KEY TAKEAWAYS

• The UK data centre market remains structurally attractive, supported by hyperscale cloud, AI and resilient enterprise demand. The constraint is not appetite, but the availability of genuinely deliverable capacity.
  
  
• London continues to dominate because its fibre density, cloud ecosystems and availability zones are difficult to replicate. However, that same concentration intensifies pressure on power, land and delivery timelines.
  
  
• Supply and demand remain exceptionally tight. New deliveries are being rapidly absorbed, leaving the market increasingly reliant on pre-commitment, while the vacancy rate has fallen from 27% in 2016 to 8% in Q1 2026.
  
  
• Supply is expanding on paper, but much of the pipeline remains early-stage and uncertain. Pre-letting has become a core feature of the market, with future capacity increasingly committed well before completion.
  
  
• Power is now the primary gatekeeper of growth. Grid reform should improve queue discipline, but it will not eliminate the underlying capacity scarcity or the long lead times required for reinforcement.
  
  
• Energy costs also matter more than before. In a European context, the UK remains a relatively expensive power market, which weakens its competitiveness for the most energy-intensive deployments.
  
  
• Land strategy is no longer just about the site itself, but about securing the wider corridor of rights needed to make development possible. Third-party land consents for cable routes, pipework and access are increasingly part of the critical path.
  
  
• Planning conditions have become more supportive, helped by stronger policy recognition of data centres and broader consenting routes. Even so, consent is increasingly conditional on schemes demonstrating credible mitigation and locally visible benefits.
  
  
• Construction has become a major execution filter rather than a straightforward delivery step. Rising costs, specialist labour shortages and long lead times for key equipment are widening the gap between consented projects and operational stock.
  
  
• Social impact is now a material part of viability. Developers that can translate national strategic value into local benefits, such as skills, partnerships or heat reuse, are likely to be better placed to secure support and reduce planning risk.
  
  
• Investment appetite remains strong, but capital is becoming more selective and more infrastructure-led in its underwriting. Value is increasingly concentrated in power-secured land, operational platforms and projects with a credible route through planning, energisation and delivery.
  
  
• Looking ahead, the UK is unlikely to lack demand, capital or strategic relevance. The decisive question is which projects can convert those advantages into deliverable capacity in a more constrained, regulated and execution-sensitive environment.

Training on paper; inference in reality

London leads the market, yet grid constraints will shape what comes next.

Over the past two years, the UK data centre market has entered a new phase, driven less by traditional enterprise cloud adoption and increasingly by artificial intelligence workloads. Hyperscale cloud, AI native “neocloud” providers, and sovereign compute requirements are converging, creating unprecedented demand for high-density, power-intensive infrastructure. Nowhere is this more visible than in and around London, which remains Europe’s dominant data centre hub, accounting for more than 80% of UK supply and acting as the primary landing zone for cloud availability zones, AI platforms and interconnection ecosystems.

AI is fundamentally reshaping demand. Large-scale training workloads require concentrated campuses with thousands of GPUs, extreme rack densities and access to tens or even hundreds of megawatts of power. This has given rise to a new category of infrastructure providers, often referred to as neoclouds, that offer GPU-as-a-service platforms optimised for AI training and inference. These players are now competing directly with hyperscalers for capacity, accelerating take-up and pushing facilities to technical limits that were unthinkable only a few years ago.

However, while announced investment and planning pipelines suggest explosive growth, structural constraints are becoming decisive. Power availability, grid connection delays, planning friction and sustainability pressures are no longer secondary risks; they are now the primary filters determining which projects will actually be delivered. Ofgem data shows that proposed UK data centre projects are collectively requesting around 50GW of grid capacity, more than the country’s current peak electricity demand, making it mathematically impossible for all schemes to proceed as planned.

This tension raises a critical question for the UK market: are we overbuilding AI training capacity on paper while underestimating the longer-term, more distributed growth of inference-driven data centres?

Evidence increasingly suggests that while a limited number of very large AI training campuses will be built, large training clusters are also more likely to reach saturation once a handful of frontier-model builders and hyperscalers have secured the scarce combination of land, cooling and 50–200MW+ grid connections they require. Notably, many of the largest training campuses are being developed in the United States, where the frontier AI ecosystem is most concentrated and where hyperscalers and specialist operators are deploying multi-hundred-megawatt sites at unprecedented scale. This concentration can further narrow the addressable market for additional UK-based mega-campuses beyond a few flagship builds.

Training demand is inherently “lumpy”, and cyclical capacity is added in step-changes around new model generations. Then utilisation can soften as workloads shift to optimisation, fine-tuning, and the reuse of trained models, meaning additional campuses quickly compete for the same relatively narrow customer base.

By contrast, inference capacity, closer to users, businesses, and population centres, is likely to grow steadily for much longer. It scales with the breadth of AI adoption across sectors, needs lower latency and regional resilience, and can be deployed in smaller increments across existing colocation and edge footprints, tracking economic activity rather than model-training cycles.

In the UK, this points to sustained demand not only within the London metro, where financial services, media, enterprise SaaS and dense interconnection ecosystems create persistent low latency requirements, but also across emerging regional clusters such as the Midlands, the North West and Scotland, as organisations embed AI into day-to-day workflows. Crucially, inference build-out can often proceed in modular tranches, such as incremental 5–20MW deployments, and through retrofits or expansions of existing campuses, making it better aligned with the UK’s constrained grid-connection environment and planning realities than the next wave of ultra-large training sites.

>>  For more information, please contact our EMEA Data Centre Advisory team

_{Read the articles within Spotlight: UK Data Centre Building Capacity below.}