A 180 kW contract approved without a capacity study

Governance failures (root causes)

1. Single-person approval. A 180 kW, high-density deployment is a material change to capacity, thermal profile and contractual risk. It required a change advisory board with facilities, electrical, network, security and commercial representation. Approval by one manager in one day is a control bypass, not a fast track.
2. No capacity model. The hall was managed on a percentage-utilisation figure (78%) that says nothing about the two constraints that matter: power per rack and airflow per rack. A hall can be 78% full and still unable to host a 30 kW rack anywhere.
3. Sales committing engineering capacity. The contract was priced and signed against a design average that the customer's profile violates by a factor of 2.5. No engineering sign-off gate existed between the commercial commitment and the technical feasibility check.
4. No post-implementation review or metering discipline. The service desk has only the contracted figure. Without per-circuit metering and a defined verification step after go-live, the operator cannot detect drift until a tenant complains.
5. No tenant impact assessment. The adjacent row was never modelled, so an SLA breach was created for a third party who did nothing wrong.

Technical failures (symptoms) The CRAH trips, the PUE rise, the 76% breaker and 88% PDU loading, and the 28 °C inlets in the adjacent row are all consequences. Treating them individually - resetting the CRAH, swapping a breaker - would leave the causal chain intact and guarantee a repeat with the next sale. The distinction matters because the corrective action budget belongs mostly in the governance column: a capacity management process, per-rack power and temperature metering, and a mandatory engineering gate in the sales workflow.

Electrical

• Convert contracted load to current with an explicit power factor and diversity assumption: 180 kW at 400 V three-phase and 0.9 pf is roughly 289 A, which on a 400 A plug-in breaker leaves only about 28% headroom before any diversity allowance. Require the design current, the breaker rating, the busway rating, the upstream PDU rating and the UPS module loading to be shown as a chain, with the highest-loaded element identified.
• Check neutral conductor sizing and harmonic profile for the customer's equipment type (switch-mode power supplies, variable-speed drives), plus breaker discrimination with the upstream device so a fault trips locally rather than the PDU.
• Confirm UPS capacity and runtime at the new load, and whether the additional load changes the battery autonomy below the contracted figure.

Thermal

• Per-rack airflow: a 30 kW rack at a 12 K air temperature rise needs roughly 2 m³/s (around 4,200 CFM) - far beyond what a tile designed for a 12 kW rack can deliver. The assessment must state required airflow per rack, available tile flow at that position, and whether local cooling (rear-door heat exchanger or in-row unit) is mandatory.
• Containment integrity, blanking, and the cold-aisle temperature effect on adjacent rows - modelled with CFD or, at minimum, with measured baseline data from the existing sensors.

Operational

• Metering: per-rack or per-row power and temperature monitoring installed before go-live, with a defined M&V period (typically 30 days) and an agreed escalation trigger.
• Reserve margin: define the maximum permitted utilisation of any single element (commonly 80% of rating) and refuse deployments that breach it.
• Documented sign-off by facilities, electrical and the affected tenants' account owners.

Immediate (24–72 hours)

1. Cap and re-shape the load. Ask the customer to spread the 26 kW working load across more racks at lower density, or cap total draw until cooling is remediated. Contractually this needs a documented variation, not a verbal request.
2. Restore local cooling at the new row. Install rear-door heat exchangers or in-row units - perforated-tile changes alone will not deliver 2 m³/s per rack. Verify with a rack-inlet survey afterwards.
3. Relieve the electrical chain. Re-balance the new row across busway circuits so no plug-in breaker exceeds the defined utilisation limit, and confirm discrimination with the PDU breaker.
4. Protect the neighbour. Add temporary supplemental cooling or airflow correction in the adjacent row and confirm inlets return below 27 °C within one week, with logged evidence.
5. Freeze further sales into this hall until the capacity model is rebuilt and the engineering gate is in place.

SLA position

• Acknowledge the breach to the neighbouring tenant proactively, before they escalate, with the measured data rather than an assurance.
• Apply the service credits defined in the SLA without requiring the tenant to claim them; that is both the contractual and the reputational answer.
• Provide a written root-cause summary and the remediation timeline, and offer an independent verification (third-party thermal survey) at the operator's cost.
• Do not attribute the cause to the new customer in external communication. The operator approved the deployment; the liability is the operator's.

Afterwards Add a post-implementation review to the change process: every deployment above an agreed kW threshold is reviewed 30 days after go-live against the predicted load, temperature and PUE, and the variance is recorded. That closes the loop that allowed a contracted figure to stand in for a measured one.