The Anatomy of a Data Center Build: A Symphony of Organized Chaos

From dirt to live load: what actually happens, who is responsible, and where quality breaks down.

Most conversations about data center construction focus on the same handful of topics: land use, power availability, cooling technology, and chip supply. Those things matter. They are the reasons a project exists and the constraints that decide where it gets built. But they are not the reason a project succeeds or fails once the financing closes and it is time to put steel in the ground.

The reason is the work itself. The on-the-ground, trade-by-trade, connection-by-connection execution that turns an empty pad into a live facility. That part of the story gets very little attention, and it is the part that decides whether a project hits its date.

Building a single data center, let alone a full hyperscale campus, is a symphony of organized chaos. A hyperscale campus is typically five to ten or more buildings, each one drawing 50 to 80 megawatts or more, each with its own power distribution, cooling, fire suppression, building management system, and security. The goal is to go from a pad-ready site to live load in 12 to 18 months. The industry is now trying to compress that to under 12. That compression is exactly where quality gets squeezed.

What the timeline actually looks like

A data center build is not one job. It is six overlapping phases, and the overlap is what makes it hard.

The first months are site work and foundations: earthwork, utilities, and concrete. Then comes the structure and shell, the steel, walls, and roof, where trade coordination starts getting complicated. Next is MEP rough-in, the electrical, mechanical, and plumbing work threaded through tight spaces by multiple trades at once. This is the highest-risk phase for installation quality, because it is where the most work happens in the least room under the most schedule pressure.

After rough-in comes MEP finishes and connections, where tens of thousands of connections get made and are supposed to be verified. In commissioning terms this is L2, installation verification, and it is the phase most likely to get compressed when the schedule slips. Then commissioning itself, L1 through L5 testing, where problems from every earlier phase finally surface. And finally handover and live load, when the facility turns over to operations and revenue starts flowing.

Here is the part that most timelines hide: these phases do not happen cleanly one after another. They overlap. The work overlaps, the risk compounds, and the time pressure moves downstream. When the early phases run long, and they almost always do, the time has to come from somewhere. It comes out of the phases at the end, which happen to be the phases where the most quality-critical work lives.

Who is responsible

A data center is not built by one company. It is built by a chain of parties who each own a piece of the outcome and who all depend on one another.

The developer or owner sets the schedule, owns the risk, and pays for delays. The general contractor manages the subcontractors and owns overall build quality. The electrical contractor is usually the largest and most critical sub, responsible for power from the utility entry all the way to the rack. The mechanical contractor handles cooling, piping, and air handling. A commissioning agent provides independent verification that the systems actually work. And QC inspectors, embedded in each subcontractor’s team, are responsible for documenting quality at every step.

That last role is worth pausing on. QC inspectors carry an enormous share of the risk on a project, and on most jobs they are overworked and under-resourced, armed with paper forms and manual checks. The people responsible for proving the work was done right are often given the least support to do it.

Where quality breaks down

Across thousands of projects, the failures tend to cluster in three places.

The first is the MEP overlap. Multiple trades, compressed timelines, and coordination failures mean connections get made but not all of them get verified. The work looks done. Whether it was done correctly is a separate question, and one that frequently goes unanswered until much later.

The second is the documentation bottleneck. Thousands of QA/QC forms have to be complete before commissioning can start. When that documentation is paper-based, it creates a backlog that delays everything downstream. The building can be physically finished while the paperwork that proves it is finished is still weeks behind.

The third is the schedule squeeze. When early phases slip, the lost time comes out of MEP finishes and commissioning. The phases with the most quality-critical work end up with the least time to do it. This is not a hypothetical. It is the default behavior of a compressed schedule, and it is why projects that looked on track suddenly hit a wall at commissioning.

The numbers behind a single building

It helps to put scale on this. A typical data center building involves:

• 30,000 to 50,000 electrical connections
• Over 3,000 individual QA/QC forms
• 200 to 400 or more workers at peak
• 15 to 25 subcontractors
• More than 3,000 labor hours spent on documentation, most of it still paper-based
• An average commissioning delay of two to four weeks caused by incomplete documentation

That is one building. A campus multiplies all of it. The construction is genuinely complex. The quality verification should not make it harder than it already is.

The one thing to remember

After all of the phases, parties, and numbers, it comes down to a single question. Whether a project gets delivered on time depends on whether every trade did their work correctly, and whether you can prove it with the right documentation.

The projects that go live on time are the ones that verify quality at the point of work instead of discovering problems during commissioning. The difference between those two approaches is the difference between a record created the moment the work happens and a record reconstructed weeks later from memory and paper. One is proof. The other is hope.

Build it right the first time. Prove it with data.

This is the first in a series of explainers on how data centers actually get built. At Cumulus, we work alongside the developers, contractors, and trades who deliver these projects, helping them capture and verify quality at the point of work. Learn more at cumulusquality.com.