A proxy only routes traffic while the cpu cycles are happening elsewhere, allegedly, the data center that went offline.
Your argument only makes sense if those proxies are more than just routing network packets but also hosting the actual code of those bots. Oh, and you'd also need to assume there's no c&c, the bots should be autonomous. Else there's the single point of failure. If there's a c&c then they can all go offline with a strike.
Don't worry about vpn failures. I have used headscale for a long uninterrupted time. Yes, eventually you may need physical access. That event is as frequent as having a broken boot partition :p be careful to read changelogs when applying updates or have some other means to access of you will... Just saying that for me it has worked perfectly fine.