Engineering the invisible: why data center sustainability starts before certification

Building a data center is unlike building an office or an apartment complex. The mechanical systems that keep it running are not details to be worked out during construction. They are the core of the project. How carefully they are designed and validated from the start is what determines whether the facility does what it promises—today and ten years from now.

Yet in practice, these decisions are routinely addressed too late. An architect delivers a concept. A contractor takes on the detailed engineering. A certification consultant handles the paperwork. Each party does their part. What often gets missed is the moment when someone needs to ask: will the system we’ve designed actually work under every condition this building will face—at full load, at partial load, when equipment fails, and five years from now when the IT load has grown? What other factors might influence the performance of the setup?

What made it possible to define the performance of the building and the data center services was a holistic, integrated design approach, in which architecture, building systems, energy strategy, and operational resilience were developed in a coordinated manner from the earliest stages, supported by advanced computational thermo-fluid-dynamic simulation and analysis tools. When these elements are considered together from the outset, certification is not an additional layer applied on top of the project, but the coherent outcome of a sound and well-informed engineering process.

This is what we learned working on the ECMWF (European Centre for Medium-Range Weather Forecasts) data center in Bologna.

The ECMWF data center in Bologna offered an unusual constraint from the start. The site was the Manifattura Tabacchi, a former tobacco factory—a protected piece of industrial archaeology with preserved arches and a historical envelope that could not be structurally altered. The solution was to build the data center as a contained unit inside the existing building: a modern, high-performance facility housed within a 19th-century shell.

The facility was designed with the capacity to handle intensive energy use typical for data centers. Cooling was delivered through air cooling units that draw warm air from the data hall, pass it through water-cooled coil exchangers, and push the chilled air into an underfloor plenum, from which it rises through floor grilles to the server racks.

The project followed a design-build approach, with the concept design already established and the executive design developed as part of the project delivery. As the project was pursuing LEED certification, it was also necessary to demonstrate through simulation that the HVAC system could maintain safe operating conditions throughout the data hall, both in normal operation and in failure scenarios.

CFD software builds a three-dimensional simulation of how air moves through a space under different conditions—the same class of tool aeronautical and motorsport engineers use to model airflow. Applied to a data center, it shows where air heats up, where it fails to reach, and what happens when part of the system goes down.

For the ECMWF project, we modeled four scenarios: two under normal operation and two with half the cooling units offline. That second category is where the real engineering value sits. A cooling design can look adequate on paper while quietly failing to protect specific rack positions the moment a unit goes down. Failure scenarios expose what nominal operation conceals.

The simulation led to concrete refinements—repositioned grilles for better air distribution, adjusted cabinet doors for more efficient rack airflow—all during the design phase, before construction, reducing costs and avoiding operational disruption.

The external environment required its own analysis. Roof-mounted chillers shared space with diesel generator exhausts, and at certain wind angles, exhaust gases reached the chiller intakes. We modeled this interaction and verified a fence enclosure calibrated to block exhaust interference without restricting the chillers’ own air intake. Not a data center problem in the conventional sense but a building physics problem, resolved with the same simulation approach.

The contractor won the Bologna project on the architect’s concept and took on the executive design. That’s standard practice. What’s less standard is when thermo-fluid-dynamic engineering gets involved.

CFD simulation requires specialized software, specialized expertise, and the engineering judgment to interpret what the results are actually telling you about a real system. Most contractors don’t carry this capability in-house—not because they should, but because it’s not what general contractors are organized to do.

In practice, validation is typically structured around compliance—proving the design meets standards and preparing documentation. But when it happens while the design is still evolving, it becomes a chance to optimize.

In Bologna, simulation was embedded in the design phase. The model let us reposition grilles, adjust cabinet door configurations, and resolve interactions between external airflow and the technical systems—all at the model level, when changes were still fast and low-cost.

This pattern holds across projects: the value of engineering analysis depends less on the initial concept than on whether it is carried out while the design can still absorb the insights. CFD engineering is most effective not as an external check, but as a technical contribution that guides development, because the questions simulation answers are design questions—and they matter most when the answers can still shape the outcome.

The scale of energy consumption in data centers is unlike any other asset class in the built environment. Among data centers reporting to GRESB under the Real Estate Assessment, median energy use runs between 17–39 million kWh per year. For offices, the equivalent figure is under 2 million kWh. The gap in energy use intensity is equally significant: data centers report a median EUI between 1,184–3,095 kWh/m², compared to 137–159 kWh/m² for office buildings.

What is striking is not the average but the range within the data center category itself. The highest-consuming decile reaches over 9,500 kWh/m² annually. The lowest sits under 200 kWh/m². Facility age, size, use type, and climate all contribute to that spread. So does the quality of the engineering decisions made at the design stage—decisions about cooling system design, airflow management, and thermal validation that determine how efficiently a facility operates over its lifetime.

CFD simulation, failure scenario modeling, and external airflow analysis are not certification requirements. They are the engineering foundation that determines whether a building performs as designed, year after year. Advanced engineering frameworks provide the structure to demonstrate and communicate that performance. But the performance itself has to be built in from the start.

The project had been developed by the design team with future expansion of the computing areas already in mind: PUE projections across the expansion timeline, chiller capacity sized for future load rather than current draw, and airflow validated under the expanded configuration while the system was still on paper. When the time came to expand, the engineering foundation was already in place.

The alternative is what happens on most projects: design for current load, treat expansion as a future problem, then reengineer an already built system when growth arrives. This results in higher cost, fewer options, and more operational risk than if the growth had simply been modeled upfront.

This matters particularly now. Data center power density is growing, and the facilities being built in this cycle will need to handle loads that were not in the original brief. If the cooling system, power distribution, and airflow management were not designed with that growth in mind, adaptation becomes expensive. If they were, it becomes a planned phase, not an emergency retrofit.

That engineering question belongs at the design stage, not after the first configuration is already locked in.

The Bologna project succeeded because thermo-fluid-dynamic engineering was part of the design process from the beginning, not layered on at the documentation stage. This is not about whether contractors or architects are doing their jobs well. It is about when the engineering questions that determine long-term performance actually get asked.

CFD simulation, failure scenario modeling, and external airflow analysis are not compliance tasks to be checked off for certification. They are design tools. For data centers—where mechanical systems are the core of the facility and performance gaps compound over decades—timing makes all the difference. The engineering decisions that guarantee a facility’s resilience, not just on commissioning day but five years later when loads have grown, do not happen during construction. They happen at the drawing board.

Across projects, there is a consistent pattern: facilities where thermo-fluid-dynamic engineering shaped the system from early design stages tend to perform better, cost less to operate, and adapt more readily to changing loads than those where it arrives as a validation exercise. That is the real lesson from Bologna. The technical work itself is standard practice. What makes it effective is simply executing it when the design is still open to change.

This article was written by Ing. Marija Golubović and Ing. Danilo Franchi, Energo Group. Learn more about Energo Group here.