I think as a rule of thumb, systems that require redundancy should scale that redundancy ~logarithmically, rather than linearly. So, the scale of the average individual production site increases at the same time as number of individual production sites to make up for demand + safety margin.