if we could really trace that much bot traffic to 3-4 big data centers, surely we could have already cut it off by filtering a few addresses?

Nope. It is trivial to set up a server to use a lot of IP adresses (think hundreds or thousands), and distribute them anong the bots.

You can lease IP addresses per month cheaply. Brokers supply you with the right ones so you appear to come from the country you claim if that is important. Usually it isn’t, as no one sees the IP except the platform, and they usually don’t care (nor should they really)