An Infrastructure Procurement Checklist That Prevents Surprises

The questions worth settling in writing before a purchase order, ordered by how expensive they become to answer late.

Most expensive if left late

Committed power per rack, in writing, and whether that figure is available today or requires the operator to complete work first. Cooling capability at the rack position, and for liquid, whether facility water reaches it, at what inlet temperature, and where responsibility for the loop ends. Floor loading against the fully populated rack weight. These three govern the delivery date more often than the hardware does, and none of them can be compressed once the schedule is running.

Expensive

Certification for the destination market against the specific model being supplied, not the product family. Fabric topology and the oversubscription ratio, since changing it later is a rebuild rather than an addition. Storage read rate measured against what the intended node count actually requires, rather than capacity alone.

Worth settling early, cheaper to fix

Support terms and escalation path, particularly where hardware spans desk-side and datacentre and could otherwise involve two vendors pointing at each other. Spares holding and its location, since parts logistics rather than engineering availability usually determines recovery time. Optics reach measured against the actual cable path rather than floor distance, and spares ordered deliberately.

The question behind all of them

Ask what the quoted lead time assumes, and what happens to it if the facility work runs late. A hardware date and a readiness date are different commitments, and projects slip where the two were assumed to be the same.