size that warrants thinking about it.

Maybe we should have made this point more directly in the post: be ready to scale, because (for many applications) it no longer costs you very much.