You should bring it in-house as soon as it becomes critical for your business. Only startups that aren't profitable yet should bet their skin on so many hosted services as long as they allow them to grow faster (that's not a given, sometimes SaaS/PaaS are complicated to use and unstable).
When you are raking in profits, you'd better use some of them to secure your business and make it independent of such risks. Which is the point the blog post seems to be trying to make.
I don't think that's a viable solution. Some of these are quite complicated and replicating them in-house is not something you just up and do in a short time. That tactic might expose you to significant downtime.
I agree they take a long time to replicate. Which is why it doesn't make any sense to me to suggest that you should build them in house before you get your product out to the world.
It depends, sometimes you really do need the big complicated thing and you don't have time/money to develop it in house.
Sometimes though you can make do with something significantly simpler if you restate the problem slightly and move the pieces around.
You might be able to do the core of the job with a cron job and some command line tools rather than the fancy SaaS platform with graphs and everything.
Building on top of the fancy SaaS APIs and then trying to scale down when you want to move away is a lot more difficult than starting with the simpler approach and then scaling up when you really need that extra functionality.
When you are raking in profits, you'd better use some of them to secure your business and make it independent of such risks. Which is the point the blog post seems to be trying to make.