Why aren’t DOI reoactively added to older papers?

As far as I can tell, if a paper is too old to have been given a DOI at the time it got released, no DOI will be retroactively given. Why isn’t this done ? I imagine there must be good enough reasons to avoid doing it, despite the benefits of having a DOI (archival/automatisation benefits mostly) I can understand the issue of there just being way too many older papers in existence to assign *all* of them a DOI, but for more popular papers that are still read today, could it not still be worth it?