Maybe I'm misunderstanding. It sounds like you agree that companies with proprietary code can't link GPL libraries, and Google's statements don't suggest any problem for companies which don't care about keeping their code proprietary.
No. This is incorrect. One cannot create derivative works with AGPL code. Linking does not necessarily create a derivative work. Whether or not it does depends on the specific case, and the degree to which the code intertwines. If I have an AGPL back-end and a proprietary front-end for an online photoeditor, that's probably a problem. If I have an AGPL database and a proprietary web app using it, that's generally not a problem.
Even in the photoeditor case, it only becomes a problem at distribution. For an internal tool, AGPL+proprietary is often okay (depending on who the internal team is). AGPL means a consumer-facing web app becomes a problem.
There is a sensible decision path for avoiding linking, but it has nothing to do with this. The arguments are:
1) Engineers aren't lawyers. For a big company, it's easier to have a hard rule (no AGPL) than to train engineers to be lawyers. For small companies, more nuance is okay.
2) AGPL is as much a legal mechanism as a social signaling mechanism. If I'm using a tool not the way the authors intended for me to use -- in an open ecosystem -- that creates its own problems.
The article points out that that is not the case.
If you are using AGPL code as a library, however, then your application as a whole must be under an AGPL-compatible license. It's just that connecting to a server does not constitute linking in the GPL, and the AGPL doesn't change that.
The downside risk is damages. Those usually aren't monumentally huge. There's a calculation, but they're based on how much damage was done:
1) How much did Google profit from the code?
2) How much did the other party lose?
3) Are statutory damages greater?
Pick the highest of the three. If it's intentional -- and in this case it isn't -- you triple it. You might toss in legal fees.
#1 is the relative cost to going with an alternative solution (build in-house, license, etc.). #2 is usually zero for AGPL code. #3 is pocket change for Google. So you're likely to go with the cost of not having gone with AGPL in the first place, twice (once in damages, and once for the migration).