Earlier quoted context omitted.
Copyright law doesn't say anything about users of software, it talks about copying , performing (no, you are not "performing" nginx for your remote users, that's not how it works), etc. In fact, courts have repeatedly ruled that the output of software (i.e. the packets nginx sends over the network, but which aren't its own code) is not covered by the software's copyright itself. If I lend you my phone, in the absence…
You seem to have trouble understanding that copyright is irrelevant to determining who the users of the software are. The users are the ones using the software. That there is a network between them and the software they're using is irrelevant. Does it make sense now?
Yet, the copyright law is what the AGPL is trying to wield as its basis.
It does not seem valid. Has anyone tested it in court?
I don't think that copyright can assert use restrictions. You cannot say that someone is not authorized to have a copy of something if they do not use it in such and such a way.
It's exactly like a "license" written in the flap of a book which says that you must read only the even-numbered pages. If you read any odd-numbered page, then you are breaching the license under which you're permitted to have a copy, and must destroy the book.
I would cheerfully deploy a modified AGPL program and not reveal the code to anyone. Nothing would happen at all.
I'm surprised Google are so allergic to the AGPL. If it came down to it, they could crush this bullshit in court like swatting a mosquito.