Why adaptability beats capacity for data centers

A long corridor with a sleek black floor, glowing green lights in the ceiling and rows of LEDS on either wall
As AI demand accelerates, data centers must adapt faster, smarter and more sustainably. (Image credit: Getty Images)

There’s always a quiet moment of calm in my morning that usually comes just after the 6 a.m. handover from the night team, and before meetings with customers and prospects get started. It’s time I reserve for reflecting and planning, looking at both the week ahead and working on the longer-term requirements our facility needs to meet.

It’s the latter of those priorities that’s taking more of my time these days, and it reflects the changing nature of the conversations I’m having with CIOs and technology leaders. Two years ago, most customers, especially those in regulated sectors such as finance and healthcare, were asking about uptime, certification and cost.

Today, the first question is almost always about AI.

Latest Videos FromTechRadar
Eduardo Espinel

Head of Data Centre at Ilkari.

They want to know if the facility can support the capacity they’re going to need, and whether it can do so on a timeline that matches how quickly their workloads are scaling. These are fair and important questions, and, in all honesty, they are not always easy to answer.

The increase in demand, driven by AI, means decisions we previously had years to make, now need to be made in a matter of months. Not impossible, but it fundamentally changes the way we think about designing and running facilities longer term.

Density, not just capacity, is what matters for AI workloads

Around the world, capacity challenges dominate headlines, but the infrastructure story underneath it is almost entirely about power density, cooling, and grid capacity. Historically, infrastructure was designed for low power density, whereas today, AI needs much more power per rack as well as more advanced cooling.

A good example is a rack that housed standard enterprise workloads at 5 to 10 kVA two years ago is now being asked to support 15 kVA within the next 18 months, and we’re already planning to support 25 to 30 kVA within the next few years. That is not an incremental change. It requires a different physical design that provides more density in an efficient way, often through modular Performance Optimized Data Centre (POD) architecture and in-row cooling. This allows us to add power and cooling gradually, controlling investment, improving efficiency, and ultimately, allowing us to increase capacity as demand grows.

The lack of density for AI workloads is why many established markets, including some of Europe's data center hubs, are struggling to add capacity at the pace required by demand.

For CIOs evaluating a provider, the useful question isn't “do you have any capacity today?” it's “how does your design accommodate the density I'll need in three years?” because a facility built for yesterday's power needs can quickly become an operational risk if it can't be reconfigured without a lengthy rebuild.

A good example of this is a POD we recently reconfigured within 30 days to meet a customer’s changing requirements, changing 40 racks at 5 kVA, to 20 racks at 10 kVA to support high-density demands. It was a significant project, and the kind of turnaround that's only possible when a facility is designed for change and adaptability, rather than attempting to bolt flexibility on afterwards.

Cooling design has to keep pace with density

Managing excess heat caused by the increased processing demands of AI workloads is becoming an increasingly important part of infrastructure design and customer consideration.

The good news is that cooling technology is improving all the time, making high density possible without needing to prioritize colder climates.

We mostly use air cooling with in-row technology, and because our design is modular, we can increase cooling as the IT load grows. For example, our customers typically join with an initial setup and gradually scale their capacity as their requirements grow.

As the customer’s IT load increases, we progressively activate the in-row cooling units, ensuring that cooling capacity is only deployed when and where it is needed. We also use a refrigerant-based cooling system rather than a water-based one, which keeps our water consumption low to begin with

This approach allows us to operate more efficiently when it comes to both energy and water use, which, as we know, is under increasing scrutiny from regulators and communities alike, and as a limited resource, needs to be used responsibly.

Local footprint must be part of the design conversation, too

It’s been hard to escape the recent tranche of negative media coverage about data centers taking valuable resources (land, water and power) from local communities. It’s true that they have an impact on the local community, but it’s unfair to frame this as entirely negative.

In my experience, a site's economic footprint is much larger than the building itself. Beyond the direct operational jobs, data centers support a much wider chain of indirect employment in sectors such as security, maintenance, cleaning, logistics, and transport. They also drive investment in the surrounding energy and connectivity infrastructure.

The ability to serve the local community and economy, as well as our customers, is an integral part of my role. It’s also something I encourage technology leaders to consider as a legitimate part of their due diligence. After all, banks, government services, cloud providers, and now AI, all depend on the same underlying capacity, and the locations that can offer credible, well-designed and well-run infrastructure, also attract more technology investment, digital skills, and the surrounding ecosystem that a growing digital economy needs.

Sustainability has to be operational, not aspirational

The scrutiny on data centers' environmental footprint is only intensifying as AI-driven demand grows, and rightly so. When it comes to managing a facility, this means building sustainability into daily operations, rather than treating it as a future aspiration or initiative that is bolted on after the fact.

In practice, this means ensuring efficient cooling design, transparent community engagement, and genuine investment in the skilled engineering talent needed to run higher-density infrastructure safely. At Ilkari, we do this through our POD-based design and in-row cooling architecture, which allow us to scale cooling capacity efficiently as IT load grows, and maintain efficient and sustainable data center operations.

It also means working with local governments to make sure the right investments are being made in local power and energy infrastructure to allow us to operate in a way that is sustainable for the local community.

The infrastructure design questions CIOs should be asking

The reality of running a data centre as AI demand increases means a fundamental shift in how facilities need to be designed and built.

For CIOs thinking about what infrastructure requirements their AI workloads need, my advice is not to start with a capacity spreadsheet, but with a short list of broader questions, including:

- How does this provider's design scale with density, not just floor space?

- Can it demonstrate its ability to meet your requirements under real conditions, not just on paper?

- Is sustainability built into day-to-day operations (cooling design, community engagement, skilled engineering talent), or is it a separate initiative?

- As density increases, has the provider planned for the point where air cooling is no longer sufficient, and what alternative cooling technologies and infrastructure would that require?

In my experience, the organizations getting this right aren't the ones with the flashiest AI roadmap, they are the ones designing and building for change from the outset.

We've reviewed and ranked the best dedicated hosting services.

This article was produced as part of TechRadar Pro Perspectives, our channel to feature the best and brightest minds in the technology industry today.

The views expressed here are those of the author and are not necessarily those of TechRadarPro or Future plc. If you are interested in contributing find out more here: https://www.techradar.com/pro/perspectives-how-to-submit

TOPICS

Head of Data Centre at Ilkari.

You must confirm your public display name before commenting

Please logout and then login again, you will then be prompted to enter your display name.