Which costs are you referring to? Are the editors paid? If not the only transfer in costs seems to be hosting a website for downloading static documents?
Previously submission by the researcher and subsequent publicaiton is effectively free; the real cost is paid by subscribers such as university libraries as well as commercial entities.
Under the open-access model the researcher pays the publisher directly for each submission and this can vary from hundreds to thousands of dollars. PLoS one currently charges US$1495 for each publication.
https://www.plos.org/publications/publication-fees/
The exact burden depends on the subject, STEM may be better funded, but this is still not a sum that one can just shrug off. Often grant money also come with strict stipulations on exactly what expenses can be paid with them - publication fee is usually not provided.
In 2002 the required payment was $500, probably about equivalent to current author fees. Presumably it offsets the administrative costs involved in running the journal website and overseeing the peer review process. I don't have first hand knowledge regarding costs of publication or profit margins, but the expense must be paid by someone, generically authors or readers.
I was willing to pay because I regarded wide exposure of the data I was reporting as important, my contribution to the benefit of the community at large. In retrospect, I guess the fact that the article has been viewed several thousand times (and cited by numerous subsequent reports) is a much greater readership than it would have garnered behind a steep paywall.
In both cases it is typically the university which pays. In one model the library pays, in the other it might be an academic department, research office or the library which pays.
Gold open access [the pay to publish model] is distinct from Green open access [where no payment is required].
Many publishers, like Elsevier are adopting a 'hybrid' model where they charge both to publish , and to read.
While there are costs to publishing one they don't justify teh huge fees [and profits] that Elsevier make. One would have to assume that the major, technical infrastructure Elsevier etc implement is to keep their texts inaccessible.
A new OA publisher, the Open Library of Humanities is trying a new model, where neither reader not author pay. Instead the suite of journals is funded by an international consortium of libraries.
Or, some kind of hybrid system where its free to non-affiliated readers but universities are expected to pay subscription on honor system and at fair price rather than Elsevier pricing.