Pure DC Series: Peering through the cloud

Just a few years ago, data centres were often described as the most important industry you’d never heard of. That is no longer the case. Today, they are firmly recognised alongside transport and energy as critical infrastructure, and they are the engine rooms of the digital economy that businesses, governments and consumers rely on every day. 

For SEGRO, this is not new territory. We have spent over 20 years developing Europe’s largest cluster of data centres at Slough Trading Estate, giving us a significant head start as demand intensifies and expectations evolve. What was once a niche has become one of the most strategically important asset classes in the world. As the sector rapidly changes, we continue to learn and adapt. We have dedicated data centre and energy functions, bringing experienced energy professionals into the business to reflect the increasing complexity of this market, the importance of power and the growing role data centres will play in SEGRO’s future portfolio.

One of the most important shifts underway is the emergence of two distinct, yet increasingly overlapping, types of facilities: cloud data centres and AI data centres. And while on the face of it they may look as alike as sugar does to salt, when you get beyond appearances, they are very different. Understanding these differences directly affects how these data centres are planned, built, powered and invested in. 

There is a third category as well called edge, which is a smaller bank of servers acting as a DC that are often scattered around cities to provide ultra-fast capability for the likes of gaming and streaming. But for the purposes of this article, we will focus on cloud and AI.

Cloud data centres are the most established model and the kind you see throughout Slough Trading Estate. They house physical servers and storage systems to deliver cloud computing services over the internet, and are operated by global providers such as AWS, Google, Microsoft and Oracle, along with specialist co-location companies like CyrusOne, Digital Realty, Equinix, GTR, Iron Mountain, Kao, NTT, Pure DC, VIRTUS, and others. 

These facilities are typically location-driven, often near major population and business centres, clustering with other data centres to form ‘Availability Zones’ or ‘Regions’. This enables resilience, low latency, and access to independent power and fibre connectivity. Clustering also streamlines planning, attracts supply chain investment and strengthens regional digital ecosystems, a key factor for policymakers and local authorities. Cloud data centres have become the silent backbone of digital transformation, supporting everything from e-commerce and financial services to media delivery and remote working. Indeed, there are entire sectors of our modern economy that wouldn’t exist without data centres. For example, Deliveroo is a food company that has no restaurants, Uber is a taxi company without any cars or drivers; and AirBnB is an accommodation company with no real estate assets.

By contrast, AI data centres can be broadly divided into two types. The first supports inference AI, which is the ‘deployment’ of trained models and is similar to cloud data centres in requiring high-speed, high-capacity fibre connectivity and proximity to major population centres. It does, however, require some additional compute. Indeed, cloud and inference AI could even be run from the same facility and we are seeing a rise of hybrid facilities. The second supports generative or machine-learning AI focused on training large models. These facilities are hugely more computationally intensive and need lots of land and access to low-cost power, neither of which the UK has in abundance, but are typically found in more remote locations, often far from major cities.

In addition, the structural demands inside AI data centres are higher. Floor loading must support heavy racks, cooling units and power distribution systems. With the speed of advancement, owners and operators need to future proof their data centres too in anticipation of the updates customers will likely want to make. 

For developers, investors and planners, the requirements of different types of data centres translates to a step-change in design complexity and flexibility, infrastructure requirements and total capital investment, and legacy planning definitions and government policy may no longer apply.

Power availability is now one of the biggest constraints on new development. AI itself is being used to optimise data centre operations, dynamically adjusting cooling, managing workload distribution and improving server utilisation. Cooling strategies are evolving rapidly, with closed-loop liquid systems emerging as a water-efficient alternative to more traditional methods.

As the sector evolves, so has SEGRO’s role. Our 2.3GW land-enabled power bank focuses on opportunities in key European Availability Zones, locations which will benefit from both cloud and inference related AI demand and where both land and power are scarce. This makes the opportunities we’ve created exceptionally rare and valuable, particularly those supporting cloud and inference AI operations. Historically, we delivered “powered shell” facilities, providing the structure and energy infrastructure while operators handled internal fit-out and cooling. In our joint venture with Pure DC in Park Royal, we are moving further up the value chain, delivering fully fitted solutions that incorporate the latest in sustainability, efficiency and performance. For example, our new designs will use air-cooled chillers operating in closed-loop systems with a Water Usage Effectiveness (WUE) close to zero, which represents over 100 million litres of water saved annually compared to best practice. 

This level of alignment between developer and operator is now becoming more valuable as planning, designing and delivering data centres now involves a more complex set of considerations than for traditional industrial buildings. Internally, the buildings must support heavy loads, advanced cooling, multi-level layouts and ultra-secure connectivity. Fit-out timelines can be critical, particularly for Hyperscalers and AI operators competing on speed to market. Not all land, and not all developers, will be capable of meeting these demands. Modular designs and construction methods can enable a faster build and reduce embodied carbon emissions through offsite manufacture and onsite assembly. Those who plan early, secure power and design with flexibility in mind, in the right locations will have a clear competitive advantage.

As AI becomes more deeply integrated into everyday applications, the role of data centres will only grow. While there is some crossover, for instance, inference AI workloads may be hosted within cloud facilities, the underlying design and property requirements of generative AI remain distinct and far more demanding, and should therefore be treated differently. The real opportunity lies in understanding where each fits, planning for hybrid requirements and investing in infrastructure that can adapt to changing technological needs.

#BuildingTheDigitalBackbone #DataCentres #PureDC

This article is part of a series delivered by the SEGRO and Pure DC joint venture which will be delivering a fully fitted data centre in Park Royal. The series combines SEGRO’s 20 years of experience delivering modern, high quality data centre shells to create Europe’s largest cluster in Slough, with Pure DC’s expertise as a leading developer and operator of critical digital infrastructure. The series will cover a wide range of topics including how data centres support communities, drive innovation and power economic growth and enable everyday life.