The Data Center Systems That Derail Commissioning (And Why Controls Is the Real Risk)

Controls is where commissioning lives or dies

The controls layer, BMS, EPMS, PLC/SCADA, DCIM, and the integration between them, is the most interdependent and the most consistently underestimated system on any data center build. Mission-critical data centers don’t struggle because of bad equipment. They struggle because control systems aren’t designed for real-world commissioning pressure.

A chiller can be commissioned. A UPS can be commissioned. A generator can be commissioned. Each is a known, bounded discipline with well-understood test scripts.

What cannot be commissioned in isolation is the response sequence when utility power drops, the generator starts, the ATS transfers, the UPS bridges, the chillers ride through on stored cooling, the BMS reads the temperature rise, the EPMS publishes the alarms, and the operator sees a coherent picture on screen. That sequence lives in the controls layer.

This is why Visibuild structures quality management around systems and assets, not just locations, because the commissioning readiness of a controls system cannot be assessed from a room-by-room inspection view.

The five systems with the deepest interdependencies

When you map dependency depth across a typical hyperscale fit-out, five systems sit at the center of the web. Get these wrong and Cx5 becomes a salvage operation.

1. Building Management System (BMS).

The BMS reads temperatures, pressures, flow rates and operating states across every mechanical asset. It is the brain for cooling. But it only works if every sensor is calibrated, every point is mapped, and every sequence of operation is loaded and tested. A BMS commissioned on partial points will appear to work in normal conditions and fail silently when load steps. Point-to-point verification is a Cx3 activity, and it takes far longer than most programmes allow for.

2. Electrical Power Monitoring System (EPMS).

The EPMS monitors switchgear, transformers, UPS, generators, PDUs and branch circuits. EPMS is increasingly the primary system for mission-critical electrical monitoring because the time resolution and fault-capture capability is fundamentally different from BMS. EPMS commissioning depends on every breaker being physically labelled correctly, every CT and PT installed in the right phase orientation, and every communication address loaded. A single mis-wired CT can invalidate weeks of testing.

3. Generator and ATS controls.

Generators and Automatic Transfer Switches have to perform on the worst day of the facility’s life. Their controls are independently complex but only matter in concert. A generator that starts in 8 seconds is useless if the ATS doesn’t transfer cleanly, and a clean ATS transfer is useless if the UPS doesn’t bridge the gap. The handshake between these three is tested at Cx4 per system and proven at Cx5 across the chain.

4. Cooling controls (chillers, CRAC/CRAH, valves, VFDs).

Cooling is not a system , it is a chain. Chillers, primary and secondary pumps, cooling towers or dry coolers, headers, valves, CRAH units, and the control logic that sequences them all. Quality issues are particularly prevalent in cooling systems, and the failure mode is rarely the chiller itself. It is the sequencing logic that controls how it responds when something else in the chain changes state.

5. Fire detection, suppression and EPO interlocks.

The most underestimated system on this list. Fire detection has to talk to the BMS (to shut down air movement), to the EPMS (to drop power in zones), to access control (to release doors), to the suppression system (to trigger pre-action), and to the Emergency Power Off circuit. Fire systems are almost always the last thing on the Cx5 schedule and the most exposed to slip.

Managing hold points and inspection sign-offs across all five of these systems simultaneously, with different trades, different Cx stages, and different sub-system readiness, is exactly the coordination challenge that structured digital QA is built to solve.

The asymmetry trap: systems don’t advance in lockstep

Here is the mental model most General Contractor’s import from commercial construction, and it is wrong for data centers: that all systems progress through commissioning levels together. That the whole project is at Cx2, then at Cx3, then at Cx4.

In reality, systems progress at radically different rates. Some have to reach Cx4 or even Cx5 before others can attempt Cx3. The commissioning programme is not a horizontal bar advancing left to right. It is a staggered ladder, and the rungs are dependencies.

Three examples make this concrete.

1. Permanent power has to be at Cx4 before almost anything else can leave Cx2.

Generators, ATS, UPS, and switchgear need to be functionally tested and producing clean, stable power before any other system can be energised for its own Cx3 activities. The electrical package runs 6-10 weeks ahead of every mechanical system on the commissioning programme.

2. The BMS has to be at Cx4 before chillers and CRAH units can complete Cx3.

Valve actuator stroke tests, pump curve verification, pressure setpoint validation, all of it is controlled through the BMS. So the BMS has to be substantially functional, with points mapped and sequences loaded, while the chillers it controls are still in Cx2. The controls contractor is doing Cx4 work while the mechanical contractor is still doing Cx2 work, on the same system.

3. Fire detection has to be at Cx5 readiness before final IST scenarios can run.

Cx5 integrated testing simulates failure modes including loss of power, loss of cooling, and zone-level fire events. If fire is still in Cx4 when the rest of the facility is ready for Cx5, the entire IST schedule waits.

This asymmetry is what makes commissioning unforgiving. Every gate that opens earlier than its dependencies allow creates rework. Every gate that opens later creates schedule slip. The PM tracking a single “commissioning percent complete” number across the project is missing the only metric that matters: which specific system is the current bottleneck, and what is downstream of it.

The right mental model is a critical-path network, not a Gantt bar. The General Contractors who deliver DC projects on time map this network at tender, not in the final quarter.

How that network translates into a programme is covered in our article “Why Data Center Commissioning Programmes Have to Work Backwards”.

For broader context on how Visibuild supports data center construction quality across multi-system, multi-contractor builds, see what teams building at hyperscale are doing differently. Read also: What Actually Causes Commissioning Blowouts on Data Centers and Why Data Centers Aren’t Big Commercial Builds.