Skip to content
// Data Centers · 2025

North American colocation operator (anonymized)

Sourced GPUs, HGX servers, and Quantum-2 InfiniBand fabric for a tenant build inside a multi-megawatt colocation hall, coordinating delivery to the facility commissioning date.

// 01Cluster size
2,000+ GPUs
// 02Schedule
14 weeks PO to live
// 03BOM lines
180+
SERVICES · GPU Procurement / Server Procurement / Networking & Infrastructure

The challenge

A North American colocation operator had a tenant signed for a multi-megawatt AI training deployment with a hard go-live date. The tenant needed 2,000+ NVIDIA H200 GPUs in HGX 8-way servers, a Quantum-2 InfiniBand fabric, and the full rack-side supply chain — staged to the data center commissioning crew, not to a generic delivery date.

What Group APZ did

  • Locked GPU allocation across authorized channels and validated lead times in writing before the PO went out.
  • Ran a multi-vendor RFQ across HGX server OEMs, normalized scoring against the tenant's facility constraints, and finalized the build configuration.
  • Scoped the InfiniBand fabric — switches, HCAs, BlueField DPUs, optics, and cabling — as a single coordinated BOM aligned to the rack elevation.
  • Coordinated logistics: kitting, customs, freight, and just-in-time delivery into the data hall against the cabling crew's day-by-day schedule.

Outcome

The cluster came up on schedule. Every line item arrived inside its commissioning window, with serial-level traceability, valid warranty registration, and a clean handoff to the tenant's operations team.

Placeholder narrative — replace once the engagement is approved for public reference.

← ALL CASE STUDIES