A free, vendor-neutral readiness checklist used to pressure-test an AI infrastructure plan before budget is committed.

Covers the questions that decide whether a deployment succeeds:

• Workload profiling — training vs inference, batch vs real-time, model sizes and context lengths
• GPU compute sizing — accelerator class, memory footprint, node count and utilisation assumptions
• Fabric readiness — InfiniBand vs RoCE, oversubscription, rail design and cabling reality
• Storage — throughput per GPU, checkpoint bandwidth, dataset staging and tiering
• Datacenter constraints — rack power density, kW per rack, cooling method, floor loading
• Orchestration — Kubernetes, GPU scheduling, multi-tenancy and quota policy
• Security and sovereignty — data residency, tenant isolation, regulated-workload handling
• Commercial readiness — build vs GPU-as-a-Service, SLA expectations, exit and portability

Use it as a pre-RFP gate. Every unanswered item is a line item you will pay for later.