Right, that's an interesting observation as well. The number of journals is
huge. What this means is that libraries usually do not buy individual subscriptions. Instead, what happens is that they negotiate so-called "Big Deals" with the big traditional publishers. In other words, they pay a large sum of money for a package of subscriptions to a large number of journals from a single publisher. Typically, these packages contain a few of the really big name journals, and a really long tail of journals of varying relevance. Publishers also own many journals which presumably mainly exist to boost those numbers in order to negotiate a higher price, and also has the effect of removing downwards price pressure by decreasing transparency in the market. Theoretically, the per-journal price is a lot lower in such a big deal than individual subscriptions would have been, but it's not quite clear how many of those subscriptions you really need.
Of course, "need" is already a vague term - preferably, all publicly funded research would be available in the first place, but that's not the world we currently live in.
Unfortunately, that's both the reason why adding a subscription journal instead of an Open Access one is bad, as well as why a boycott doesn't affect much change other than that resulting of its publicity. On the scale of those big deals, saying "we don't need that specific ML journal since interesting research hardly gets submitted there" doesn't drive the collective price down that much.
(In fact, I'd wager that those prices are mostly driven by how far librarians managed to stretch their budgets, rather than by the value provided by the specific included journals or the cost spent producing them.)