The key characteristics of data centres in the US and Europe – an overview for those involved in financings

Key features of data centres

Real estate fundamentals

So what is a data centre?  A data centre is a vast facility which houses computer systems and related infrastructure such as networking equipment, which allows data to be stored and processed.  Technology companies also use data centres to train AI models such as ChatGPT.  A “standard” data centre (which has not been re-purposed from existing real estate) is typically a low-rise square or oblong building, which is not particularly eye-catching from the outside, and is often located in an industrial or brownfield area, and as we detail later in this article, a key driver as to location is available power and grid connection.  The “shell” of a data centre will include reliability systems such as back-up generators, hardware, cooling towers, and fibre optic paths.

Inside, however, a data centre contains data halls, which house racks of servers and other IT equipment.  Data halls are fantastically complex, and represent the magic of a data centre.  Data halls normally sit on a raised floor.  When financing data centres, the size of the raised floors is one of the metrics which is looked at, together with a data centre's megawatt capacity and power availability (in contrast with more traditional real estate, where square footage alone tends to be the key metric). 

Data centres should not be thought of solely as real estate assets, however.  They are inherently operating businesses. Their services component, set out in the relevant service level agreements, and the credit quality of their tenants, often secured under long leases, form part of the package as to attractiveness for investors. 

Usage and tenants

Typically, data centre operators lease out space and provide certain services such as internet, power and capacity, cooling systems, emergency power supply, and/or maintenance and repair services.  Tenants may then provide some of their own equipment, and may reimburse certain of their expenses depending on their lease.  Data centre leases will typically be gross, modified gross (requiring the tenant to only cover electricity, having covered operational expenses in the base rent) or triple net (requiring the tenant to pay or reimburse the data centre operator for its operating expenses including taxes and insurance).  Data centres are typically staffed around the clock by employees for operational and security reasons.

We set out below the key types of data centre:

• Hyperscale: Hyperscale data centres are immense in size (the consensus definition is that to be “Hyperscale” a data centre has to contain 5,000 servers, spanning a minimum of 10,000 square feet (930 square meters) and offering at least 40MW of capacity with an ultra-high speed, high fibre network. A data centre under construction in Langfang in China will be over 6 million square feet in size - equivalent to 80 football pitches the size of Wembley.  They are used by household name cloud service providers usually containing one to three tenants under leases with a term of between 10 to 20 years.  The occupational leases are typically triple net lease or modified gross. Hyperscale tenants will often require high performance guarantee standards from their data centre operator.
• Colocation/retail: Colocation/retail data centres house small to mid-size tenants with more modest data storage and/or processing needs and shorter-term lease contracts (or even simply services agreements, which provide both parties with more flexibility than a lease).  They tend to be smaller and use less power, to be more commoditised, and to have more churn, sometimes with tenants with lower credit profiles; but they can have greater potential for upside.  Colocation centres may take longer to liquidate than a hyperscale due to their tenant diversity; but this also provides special servicers with more flexibility.  Colocation facilities may be older and in more densely populated areas.  The occupational leases are typically gross or modified gross.
• Enterprise: Enterprise data centres (which can be considered a subset of wholesale data centres) vary in size but are normally owned and operated by their end-users, often for their own infrastructure – such as by banks and trading firms.  They may have in-house operational staff.  These facilities can also be subject to sale and leaseback arrangements.
• Wholesale: Wholesale data centres are similar to enterprise data centres but are not owner-occupied.  They have tenants who are responsible for the maintenance and management of their racks, storage and networking equipment (in contrast to colocation tenants which receive certain support).  Wholesale data centres tend to be used by enterprises such as cloud computing providers, AI companies and healthcare providers, over long periods of time.  Rent for wholesale data centres is often priced in terms of power availability or critical IT load.  Some enterprise data centres are sold and leased back and then repurposed as wholesale data centres.
• Edge: Edge data centres are a further subset.  They are small in scale, and typically located on the outskirts of major markets and close to cell towers.  They process, collect and distribute time-sensitive content requests from end users' Internet of Things devices (such as smartphones and laptops, autonomous vehicles and smart electrical grids).  They are critical to cultivating 5G density.  Edge data centres are smaller deployments which have not yet become part of the data centre securitisation space in the US or Europe.

Data centres also vary in type from a construction perspective:

• Turnkey: Turnkey data centres are “built to suit”, with landlords investing significant capital towards building a state of the art data centre facility, with all infrastructure.  They typically contain no servers or related equipment, which are supplied by the tenant.  Most turnkey assets are built by landlords who specialise in data centres, and they tend to be leased on a gross basis or triple net basis.  Tenants will pay more for them.  By their inherent nature, they risk becoming obsolete quickly due to technological changes, so they require regular investment and maintenance.  Turnkey facilities typically become colocation centres.
• Powered shell: Powered shell data centres are a building shell, often leased to a single tenant, with only power and network connectivity in place, into which the tenant installs its own equipment.  Tenants invest significant capital to convert this shell real estate into a functioning colocation or wholescale data centre.  These tenants typically have triple net leases for a length in excess of 10 years, charged by the square metre.  The tenants are usually entities who want control over design and infrastructure, such as tech titans. 
• Carrier hotels: Carrier hotels aren't data centres in the traditional sense, but are rather central internet exchange points where metro fibre lines link with long-haul or sub-sea fibre.  They are often located in repurposed office buildings in central business districts and allow interconnection between multiple telecommunications service providers and colocation providers.  They offer excellent network connectivity and interconnections because of the presence of multiple network carriers, and they can be seen as trophy assets because they stitch internet networks together.  Powered shell data centres and carrier hotels are inherently less risky than turnkey data centres.

Rating agencies will often also categorise data centres and/or their markets into different “tiers”, based on capacity and maintenance requirements, redundancy and/or connectivity.

Growth

Data centre construction and usage has been growing massively due to increased internet demand (global internet traffic has increased by an average of 30% since 2018).  Cloud computing, which allows businesses to use multiple cloud based platforms instead of traditional IT systems, has been increasing.  The AI revolution continues, as does the roll-out of 5G networks, and the previous energy crisis has abated.  Worldwide data centre demand is expected to rise around 20% annually over the next few years, and around $700bn is expected to be spent on the construction of new data centres between 2025 and 2028. There is currently 1.7 gigawatts of new data centre capacity in development in Europe and a further 1.3 gigawatts in the planning stages across core and secondary markets in Europe.

Having said this, it is important to keep in mind that the data centre industry is a relatively nascent one (for lenders and investors).  It is still subject to risks, in particular in relation to energy, electricity and water availability (see below for more on this).  Additional pressures can come from regulatory demands around environmental impact and performance, privacy and AI, opposition from local communities (due to data centres' relatively low job creation combined with their need for space, water and power), high construction and operational costs, high demand for key components, supply chain bottlenecks, low supply of electricians and engineers, cyber-security and energy-security threats, competing technological advances, and geopolitical risks such as potential new tariffs on the AI and semi-conductor industries.

As a result, major Cloud providers have been seen to pause or slow select projects, even as demand for computing power remains strong.  There is also some tension between the demand for renewable energy and the 24/7 energy needs of data centres.  Data centres need to be constantly on, which makes them almost uniquely unsuited to grids relying on renewable energy sources (the instability increases as the renewable component increases due to basic laws of physics).  Power outages such as the ones seen recently in Spain and Portugal cannot be allowed to impact the data centre, so nuclear or fossil fuels need to be used as primary or 100% back-up (battery storage does not currently exist at anywhere near grid scale).

The medium to longer-term leases often seen in data centres, and the fact that energy costs are often passed on to tenants via triple net leases, help to alleviate concerns around stability.  Additionally, supply constraints might actually benefit existing data centre assets, as they reduce competition from newer assets being built.  High entry costs for participants can ensure that tenants renew their leases.  However, with these stressors in place, diverse capital stacks and innovative tools can be helpful to ensure efficiency.

Locations in the US and Europe

Choice of location of a data centre will typically be driven by access to a grid connection, renewable energy, fibre and internet connectivity, climate, proximity to users, political and geological stability, favourable tax incentives, and tenant demands.  A data centre operator will also need to take into account local regulations, planning requirements, environmental impact, price of electricity, and secrecy and security when choosing where to locate its data centre. 

Data centres will often be put in areas that were previously industrial or manufacturing centres, and near major fibre hubs.  Data centres which are less reliant on uninterrupted power supply, such as to train AI models before those models are deployed, can be put in more remote locations. In contrast, edge data centres are situated geographically close to the people they serve.  Availability of land is clearly less of an issue in the US than in Europe due to the US's far greater scale, although existing grid and fibre infrastructure (or cost and time to optimise them) are issues that have to be considered in relation to any siting decision.

The US is the largest and most mature data centre market in the world.  Within the US, northern Virginia is the largest data centres hub.  This is partly for historical reasons, as northern Virginia was the location of the first undersea cable that went from the US to Europe, and also because the government of the state of Virginia has encouraged data centres growth through tax incentives and low power costs.  Santa Clara, Chicago, Dallas and Colombus are also hot markets, some of which have doubled or tripled in the past year or so, as well as Salt Lake City, Seattle, Houston, Denver and Minneapolis.  Additional less obvious, but still growing, markets in the US are Kansas, St. Louis, Boise and Tulsa.

The European data centres market is smaller than that of the US, but is also growing rapidly, particularly due to a recent spike in AI companies in Europe.  It is estimated that it will grow 25% in 2025.  In Europe, data centres are primarily concentrated in Frankfurt, London, Amsterdam, Paris and Dublin (known as the “FLAP-D Cities”, which offer robust connectivity and attract major tech players).  Secondary markets such as Milan and Warsaw are growing, as are the Nordics with their cooler climates and abundant renewable energy supplies.  The Iberian Peninsula is popular due to its location as a connection point for Africa, the Middle East and the US, and locations for transoceanic cables, such as Marseille and Athens, are also seeing continued growth.  AI Growth Zones were introduced in the UK earlier this year (January 2025) as part of a series of governmental policy decisions aimed at facilitating the expansion of data centres in the UK and the UK becoming a leader in AI. 

Energy and electricity

A data centre is power-hungry – merely cooling the infrastructure takes a significant amount of power.  Therefore, access to a reliable energy supply is paramount.  This is typically obtained through access to the local power grid, but in some cases national and local grids have been struggling to build and maintain infrastructure at the rate and level required to match the growth in demand.  Ireland, for example, currently has unsustainable electricity growth projections due to its data centre growth and the national grid in the UK is currently undergoing a grid reform process in an attempt to address the backlog of grid connections by re-prioritising the queue such that proceedable projects (i.e. those with secured land rights and planning permissions / development consent orders) are prioritised.  Robust back-up systems are also required at data centres to ensure no or minimal power failures, through generators, batteries and/or UPS systems. 

Already highly energy-dependent, data centres are expected to require an increasingly vast amount of energy, particularly due to the rise in AI.  AI interrogates language and learns from interactions, which requires a heavy amount of processing.  As a result, global demand for electricity to power data centres is expected to double by 2026. It has even been suggested that data centres may start to be placed on the moon, which would offer unparalleled security, no disruption to locality and unlimited solar energy (although these data centres may be more expensive to construct)!

Interestingly, a decommissioned power station can be a unique opportunity for a data centre to be placed, particularly for a hyperscale tenant. These sites were originally designed for high power usage, and often come equipped with power transmission infrastructure. They are often also located near water sources, which can be a distinct advantage in planning the cooling systems required by data centres.

Heat, cooling and water

A data centre by its very nature produces a lot of heat, due to the energy that it uses.  It therefore needs an effective cooling system.  A cool atmosphere is also essential for the smooth functioning of servers housed in data centres.  Data centres achieve this through air cooling, liquid cooling, and even AI.  For example, rear-door heat exchangers combine cold air with liquid cooled heat exchangers; direct-to-chip technology uses a liquid that circulates through a cold plate in direct contact with the most power dense electronic components; and liquid immersion cooling entails placing the servers in a tank filled with dielectric fluid. 

These cooling systems often in turn require further power, especially those cooling systems which use less water.  They also mean that data centres consume a lot of water, sometimes creating concerns about putting pressure on local water services.  Certain hyperscale tenants have been aiming to become “water positive”, meaning that they offset water consumption in their data centres by providing water to areas of scarcity elsewhere.  However this can be more complex than carbon offsetting, because water supply is localised in nature, in comparison with carbon which is global in nature. 

ESG

As a result of the above concerns, ESG factors have become a topical concern in connection with data centres.  In recent years, data centre operators have found novel ways to recycle the heat produced from data centre servers, such as a data centre in Devon which heats a local swimming pool.  Other technologies, such as small module nuclear reactors and "direct to liquid" cooling systems, which use little water, are coming online, which will help mitigate some of the environmental impact of data centres. 

Data centres also require lithium and cobalt for their components, which come from mining in places such as DRC, Chile and Brazil.  There is concern that these sources may be connected with systemic human rights and labour abuses, giving rise to a focus on the “S” in ESG as well.  As a result, some data centre operators have put in place a green bond framework for their bond issuances.  Many data centre operators have also been focusing on investing in renewable energy.

Conclusion

Data centres are a relatively new real estate asset class, which have grown and are growing rapidly in both Europe and the US.  Whilst they might appear unassuming from the outside, data centres contain an intricate network of equipment and are power-hungry, requiring reliable energy supply and complex cooling systems.   The data centre industry is not without its challenges, such as regulatory pressures and ESG concerns, but it is also a creative one. With the AI revolution continuing, we look forward to supporting the continued growth and innovation of this asset class.

Look out for our next article where we will consider how data centres are financed, both in the US and Europe, with a particular focus on the securitisation market.

Authored by Isabel Tinsley, Emil Arca, Sian Howes, Tim Hinrichsen, and Daniel Norris.