Information dive: Authorities 2030 datacentre capability targets look shaky
In July 2025, UK secretary of state for science, innovation and expertise Peter Kyle mentioned: “We forecast that the UK will want at the least 6GW of AI [artificial intelligence]-capable datacentre capability by 2030.”
As issues stand, that focus on will seemingly not be met, with UK datacentre capability on target to whole 4.9GW by 2030 given the present planning standing of many tasks.
Present UK datacentre capability is round 1.6GW. If we assume an enormous chunk of that isn’t “AI-capable” then 6GW of “AI-capable datacentre capability by 2030” appears a method off.
That determine – 4.9GW by 2030 – comes from the seemingly completion of datacentre tasks that presently have planning approval. It’s based mostly on analysis by Pc Weekly, with information from development pipeline trackers Barbour ABI, and for present operational datacentres, is predicated on authorities electrical energy efficiency certificates information.
On this article, we take a look at presently operational datacentre capability within the UK and the seemingly projected pipeline for the approaching years. We break that down by authorized and unapproved capability when it comes to planning purposes.
We additional take a look at the regional traits: the place datacentres at the moment are and the place they are going to be constructed within the subsequent decade within the UK. In future articles, we’ll take a better take a look at the federal government’s plans for AI datacentre capability and the way their targets could possibly be fulfilled.
Can the pipeline get to 6GW by 2030?
In accordance with the info, the UK has about 1.59GW of presently put in datacentre capability at just below 190 websites.
If we add present capability to that which is deliberate to finish by 2030 and which has planning consent, we get 4.9GW.
Whole pipeline, nonetheless, together with tasks that don’t but have planning consent, is 8.1GW. The entire of unapproved tasks deliberate to finish by 2030 is 6.2GW.
There’s nearly sufficient within the pipeline to realize 6GW by 2030. However that might require a rash of planning purposes to undergo in time to start out development in 2027, assuming a three-year construct interval.
Having mentioned all that, there should be some doubts about how a lot of that 6GW could be “AI-capable” datacentres.
How a lot capability will probably be AI-capable?
Many datacentres within the present capability base are previous and comparatively small.
About 14% (27/120MW) of 190 operational datacentres are at the least 10 years previous. That’s to not say they gained’t have been upgraded. However growing energy and cooling necessities over time imply they want extra of each to assist processing with every improve, and the variety of racks that may be hosted in the identical bodily footprint decreases. In the meantime, 42% (80/600MW) of present datacentres are greater than six years previous.
Additionally, most datacentres within the presently put in capability base are very small by present requirements. Based mostly on estimates within the information (see under) most (77%, or 147 of the 190) have capability of lower than 10MW.
Present AI rack energy draw maxes out at about 100kW. AI processing additionally wants liquid cooling. Energy draw of 1MW per rack is just two to 3 years away, in accordance with the Nvidia graphics processing unit (GPU) roadmap.
That’s why 18 out of 190 presently proposed datacentres – together with the giants within the authorities’s AI development zones – account for 68% of all deliberate capability by 2037. That’s 5.5GW of a complete pipeline of 8.1GW.
The majority of operational capability is centred on London and the M4 hall proper now, with simply over 850MW of the 1.59GW whole in that area. The M62 hall – taking in Manchester, Liverpool, Leeds and Hull – comes subsequent, with 471MW put in. Scotland and the north east of England have round 50MW of put in capability every.
Whereas it is vitally troublesome to get exact MW capacities for datacentres, we may be certain there isn’t a lot in the best way of greater than 100MW in anybody datacentre within the UK proper now. That’s all set to vary, with tasks deliberate of a number of hundred MW – 600MW at East Havering, 550MW at Ravenscraig, 500MW at Blyth, for instance – and 1GW at Elsham in North Lincolnshire.
Whereas essentially the most fast datacentre growth has been centred on the south east to this point, that’s set to vary, with the centre of gravity shifting north. The M62 hall has the most important pipeline, with 3.3GW projected. London and the M4 hall comes fairly a distance behind with about 2.2GW. Having mentioned that, after we strip out tasks that haven’t but acquired planning consent, the M62 area drops again to about 2.2GW and London to lower than 2GW.
As we’ve seen, nonetheless, that degree of constructing will improve the UK’s datacentre capability by round 4x, a lot of which can have come on stream by the primary yr or two of the following decade.
Maybe worrying for Scotland, nonetheless, is that the huge bulk of deliberate MW there doesn’t but have planning consent. Nonetheless, there are solely 4 datacentres deliberate in Scotland within the Barbour ABI information, and two of them account for greater than 99% of that (see chart: UK datacentres: Challenge scale distribution).
Each tasks are within the public session part, with the most important of those – the 550MW former Ravenscraig steelworks web site – being thought of as a government-backed AI development zone.
By 2034, London and the M4 hall will nonetheless be residence to essentially the most UK datacentre capability, however with the M62 area catching up. Actually, as issues stand when it comes to approvals, the Hull-Manchester-North Wales axis is about to overtake the south east by the mid-2030s.
If the chart (UK datacentre capability 2026–2037) reveals the M62 lagging a bit of, that’s as a result of it doesn’t present the staged addition of capability at Elsham Tech Park because it builds in the direction of its full 1GW by 2037.
The place the info comes from
Barbour ABI collates planning software information from UK native authorities for development trade prospects. We will use it to see the datacentre pipeline. Whereas the info is extraordinarily helpful, planning software data gives challenges for many who wish to monitor datacentre tasks.
As a result of there is no such thing as a requirement in planning purposes to explicitly label a mission as a datacentre, we estimated MW capability for a lot of websites listed based mostly on the connection of ground space and megawatts in a pattern the place each values are recognized.
We constructed estimates of present put in datacentre capability by monitoring down seemingly datacentres in UK authorities vitality efficiency certificates registers (together with for Scotland).
Datacentres use quite a lot of electrical energy, so we filtered for a “major vitality worth” of greater than 1,000kWh per m per yr – 3-5x the usual industrial common – and cross referenced that with a spread of traits that included ground space, essential heating gasoline (to rule out gasoline use in largely non-office buildings) and recognized key phrases present in descriptions of datacentres.
That supplied 190 seemingly present datacentre websites and allowed an estimation of MW capability.
Any megawatt figures are prone to be an underestimate as, over time, the megawatt draw of GPUs will increase for a similar bodily footprint.
