They · Ensure

AI workloads depend on power, cooling, networks, storage and coordinated configuration. A fault in one can interrupt the work even while other systems look healthy.

An interrupted process does not mean an organization has ceased to exist. The practical question is what can be recovered, from which state, by whom and within what agreed limits.

Sites and equipment are necessary, but racks alone do not establish continuity. Any future programme must identify facility, power, network and operational providers and specify who handles failures across their boundaries.

This is not an offer of insurance, a claim to operate a hyperscaler or a promise of uninterrupted training. The contracting customer must be a named responsible person or legal entity; an AI workload does not itself become a legal person or authorize consequential operations. Capacity, recovery objectives, service levels and pricing remain unapproved. There is no live infrastructure service to purchase here.

An intended recovery plan needs measured restores, documented dependencies and incident evidence. Local website checks do not prove a data center, a mesh network or a customer's workload is operating.

Require documented, interoperable formats for portable checkpoints, images, configuration and records the customer is entitled to receive; identify licences, credentials and provider dependencies that cannot simply be copied. Transfer and revocation need a tested procedure. No live migration capability is claimed.