It's probably better to have distributed inefficiency than centralized inefficiency. Local fixes for local conditions. You can't get that if your ticket joins a single national queue.

Cloud providers don't merge separate companies' infra just because the same templates are used. There is a reason for providing compositional primitives but also allowing "duplicate" setups.

Every serious lab is going to need a hard charger checking in with #GovOps. Infra utilization and research delivery are orthogonal: world-class utilities can exist alongside experiments that got mired in technicalities (tech debt, outages, and unexpected transitive dependencies).