Most data center commissioning defects are never found. That is not an exaggeration. The building passes commissioning. It receives its certificate of occupancy. Tenants move in. Years pass. The defect continues operating exactly as it was installed.
Sometimes it is a CRAC unit with a misconfigured supply setpoint. Sometimes it is a UPS module that has never synchronized correctly with its bypass circuit. Sometimes it is an airflow path that pushes hot exhaust toward an intake because a blanking panel was installed wrong in rack 14 of row 7.
These defects do not announce themselves. They do not grow. They just exist. And they compound every single day the facility is operational.
A 2 percent efficiency loss over a ten-year facility life does not cost you what you lost on day one. It costs you ten years of losses, compounding.
Traditional commissioning verifies that equipment powers on, reaches setpoint, and responds to control commands. It is point-in-time testing under conditions that rarely match actual production load.
A CRAC unit that holds 68°F under 40 percent load may not hold 68°F under 85 percent load with the specific airflow patterns created by the actual rack deployment. That interaction is not testable during commissioning. It only emerges under operational conditions.
By the time the facility reaches production density, the commissioning team is gone. Nobody is looking for the subtle cross-system interactions that produce a persistent 2 percent EA gap. Operations teams are managing incidents. They are not running a baseline analysis to find steady-state losses.
A commissioning defect is not a spike. Spikes are incidents. Operations teams find those.
A commissioning defect is a floor. It sets the minimum performance level of some subsystem, and that subsystem never performs below that floor because the defect is always active. The facility adapts around it. Cooling capacity is provisioned to compensate. Power reserves absorb the inefficiency. The defect becomes invisible because the system accommodates it.
The signal you are looking for is a persistent gap between what your infrastructure should be delivering and what it is delivering. That gap is the accumulated cost of every commissioning defect running in parallel.
| Defect Type | Typical EA Impact | Detection Method | Latency to Detect |
|---|---|---|---|
| CRAC setpoint drift | 0.3 – 0.8% | Thermal baseline deviation | 30 – 90 days |
| Airflow path obstruction | 0.5 – 1.2% | Row-level temperature gradient | First dense rack deployment |
| UPS bypass sync failure | 0.1 – 0.3% | Transfer time anomaly under test | Never without active testing |
| PDU branch imbalance | 0.2 – 0.6% | Phase current monitoring | 60 – 120 days |
| Cooling unit capacity miscal. | 0.4 – 1.0% | Δ supply vs. return air vs. load | Requires cross-domain correlation |
Consider a 200 MW facility with a 2 percent total commissioning defect burden. That is 4 MW of capacity that is technically available but not actually delivering compute. At $200 per kW per year in GPU compute value, that is $800K per year.
Over ten years, with conservative 5 percent annual compute value growth, that commissioning defect portfolio costs over $10 million. That is the cost of not finding these defects in year one.
And that is a conservative estimate. It does not account for the secondary costs: the cooling overcapacity built in to compensate, the power reserves held against instability, the operational complexity added by a system running outside its design parameters.
The approach that works is establishing an EA baseline under actual production conditions, then hunting for persistent deviations.
A properly constructed EA baseline tells you what your facility is theoretically capable of given its installed power, cooling, and compute capacity. Any persistent gap between that theoretical maximum and actual output is a candidate for a commissioning defect. The gap will not close on its own. It has to be investigated.
Synestra's 30-day discovery baseline is designed specifically to surface these defects. We correlate telemetry across thermal, power, compute, and network domains to identify where the infrastructure is underperforming its spec. Those findings are delivered at day 30 — before any contract commitment, before any fee.
If we find a commissioning defect that has been running since facility launch, we want you to know about it. The economic case for fixing it is straightforward. The cost of not fixing it is now quantified.
That is the conversation we want to have.