London still hosts the biggest concentration of UK data centre capacity, but the centre of gravity is starting to move. AI workloads are changing the infrastructure maths, pushing power, space, and planning considerations up the decision list. That is exactly where regional locations start to look like the sensible option.
Government data shows how concentrated the market remains: as of Autumn 2024, London is estimated at 1,048MW of colocation it load. Compare that with 44MW in the east of England, 17MW in the northeast and 30MW in Scotland. The gap is huge, yet it is not a permanent advantage.
Many AI use cases do not need to sit inside London. They need predictable performance, secure hosting, strong connectivity, and a viable path to scaling. That shifts the location question away from postcode gravity and towards energy access, latency requirements and build timelines.
Reuters, citing Barbour ABI, reported UK data centre spending projected to rise to £10 billion a year by 2029, up from £1.75 billion in 2024, driven largely by AI capability needs, with close to 100 new projects in progress. Even if the exact mix of projects shifts, the direction is clear: more sites, more power, more urgency.
Here, Mark Lewis, Chief Marketing Officer at Pulsant, offers his insights.
Why London’s dominance is starting to fray
London’s traditional edge has been proximity to finance, dense connectivity, and established ecosystems in places like the M4 corridor. The constraint is no longer demand. The constraint is delivery.
National planning and policy bodies have started to address data centres as an infrastructure class with real trade-offs. A UK parliament briefing notes TechUK’s view that west London is “beginning to reach saturation point”, citing land and grid capacity constraints. London’s own institutions have been blunt about electricity demand pressures too, linking data centre growth to wider constraints in the capital.
AI workloads intensify the issue because they change the shape of demand. Training, inference and model fine-tuning pull large, sustained loads. That makes grid access, connection queues, and substation availability the gating factors. The sites that win tend to be the ones that can secure power with fewer unknowns.
Lewis comments: “A lot of organisations still default to London in early planning, then run into delivery friction. AI has made the power question impossible to defer. The smart move is to start with the workload, the latency tolerance, and the power profile, then choose the geography that can deliver on those constraints.”
AI growth zones and why they matter outside London
The UK government’s AI growth zones programme is effectively a signal that regional build-out is now part of national industrial policy. The stated aim is to unlock investment in AI-enabled data centres and support infrastructure by improving access to power and providing planning support. That framing matters because it moves the conversation from “regional capacity is nice to have” to “regional capacity is a planned route to delivery”.
The policy detail is explicitly tied to power system efficiency. The government’s “Delivering AI growth zones” paper sets out potential electricity cost discounts linked to location, giving a north east example of up to £14/MWh for a 500MW data centre. That is not a marketing line, it is a system-level argument: place large new loads nearer to where generation and network capacity make the whole system cheaper to operate.
Lewis notes: “Growth zones make one point unmissable. Power and planning are now first-order design constraints for ai infrastructure. Regional sites can move faster on both, and that changes the investment case. New sites may be years away, but there’s plenty of regional capacity available right now. Existing regional operations are a vital part of the mix.”
The “regional edge” angle is becoming the smarter play
For inference, retrieval, and real-time decisioning, distance starts to matter again. Latency budgets can be tight, especially where AI is embedded into customer-facing services, industrial control, fraud detection, or near-real-time analytics, which increases the value of regional data centres and edge-adjacent capacity.
This is not a call to abandon London, but it is a recognition that a single-centre strategy looks brittle as power becomes contested and AI load grows. Hybrid architectures are more common now: core platforms in established hubs, paired with regional capacity that keeps data closer to users, sites, or operational systems.
With UK regional data centres and high-performance connectivity options, Pulsant can support architectures where workloads sit closer to where they are generated or consumed, plus provide resilient interconnect paths into public Cloud and partner ecosystems.
Lewis adds: “A lot of AI projects stall because the infrastructure plan arrives too late. If you start with regional capacity in mind, you can place data, compute and connectivity in a way that keeps latency predictable, reduces exposure to one grid area and gives you options as your workload grows.”
Grid reform adds urgency to location choices
Energy regulation is moving in parallel. Ofgem launched a consultation in February 2026 on overhauling grid connection rules, with data centre growth as a focal point. That matters because connection queues and allocation criteria can reset the competitive picture. Projects that look viable on paper can slip if grid access cannot be secured on the required timeline.
Future-proofing AI infrastructure means making choices that survive real constraints: power access, build time, resilience, latency, operational control, and regulatory pressure. Regional capacity increases optionality. It gives organisations a way to scale without betting everything on one congested zone. The next wave of investment looks more distributed, driven by ai load and a policy environment that is explicitly trying to pull capacity into regions.
Lewis concludes: “The organisations that do best with AI infrastructure planning are the ones that treat geography as part of the architecture. Regional data centres are not a fallback. For many workloads, they are the route to predictable delivery and a cleaner scaling path.”
