The UK wants to lead in AI – but its data centres are running low on the one resource they cannot function without.
Government ministers have become accustomed to calling data centres “the engines of modern life”, and they are right to. Everything from routine cloud storage to the frontiers of AI rely on a rapidly expanding number of sites running flawlessly, every second of every day.
Equinix's £3.9bn purchase of an 85-acre site in South Mimms, announced in October, was both a vote of confidence and a stark reminder of the scale of what's required to keep the UK's technical ambitions on track.
This was a rare thing: a site with planning secured and a confirmed power connection. Equinix deserves credit; in a market beset by constraints, it moved decisively with the clarity and speed that hyperscale operators increasingly need.
However, the power demands of the site speak to the central challenge facing our digital future. Even the most sophisticated engine is useless without fuel, and a lack of power risks hobbling many hyperscalers as demand accelerates.
The UK’s electricity grid was never designed for the kind of always on, high density computing that now underpins artificial intelligence, cloud services and our minute-to-minute digital interactions. Substations feeding London, the core of the UK’s hyperscale market, are at or near capacity. Power for new data centre campuses often arrives five to ten years too late.
The result is a queue that has swollen to 125 GW of demand – more than double the UK’s peak winter load. For years, individuals could apply for power allocations without holding planning consent, owning a site, or even having an option on the land. These speculative placeholders, now widely referred to as “zombie connections”, have clogged the system. Until recently, the connection process has been unable to distinguish between a targeted hyperscaler preparing a multi billion pound campus and a party making a speculative bet.
Thankfully, Ofgem and the National Energy System Operator (NESO) have already started to reshape the connections regime around credibility rather than chronology. Under the new Gate 2 process, projects are no longer able to sit in the queue on the strength of a speculative application. Developers are required to evidence control of land, demonstrate project viability and show that the scale of their power request matches the site they are promoting.
In theory, this should begin to flush out zombie connections. NESO’s current review of the demand queue is the first step in that direction – the next six to twelve months will reveal where real power capacity exists.
For data centre developers, operators and hyperscalers, the question is not simply whether power is available – it is when. Most operators building at scale in the London market need connection dates no later than 2032.
This is where the nation’s internal rhythms collide. Planning and grid investment move on multi year, sometimes decade long horizons. Hyperscale business models do not. Power promised for 2035 is just not good enough for most operators.
This mismatch is becoming the defining friction in the UK market. Land is hard to find, but land with power inside the right deployment window is harder still. Until energisation dates and platform timelines are brought into alignment, Britain’s most important digital investments will remain hostage to a slow, unpredictable process.
The government understandably wants to spread the load by fostering AI Growth Zones in Scotland and across the North of England, but geography alone cannot determine where hyperscale campuses go. Data centres obey the physics of latency, fibre routing and network architecture. London remains the UK’s digital epicentre because it is where cloud providers cluster and the country’s main financial, media and corporate workloads reside. It is right that data centres are considered Critical National Infrastructure (CNI), but planning rules could be relaxed further to ensure they can be built where they are needed most, rather than potentially being restricted to AI growth zones.
It is possible that demand becomes more geographically distributed – the future mix may well become more hybrid, with regional sites handling bulkier AI, or lower latency processing while London absorbs high intensity demand via the cloud and co-location. But even then, power costs would remain a barrier to the success of AI growth zones, because wholesale electricity costs are so much cheaper in competing locations like the Nordics and the Iberian Peninsula.
To be clear, the government and NESSO deserves credit for much of the reform now underway. Clearing out zombie connections, requiring evidence of land control and project maturity, and allowing schemes of more than 50mva to apply for planning consent directly from the Secretary of State are all steps in the right direction.
However, aspects of the policy landscape need better tailoring to the needs of the hyperscalers. Under Gate 2, for example, developers must control their land before they can apply for power, yet hyperscalers will not commit to land unless they know when power will arrive. Until this catch-22 is resolved , the system will depend on joint ventures or the rare approach that unlocked the Equinix deal: developers with the expertise and appetite to take early risk, assemble land, secure planning and establish a credible power pathway before selling a fully packaged site to an operator.
Sequencing is the next challenge: aligning power delivery with the timelines global operators actually build to. If the grid, planning system and investment cycles can finally move in step, sites like South Mimms will become the rule rather than the exception.
This article first appeared on Green Street News.
Click here to find out more about how we deliver confidential real estate advice, that best supports data centre operators and investors needs across EMEA.
