Densification and the Geometry of Change: Designing Data Centres in an Age of Uncertainty

In this period of flux, we face a dilemma: how do we commit to design and construction when we don't yet know what the future of IT load will require? 

At Bryden Wood, we’ve been exploring this challenge not only through our project work on high-density, next-generation facilities, but also through the conversations sparked at the Accelerate Data Centres summit. This event, hosted in London in early 2025, brought together specialists from across the data centre ecosystem: developers, operators, engineers, architects and integrators. The aim was to share insights, debate dilemmas and confront the realities of delivering infrastructure in an industry moving faster than ever. 

What follows is a perspective shaped by that discussion, by practical delivery experience and by a belief that our approach to data centre design needs to change — not to chase the future, but to enable the present.  

 

The Density Dilemma

 

 

It wasn’t long ago that 3 to 5 kW per rack was standard. Today, hyperscalers are routinely designing for 30 to 60 kW, and in AI-focused clusters, densities of 80 to 150 kW per rack are becoming the norm. Some forward-looking programmes are already contemplating 400 kW or more. 

These aren’t abstract numbers. They’re driving fundamental changes to the scale, shape and logic of data centre buildings. 

At the same time, it’s increasingly difficult to predict where the market is heading. Will AI training continue to dominate, or will AI inference shift workloads back to distributed, lower-density environments? How quickly will liquid cooling be adopted across the sector? What will the average rack look like in five years? 

We don’t know — and that’s the problem.  

 

Designing in the Dark 

 

Data centres often take three to five years to deliver, constrained by power availability, planning delays, regulatory hurdles and long-lead infrastructure. In that time, IT requirements can change dramatically. A facility designed for 30 kW racks might need to support double that, or more, by the time it goes live. Yet developers are being asked to commit early: to secure power, navigate planning and procure long-lead plant. 

This creates a tension. Clients want to wait to see where density trends go. Developers want to move. Designers are caught in the middle. 

We often talk about future-proofing in this industry, but right now, that concept feels increasingly hollow. No one can predict the future of compute, but that doesn't mean we can't design for it. 

 

Present-Enabling Infrastructure 

 

This is where we shift the design mindset: present-enabling, not future-proofing. 

Instead of waiting for perfect clarity, we define a stable foundation. We design the primary infrastructure to enable progress today, while keeping the IT spaces flexible enough to adapt as density and load types become clearer. 

The infrastructure layout itself remains broadly consistent across scenarios. What changes later are the final elements: distribution topology, cooling loops, and the configuration of data halls to suit specific IT demands. 

In this model, the heavy lifting happens up front. Generators, primary electrical switchgear, UPS systems, chilled water systems, structural cores, risers and critical ancillary plant are all fixed and built early. The data halls follow, shaped around this infrastructure rather than the other way around. 

 

The Inversion: From White Space Outward to Infrastructure Inward 

 

At low densities, data centres have traditionally been white space led. The architectural layout of IT racks drove the floorplate. MEP systems were arranged around it: on rooftops, in back-of-house plantrooms, in yards or on gantries. 

But as rack densities rise, and especially as liquid cooling becomes prevalent, that relationship flips. 

Power and cooling infrastructure has become the spatial driver. CDU rooms, PDU zones, pipework distribution routes, and vertical risers all require large, highly coordinated spaces. These systems can no longer be arranged around the IT layout — they define it. 

At high densities, the building is no longer designed from the white space outward. It is designed from the infrastructure inward. 

This shift changes everything. Floorplates may need to break into smaller, service-proximate modules. IT rooms might cluster around risers rather than stretch out in long rows. In multi-storey designs, upper levels may give way to plant decks instead of additional data halls, as the weight and supporting infrastructure of high-density, liquid-cooled racks can make upper-floor deployment progressively less viable.  

 

Innovation in action — this breakthrough cooling system revolutionised data centre deployment. Scaled into full production by a supplier, it’s now used in dozens of new DCs around the world.

 

Can We Create a Flexible Chassis? 

 

This leads to an idea we’re exploring at Bryden Wood: the creation of a plant-first chassis, a repeatable building infrastructure that can support a wide range of eventual IT densities. 

The concept: 

• Design a fixed MEP backbone: generators, switchgear, UPS systems, chillers, dry coolers, primary risers, and so on. 

• Engineer the structural frame for adaptability. Fix primary column grids, vertical load paths, and cores early, but defer intermediate floors. Use flexible structural tie-ins and, where appropriate, consider temporary or modular roofs for early weather protection without locking in the final form. 

• Shape the white space later, based on confirmed IT density and load type: 

• For low-density air-cooled: stack two or three lightweight halls above one another. 

• For high-density liquid-cooled: concentrate IT on the ground floor and use upper levels for CDU rooms, power and chilled water distribution, and heat rejection. 

• Maintain modular plant bays and risers that allow deferred or partial fit-out based on demand. 

This approach enables early procurement of long-lead items, keeps the delivery programme moving, avoids locking in density assumptions too early and allows time for the IT requirements to clarify. 

It is, in effect, a two-speed delivery model. Build the plant early; flex the compute space later. 

 

But Can the Plant Itself Evolve? 

 

Here, we turn the lens around. If IT hardware is becoming more compact and dense, should the same not be true for the plant? 

Right now, plant is the constraint. It scales disproportionately. The more power and cooling you need, the more space and complexity the MEP systems require. That often cancels out the gains in IT density. But that might not always be the case. 

Emerging technologies offer hints at a different future: 

• Small modular nuclear reactors (SMRs) promise dense, self-contained power generation 

• Fuel cells could replace diesel gensets with quieter, more compact, zero-emissions units 

• Direct current power distribution reduces losses and component size 

• Two-phase or dielectric cooling may allow for more efficient heat removal at the rack 

• Heat reuse networks are already technologically viable but remain underutilised due to planning challenges, economic factors and integration with local energy systems. Where implemented, they can significantly reduce or even eliminate the need for on-site heat rejection systems 

These technologies are progressing at different speeds. But collectively, they point to a future where the plant may once again give way to other drivers of form. 

 

The Shape of Things to Come 

 

Right now, however, the physics hasn’t caught up with the ambition. Power must still be generated and transformed, heat must still be rejected, and maintainability must still be guaranteed. These realities keep plant large and force it to the centre of design. 

So, in the near term, we must design accordingly. The data centre is no longer a building with infrastructure around it. It is infrastructure, with space for compute inside it. 

In the geometry of the high-density data centre, the plant leads and the architecture follows. 

That may seem like a shift in priorities. But it’s also an opportunity. 

 

A Quiet Shift in Priorities 

 

What’s perhaps most interesting about this shift is that, as rack densities rise, the centre of gravity in data centre design quietly moves. Where once the form of the building followed the logic of the IT layout, it now increasingly follows the infrastructure: cooling loops, electrical topologies, service access and resilience pathways. In effect, the services are shaping the space. 

This doesn’t diminish the role of architecture. It reframes it. As plant systems become more central, the challenge is one of integration. The goal is to create buildings where spatial organisation, utility logic and future adaptability are considered as one. 

At Bryden Wood, with our roots in architecture and a long-standing commitment to integrated design, this shift is less a disruption and more a continuation of how we’ve always approached complex challenges. We work across disciplines, ask better questions and evolve our thinking as the context demands. 

 

Looking Forward 

 

 

Of course, the story doesn’t end here. As technologies like SMRs, fuel cells, direct current power and advanced cooling continue to mature, the infrastructure itself may begin to shrink or at least simplify. That might enable new typologies, where plant and white space coexist more seamlessly, and perhaps see a renewed architectural expression in the form. 

But for now, we work with what we know. We accept that compute is changing faster than the systems that support it. We build infrastructure that gives us optionality. We delay what we can. We fix what we must. 

We design not to predict, but to be ready and adaptable, enabling what’s needed now — and whatever comes next.