There's a ton of infrastructure to build and maintain at Twitter. Yesterday's solutions don't scale and need replacing. Yesterday's "just get it done, we'll worry about cost (or quality, or operational burden) later" have happened and it's time to fix. Lather, rinse, repeat.

This post should give you an idea of scale at Twitter. When I worked there, I spent about 2 months focused just on creating software to help automate the Clos migration mentioned. And there are just tons of things like this that are constantly being worked on.

https://blog.twitter.com/engineering/en_us/topics/infrastruc...