AI Readiness and GPU Colocation Checker
Assess what your GPU or AI workload needs from a colocation facility and what to look for when evaluating providers. Cooling, power density and certifications explained.
Your workload
AI readiness assessment
Assessment: 20.0 kW/rack
Standard air cooling sufficient
Facility requirements
- ✓Standard raised-floor or overhead air cooling is generally adequate below 30 kW per rack
- ✓Hot aisle containment recommended for optimised efficiency
- ✓High-density power feeds: 32A or 63A per rack (single or dual)
- ✓Dual A+B power for GPU workloads (minimum N+1 UPS)
Questions to ask operators
- ►Confirm maximum power density per rack in writing
- ►Verify per-rack power capacity matches your deployment plan
- ►Ask whether the facility holds NVIDIA DGX Ready certification
- ►Confirm diverse fibre entry paths for AI training data movement
- ►Request reference from existing GPU or AI tenant if available
- ►Verify network capacity for AI training traffic (100G+ uplinks)
Are you a data centre operator? See how Cagelab can help you rank for AI colocation search terms.
Understanding AI colocation requirements
AI and GPU workloads are fundamentally different from standard server deployments in their power density requirements. Where a typical enterprise server rack runs at 5 to 15 kW, a rack of NVIDIA H100 or similar GPU servers can draw 30 to 80 kW or more. This has implications for facility power, cooling and physical infrastructure that standard colocation facilities may not be able to accommodate.
The critical threshold is approximately 30 kW per rack. Below this, standard raised-floor air cooling with in-row supplemental cooling is generally adequate. Between 30 and 40 kW per rack, the facility needs to assess its cooling capacity carefully, and some standard facilities will struggle. Above 40 kW per rack, liquid cooling, direct-to-chip cooling, or rear-door heat exchangers are typically required.
When evaluating facilities for AI workloads, ask operators directly about their maximum sustainable power density per rack, their cooling technology (CRAC, in-row, rear-door, direct-to-chip), their PUE at high density, and whether they hold NVIDIA DGX Ready certification. The last point is not mandatory but indicates the facility has been assessed for GPU workload suitability.
See the AI colocation UK guide and the high density colocation guide for more detail on what to look for in a facility.