The question of where to put your servers has changed character over the past two years. Latency to a financial market still matters for some workloads, but for many others it has been overtaken by a more practical concern: can the facility supply the power, and can it actually cool the hardware you want to run?
The UK remains Europe's largest data centre market, but the supply picture is tightening. Power availability, grid headroom, and cooling design now sit alongside connectivity as primary selection criteria. Carbon-Z operates UK colocation sites in London, Swindon, and Manchester and is certified to ISO 14001, ISO 27001, and ISO 45001, and we work with hardware ranging from standard air-cooled racks to immersion pods. This guide sets out how we think about the UK hub categories, what each one is suited to, and the issues we see come up most often during planning.
The state of the UK colocation market
According to a UK Parliament research briefing on data centres , the country hosts more than 240 operational facilities, contributing roughly £4.7 billion in gross value added each year . National colocation capacity sits at around 1.6 GW, with forecasts placing this between 3.3 GW and 6.3 GW by 2030 . Data centres consume about 2.5% of UK electricity, a figure expected to roughly quadruple within the decade.
In September 2024, the government formally recognised data centres as Critical National Infrastructure , the first such designation since 2015. Subsequent announcements introduced AI Growth Zones intended to encourage development outside the established clusters.
The headline trend, set out in the CBRE 2026 UK Data Centres Outlook , is that take-up has exceeded new supply for the past four years and is forecast to do so again in 2026. If you are planning a deployment in 2026 or 2027, you are competing for capacity that is, in many places, already spoken for.
The four types of UK data centre hubs
Rather than thinking in terms of cities, we find it more useful to consider hub types. Each one solves a different problem and comes with its own trade-offs.
The historic heart of UK colocation is where financial trading, cloud availability zones, and dense fibre routes converge. Most of the country's largest facilities sit within or just outside it. It is the right choice when low latency to financial markets, content delivery networks, or major cloud providers genuinely matters.
The constraint is power. Available grid capacity in the densest parts of the cluster is now the binding factor on new development, and as a result, new supply is being pushed outwards into surrounding counties where headroom is greater, but the latency advantage softens. Our own London colocation site is positioned for this category of workload, supporting deployments from standard racks through to high-density GPU and AI environments where space, power, and infrastructure are under constant pressure.
Best for: financial services, ad tech, cloud peering, content delivery, and latency-sensitive enterprise workloads.
Watch out for: premium pricing, multi-year wait times for greenfield capacity, restrictions on high-density deployments in legacy halls.
The second tier has matured noticeably over the past five years. These locations offer credible peering, decent fibre routes, and more accessible power than the capital cluster, often at a meaningful discount on rack and kilowatt pricing. They tend to suit AI training, rendering, and HPC workloads where the latency premium of the capital is not needed.
A regional deployment also gives you geographic diversity for disaster recovery, which has become a more pressing concern as resilience requirements tighten across regulated industries. Our Swindon colocation facility sits on the M4 corridor and is a good example of this category: strong connectivity into London without the metropolitan power constraints, and the flexibility to scale into private suites or high-density pods as workloads grow.
Best for: AI training, GPU clusters, HPC, disaster recovery, regional enterprise IT, public sector compute.
Watch out for: thinner cloud on-ramps in some locations, fewer carrier choices than in the capital, variable provider quality.
A newer category is forming around sites with abundant power, often near former industrial assets, retired generation infrastructure, or planned renewable build-outs. These locations have attracted hyperscale and neocloud investment because they solve the one problem the capital cluster cannot at scale: tens or hundreds of megawatts of available grid capacity.
If a deployment is power-hungry, latency-tolerant, and built around AI or large-scale batch compute, the cost-per-kilowatt comparison often favours these hubs. The trade-off is that they tend to be less mature on fibre redundancy and ecosystem depth, although that is changing as operators commit long-term capital. The North West sits firmly in this category, and our Manchester colocation environment is purpose-built for high-density compute with N+1 power and cooling, immersion pod support, and headroom for sustained AI and HPC workloads.
Best for: large-scale AI infrastructure, training clusters, neocloud GPU rental, hyperscale tenants, sustainability-led deployments.
Watch out for: longer build timelines, fewer immediate cross-connect options, dependency on planned grid upgrades.
The fourth category is less about geography and more about distribution. Edge and micro deployments place small amounts of compute close to the workload, whether that is a manufacturing site, a logistics hub, an NHS trust, or a high-density urban area. We deploy these on demand within existing commercial property footprints, which keeps lead times short.
Edge is a complement to a primary colocation site rather than a replacement. The pattern we see most often is a core deployment in a regional hub paired with edge nodes wherever real-time response matters.
Best for: IoT processing, real-time analytics, content caching, low-latency local services, and regulated workloads needing data locality.
Watch out for: operational complexity at scale, the temptation to over-distribute compute that would be cheaper to centralise.
How to compare hubs honestly
Marketing materials make every facility sound much the same. When evaluating options, the questions below tend to separate the genuinely suitable from the broadly adequate.
A note on pricing: traditional colocation contracts are priced by the rack, which means consolidated, high-density deployments effectively subsidise low-density ones. Power-based billing reverses that incentive, which is why density-led operators have increasingly moved to it.
Why has cooling design become a hub-level question
When evaluating a facility on cooling, the underlying question is whether the hall was actually designed for the hardware you want to run. Many older UK halls were built around 4kW to 10kW air-cooled racks. Modern GPU servers can pull several times that, and the gap widens with each hardware refresh.
Carbon-Z deploys Immersion Cooling and Direct-to-Chip (DTC) cooling alongside conventional air-cooled racks across all three of our UK sites. Our immersion environments, built in partnership with Midas Immersion Cooling, support pod densities from 25kW to 100kW. The principle is straightforward: submerging hardware in a dielectric fluid removes heat at the source rather than relying on airflow, which is the relevant capability when sustained density rather than peak density is the design goal.
The reason this matters at the hub-selection stage is that not every facility can host immersion or DTC, even where the marketing suggests otherwise. Retrofitting these systems into a hall built for traditional air cooling involves structural, plumbing, and floor-loading work that can be more expensive than starting in a purpose-built environment.
Explore our Immersion Cooling service to see how it fits into a wider hub strategy. If you would prefer a hardware-specific review, book a free power assessment, and we will share what we see.
Three patterns worth flagging during planning
These come up often enough in industry discussion that they are worth listing explicitly.
Defaulting to the capital because the previous deployment lived there. This can lock teams into longer wait times and higher costs than a regional or power-led hub would impose, with no measurable latency benefit for the workload in question.
Under-specifying cooling for AI hardware. Air-cooled racks rated at 8 to 10kW will not sustain modern GPU densities at full load. Planning the cooling architecture before the hardware refresh, rather than after, avoids significant rework.
Signing rigid contracts for changing workloads. AI hardware roadmaps move quickly. The contract should allow you to start in air-cooled staging racks and transition to immersion or direct-to-chip environments as the hardware evolves, without renegotiation.
How we would approach the decision
We suggest starting with the workload profile rather than the map. Latency-sensitive financial or cloud-adjacent workloads still point towards the capital cluster, where our London facility sits. AI training, rendering, and HPC tend to favour a regional or power-led hub with purpose-built liquid cooling - patterns we see play out in both Swindon and Manchester . Resilience workloads benefit from geographic separation, and edge nodes belong wherever the data is generated.
That sequence sounds simple, but is often inverted in practice, with the location chosen first and the workload squeezed in afterwards. If you would like a second opinion on the fit between your hardware and a particular environment, our engineers are available for a technical conversation.
Carbon-Z is a UK data centre operator certified to ISO 14001, ISO 27001, and ISO 45001.