What I don't understand about bot farms is how they don't get IP banned.

If you do buy such a rack, how do people in practice get a rack full of devices to look like they're coming from valid ips that aren't in a VPN or cloud provider's ip range?

IP banning is way more difficult in practice than most people realise due to NAT. How do you distinguish a bot farm that is heavily posting on some trendy topic from a school full of highly engaged kids? They'll both have a ton of traffic coming from a single IP as far as your server is concerned. And if any IP ever does get flagged (which only makes sense for static ones anyways), it is trivial to move to a new one.

Mobile phones end up sharing an external IP address with may other (real human) mobile phones on the same cellular network.

Residential proxies and other such techniques, I imagine.

Most (all?) commercial social networks sell ads. Outrage-farm bots drive activity in two ways: the bot traffic itself and human response to the outrage bait.

I think the sad truth is the bots are good for business.