3 ways material choices can improve data centre construction and reduce carbon
April 28, 2025
To design and build more efficient data centres, the industry must consider both new materials and how it uses existing materials
In October 2024, Microsoft made headlines when it announced plans to build two data centres in northern Virginia in the US using wood. This move is part of its efforts to create some of the most sustainable data centres in the world.
The trial explores how using wood can help reduce carbon emissions in data centre construction and designs. Microsoft estimates that its plans for these two data centres will reduce their embodied carbon by 35 percent compared to conventional steel and 65 percent compared to typical precast concrete.
It¡¯s a timely example of how the data centre construction industry is rethinking material choices to meet its looming net zero targets. My US colleagues have summarised how their clients are using mass timber in data centres. Meanwhile, in the UK, there are still challenges¡ªprincipally insurance¡ªthat we need to address before we can use timber more widely.
For me, low-carbon design starts with the idea of ¡°using less stuff¡±. As a designer, I see this as a great challenge in being efficient with the materials you have. On this basis, beyond new materials, it¡¯s also important for the industry to reassess how it uses its existing materials palette. Doing so can achieve the design utopia of lower carbon, lower cost, and even quicker construction.?
Aerial view of large data centre.
On recent schemes and in the work we¡¯re currently doing, we have been exploring this for realists and working out how to get there. We have asked ourselves this: How can we prevent over-specifying and overdesigning data centres? Knowing the answer will help us reduce inefficiencies, costs, and carbon emissions.
There are three big opportunities for potential savings:
- Questioning baseline requirements
- Rethinking standardised structural components
- Exploring new materials
Let¡¯s dig into them one at a time.
Opportunity one: Questioning baseline requirements
The baseline requirements for data centres often come from logistics and distribution (L&D) sites due to a perception that they have similar uses. They are spaces for storage and process, as opposed to habitation. However, data centres are different to L&D sites in both their use and function. If baseline requirements aren¡¯t examined closely, a data centre could end up with an over-designed structure using larger, heavier, more expensive, and more carbon-intensive components than necessary. I see this all the time.
Here are two common examples:
- Floor surface regularity and tolerance criteria (i.e., the maximum allowable deviation in the floor along a fixed length)
- Deflection criteria (i.e., the maximum allowable movement under the application of load)
In L&D sites, these are strict criteria because they support tall racking. If this equipment is installed on an uneven surface, it can exhibit significant vertical out-of-tolerance, leading to possible instability.?
Rows of server hardware in a data centre.
Data racks are typically only 3.5 metres tall. This makes out-of-tolerance less critical in data centres. Furthermore, in a cold aisle containment configuration, data racks are installed on a raised access floor. This can accommodate significant out-of-tolerance floor levels. If high-tolerance floors are required in data centres, it leads to stringent construction-stage deflection criteria or the installation of levelling screeds. These are high in embodied carbon and impact the construction programme.
In one study for a major data centre operator, we found that by relaxing the floor flatness criteria and construction-stage deflection limits we could reduce the steel tonnage within the floor plates by 12 percent. This created a possible savings of more than 75 tonnes of steel. This is the kind of re-evaluation of data centre construction that is essential. It will help us improve data centre efficiency and build the most sustainable data centres possible.
We can also find efficiencies in the tenant¡¯s loading requirements and equipment loading specifications. We do this by developing a clear understanding of the operator¡¯s activities within the facility. On a pair of previous projects for a major data centre operator, both for the same hyperscaler client, we were handed loading requirements that were twice as much on one project as on the other. However, they were both based on the same baseline specification. The projects were so far along that we couldn¡¯t change them. But it¡¯s a good lesson.
The equipment in data centres is heavy, but the space is often sparsely populated due to the need for airflow. As such, understanding how the equipment is to be moved around the building and deployed during installation can offer occasions for more rationalised loads.?
Baseline requirements are the starting point for all structural design. In my experience, getting this right will always be key to efficiency. This is especially critical in this sector, where the requirements are onerous and the facilities are large. Therefore, we need to design them efficiently. If we don¡¯t, they will consume significantly more materials.
Opportunity two: Rethinking standardised structural components in data centre construction
Something else I find myself frequently talking about is how data centre construction is often highly standardised, with a lot of repeatability. This is especially true in North America, where space isn¡¯t as constrained as in the UK, allowing ¡®cookie cutter¡¯ designs to be deployed for most projects. There is still scope for repeatability on the more constrained sites in the UK. However, this is more likely to be through a project-specific standard cell, rather than the operator¡¯s standard design.
The rise of AI and other new technologies has led to a need to revisit these standard designs. Now we see that design turnover is faster. This puts the onus back on project engineers to develop a scalable, standardised project-specific design that delivers optimal efficiency.??
For example, ¡ we challenged the use of standardised beams and columns within a design received at RIBA Stage 4. This ¡ led us to save over 25 percent of the steel.
It¡¯s key to understand where standardisation adds value and boosts data centre efficiency. ?Then, it can help us deliver an optimised structural solution.
The common misconception in data centre construction
There is a common misconception that using standard element sizes within steel frames will make the manufacturing and construction more efficient. This can lead to large inefficiencies in design and is something the industry must challenge. Let¡¯s explain.
It¡¯s true that data centres are often formed of repeatable bays. But within these bays there are often different spans, loadings, or constraints. If element standardisation is applied, it can lead to inefficiencies. And since these bays are repeated¡ªsometimes more than 100 times on a single development¡ªefficiencies found through design rigour on a typical bay are scalable, leading to an opportunity for significant material savings.
For example, on a recent project for a colocation provider, we challenged the use of standardised beams and columns within a design received at RIBA Stage 4. This, alongside a more rigorous approach to design, led us to save over 25 percent of the steel. That¡¯s an embodied carbon saving of around 1750tCO2e. Or, put more simply, 2,000 one-way flights from London to New York. This approach was applied again to another project, this time at RIBA Stage 3, where we realised steel savings in the order of 30 percent.
I¡¯m not suggesting that structural standardisation has no place in data centre design and construction. Buty how and where it is used needs careful consideration. For instance, where there are interfaces with the building fabric, standard detailing of highly repeatable details allows for speed on site and better quality assurance.
Where standardisation is required, it¡¯s important to check that the design inputs, spans, loads, and constraints are also standardised. This helps to prevent material inefficiency.
In some cases, repeatability and standardisation can help us to improve data centre efficiency. Repeatability reduces the cost and programme impact of using structural assemblies in place of individual elements. These can be vastly more materially efficient than a single element, saving cost and carbon. I think this is an area where we have only scratched the surface of the potential savings available. ??
Server hardware inside a data centre.
Structural zones
As discussed above, a key influence on structural efficiency is assuming the correct constraints within the design, such as the structural zone, which dictates member depths. If this is highly constrained, it can lead to the use of members that have a higher weight, cost, and embodied carbon.
In some sectors, the neat demarcation of spaces for different members of the project team across a building cross-section can be useful. However, these silos break down in a data centre because of the complex and specialist systems required to keep the building cool. This leads to variable zones within the building. In some instances, it can lead to different zones depending on which direction the element is spanning.
For example, within a data hall with hot aisle containment cooling, beams are located within the cooling plenum. As such, the depths of beams that are perpendicular to the airflow act to constrain this airflow, while those that are parallel do not. This leads to different design constraints depending on the airflow direction and beam orientation.
Understanding these constraints at the detailed design stage can allow for the efficient design of a given arrangement. However, applying this understanding in the concept stage of a design can lead to even greater material efficiency. And, in some instances, better performance of the mechanical systems.
On a recent project, we reviewed the column placement adopted by the incumbent engineer. The columns and therefore primary beams were in the computer room air handling corridor cage wall. Air velocities within the plenum at this location are at their highest, which means any obstruction to this flow has an adverse impact on pressures. This arrangement results in significant constraints to element sizing, so as to not compromise the performance of the mechanical systems.?
We found an oversized plenum and an increased building floor-to-floor height. Due to the stage of the design at the point we were engaged, there were limits to what we could do to increase efficiency. We adjusted the beam and column arrangements where possible, which improved the airflows and reduced the plenum height. Where we could not move the beams and columns, we reviewed the continuum fluid dynamics modelling of the airflows in the plenum alongside our building physicists and mechanical engineers. We found that introducing holes in the beams improved the airflow and reduced the plenum depth.
When we understand the constraints that define structural zones it allows for more efficient structural members and material use. Or it can deliver a smaller or more efficient building. This reduces data centre construction and operational costs.
Ground works and foundation design
Challenging constraints can work just as well with ground works and foundation design as it does with the superstructure. For example, on another project, we saved approximately 11,000m3 of concrete in the building¡¯s substructure by challenging the initial design. We discovered that the ground was suitable for the use of ground-bearing slabs and strip and pad foundations instead of the proposed 20-metre piles, suspended slabs, and deep raft foundations. This was equivalent to an embodied carbon saving of 6,300 flights from London to New York and avoided more than ?2 million in capital costs.?
A data centre under construction.
Opportunity three: Exploring new materials in data centre construction
When it comes to building the most sustainable data centres, there are ongoing debates about the future of timber data centres in the UK. However, we need to start thinking creatively about what a hybrid structure could look like. This is especially true if we are to hit critical sustainability targets and improve data centre efficiency.
When exploring new materials, we must use them to their best effect. We can do this by having a fundamentally efficient design. This is achieved through in-depth review and qualification of requirements and constraints.
The answer won¡¯t be as simple as swapping higher-carbon materials for lower-carbon ones in data centre construction. I strongly believe we need to go back to the drawing board on the entire structural form or even that of the building fabric. We need to consider whether there are ways to explore more efficient structural systems, by responding to the way the structure of a data centre interacts with the systems within it. Beyond the focus on material efficiency, we need any new approach to not compromise the speed of delivery. And we need to explore whether there are structural systems that may even improve this.
The answer lies in the effective deployment of new design for manufacture and assembly (DfMA) structural systems that can afford high material efficiency and speed of on-site construction. And we should use more low-carbon materials in place of some of the more carbon-intensive materials used now. This needs a mindset shift away from considering DfMA and modern methods of construction as either volumetric or 2D technologies offered by the supply chain. As designers, we need to consider how we can design off-site manufactured components and assemblies that can form a portion of the kit of parts within our client¡¯s standard designs. ???
The future of data centre construction
There are some exciting ideas about how to use timber to reduce the carbon impact of data centres. But we are just scratching the surface in terms of what can be achieved in data centre construction.
Of course, big changes can feel overwhelming. Especially when the pressure to deliver projects is extremely high. But, as the industry has shown time and again, new ideas can deliver multiple benefits and support the needs of this fast-growing market.?
