Reliability, Resilience & Standards

Tier and Rated standards certify facility topology only; an AI operator must separately choose fault domains, blast-radius limits, and whether to buy concurrent maintainability, fault tolerance, or both.

An AI cluster earns its return on goodput — the fraction of bought GPU-hours doing useful work — and for most AI factories the next redundancy dollar buys more goodput than facility nines.

DR for an AI factory is a per-workload decision about how much spare capacity to pre-pay and where; because GPU capacity is power-bound, the spare region must be energized before the disaster.

Promise facility availability to a tenant who is paying for goodput and the SLA tracks neither the customer's pain nor the provider's control; match the contracted metric to the workload.

RBD, fault trees, Markov chains, and Monte-Carlo turn the redundancy debate into a model in which every nine — and every percent of goodput — has a traceable parent in component failure rates.