At our very large org we do both. At least two clusters per region to isolate platform changes, all hardened to the same standards using automated tooling.