There are plenty of cases where infrastructure can be split into "critical" and "nice to have".

In HPC, it is reasonably common for people to run RHEL on the scheduler, database and login nodes (totalling maybe a dozen servers) but then use CentOS (now Rocky) for the compute nodes, which might be hundreds of nodes. Those customers simply can't justify the cost of RHEL everywhere, even with more recent changes which make it cheaper than it has been historically in this use-case.