The UK’s existing stock of datacentres is smaller than suggested by most of the numbers in circulation, and so is the carbon it emits today. A fresh count by Computer Weekly puts the running estate at about 1.5GW (gigawatts) across roughly 90 to 110 sites, most of them clustered in London and the M4 corridor.
On today’s electricity grid, that estate emits about 1.5 million tonnes of carbon dioxide equivalent (MtCO2e) a year. That is roughly the annual footprint of UK domestic aviation, or a third of the country’s cement sector.
But the committed pipeline of datacentres in the UK planning system – projects that have entered planning but are not yet built – stands at 173 projects and more than 14.6GW of capacity. That’s several times the size of the estate running today, as Computer Weekly has reported, and the carbon that pipeline would add is a different order of magnitude.
In this piece we estimate the carbon footprint of the UK’s datacentres three ways: what they emit in operation, what is locked into their construction and equipment, and how these compare with the rest of the UK economy. We also show why the answer depends so heavily on two things the industry would rather not dwell on – where a datacentre is located and how it gets its power.
The UK datacentre estate today
The 1.5GW running today sounds large but isn’t the full picture. We apply two conversions that most headline numbers skip.
The first is utilisation. Rated capacity is a ceiling, not a constant draw. The Uptime Institute finds that more than half of datacentre operators do not even track server utilisation, and industry benchmarks put typical enterprise servers at 10% to 20% of their compute capacity. A datacentre drawing every watt of its rating around the clock is a rarity. We assume a facility draws 60% of its rated capacity on average, with a range of 45% to 70% to reflect the uncertainty.
The second is power usage effectiveness (PUE), the ratio of a facility’s total energy to the energy its IT equipment draws. The Uptime Institute’s global average is 1.5 to 1.54, meaning half as much again in cooling and losses. We use 1.3 to 1.7 for the existing stock and 1.15 to 1.4 for new build.
Applying those to 1.5GW, on a UK grid that averaged 125g of CO2e per kilowatt-hour in 2024, gives an operational footprint of about 1.5 million tonnes a year – with a plausible spread of 0.8 to 2.5 million depending on the assumptions. Count the carbon of the grid where each site actually sits, rather than the national average, and the figure rises to about 1.9 million tonnes, because the London and M4 corridor’s Slough and Reading hubs sit on a dirtier slice of the grid than the UK as a whole.
The grid is not one number
That last point indicates a potential hole in datacentre carbon accounting.
Namely, the carbon intensity of UK electricity is not a constant. Across 2024 it swung, half-hour by half-hour, from 14g to 303g of CO2e per kilowatt-hour. The annual average was 125g, but the average hides almost everything. On a calm, cold evening in November, the grid burned gas and the intensity climbed past 170g. On a bright, windy August afternoon, it fell below 40g. The same datacentre, drawing the same power, emits several times more carbon at one hour of the year than another.
There is a second distinction. The average intensity describes the grid as it is. But when a new datacentre connects, the grid does not draw a little more of everything – it dispatches more of whichever generation source sits on the margin, and in Britain that is usually a gas-fired power station, at roughly 400g of CO2e per kilowatt-hour. So, the additional carbon caused by new capacity is closer to three times the average-based figure. The existing estate can fairly be measured against the average grid, but the pipeline cannot.
The carbon you build
Even that understates the footprint, because it counts only what the datacentre burns. A datacentre is also a shed full of servers on a concrete base, and all this carries a carbon cost before a single watt is drawn.
We estimate the embodied carbon of the running estate – the shell, the IT equipment and the electrical and mechanical plant – at roughly 3.5 million tonnes, and of the full pipeline at about 31 million tonnes. Those are one-off figures, though. Annualised over the life of the asset – 40 years for a building, but only about four years for a server, which is refreshed repeatedly – the embodied carbon works out at roughly 40% to 50% of the operational figure.
In other words, close to half of a UK datacentre’s carbon is locked in before it is switched on, and the biggest single contributor is the equipment, which is replaced every few years.
The pipeline problem
Apply the same arithmetic to the committed 14.6GW and the picture changes scale.
On today’s average grid, the pipeline would add about 12 million tonnes of CO2e a year – roughly eight times the footprint of the estate running now. On the marginal, gas-based figure, it would add about 38 million tonnes a year. The true figure, as the grid decarbonises over the coming decade, will sit somewhere between the two.
That is not the end of the build-out, either. The 14.6GW figure counts only projects carrying a published capacity. Of the 173 projects, 28 carry none, so the real committed figure is higher, while a further 4.4GW sits in pre-planning – announced but not yet in the system. The pace matters because the government’s own 2030 capacity ambitions already look hard to reach.
Location is a lottery
Where a datacentre is built is the single biggest determinant of its carbon — and it is decided by grid connections and demand, not by carbon. The same datacentre lands on 22g or 254g per kilowatt-hour depending on where it happens to connect.
In North East England the grid ran at about 22g of CO2e per kilowatt-hour in 2024, and in South Scotland at about 25g, thanks to wind, hydro and nuclear. In South Wales it ran at 254g, and in South-West England and East Midlands above 200g, where gas does more of the work.
Weighted by where the capacity actually sits, the existing estate draws a grid of about 152g per kilowatt-hour – dirtier than the UK average, because of its London and M4 concentration. The committed pipeline, by contrast, draws about 118g per kilowatt-hour. The reason is that a meaningful slice of it – South Scotland’s 2.4GW and North-East England’s 1.6GW, among the largest schemes in the country – is heading to the cleanest grid in Britain.
That drift is real, but it is modest and roughly a third of the pipeline sits in the dirtiest regions – South England, Yorkshire and South Wales, at 143g to 254g per kilowatt-hour. The UK’s datacentre locations are chosen for reasons that have nothing to do with carbon. Namely, where the grid connection is available, where the land is, where the demand sits. So, the carbon footprint is usually a by-product of a decision made on other grounds. The operator is chasing grid access, and the clean/dirty outcome comes out however it comes out.
Where is the clean energy?
One might expect the planning system to record how each of these projects intends to power itself. It mostly does not.
A handful of the largest schemes do name an energy arrangement of sorts. According to the carbon footprint statement in the planning application for Elsham Wolds, a 1GW campus in North Lincolnshire, the region is well-suited to such a development because of its proximity to clean electricity generation centres.
It said: “The Humber accounts for approximately 20% of the UK’s total electricity generation and 33% of the nation’s offshore wind capacity.” But the planning application went on to apparently miscalculate the likely carbon footprint.
Meanwhile, the former Ravenscraig steelworks site in Scotland proposes a 550MW datacentre alongside 650MW of battery storage. Cottam, a 1GW campus in Nottinghamshire, promises up to 12 small nuclear reactors – though it has not yet entered planning. Microsoft’s Skelton Grange campus in Leeds will run on gas fuel cells until 2037.
But these are the exceptions. For most of the 173 projects in the pipeline, planning documents say nothing about clean energy. Any renewable claims rest on power purchase agreements and certificates that never touch the planning system – and that gap is where greenwashing can live. A datacentre that buys renewable certificates in Yorkshire still physically draws from the same gas-burning grid when the wind drops.
It’s not that potential for clean energy doesn’t exist, however. The Scottish government is seeking views on raising its offshore wind ambition to up to 40GW by 2040 – enough to power the equivalent of about 45 million homes a year, up from an ambition of 8GW to 11GW by 2030. That is why battery storage is springing up alongside the northern schemes, such as Ravenscraig’s 650MW (megawatts).
But the datacentre industry is betting on a cleaner grid to come rather than documenting a clean supply that exists today.
How that compares
If we set the datacentre numbers against the rest of the economy, the scale of things becomes clear.
Combining operational and annualised embodied carbon, the running estate accounts for about 2.2 million tonnes of CO2e a year. The committed pipeline would add about 18 million tonnes a year on the average grid, or 44 million tonnes on the “worst case” gas figure.
Against the UK’s 2024 emissions of 373 million tonnes, the pipeline’s average figure is about 2.6 times the whole cement sector (roughly seven million tonnes) and larger than all household electricity use (14.7 million tonnes). On the worst-case figure, it approaches the size of the entire industry sector (46.5 million tonnes) and about 12% of all UK emissions. At the UK’s per-person footprint of roughly 5.5 tonnes a year, the pipeline’s average figure is the equivalent of about 3.3 million people’s entire annual emissions.
There are caveats here. These datacentre figures include embodied carbon, while the sector figures are territorial production emissions that do not. And datacentres are not a sector in the national figures at all – their electricity sits inside the “business” category, and its generation inside “electricity supply”. The comparison is illustrative, not additive. It shows scale, but does not claim the datacentres are additional to everything else.
The methodology behind the data
The figures rest on two established datasets. The existing estate comes from Computer Weekly’s count of the non-domestic Energy Performance Certificate registers for England, Wales and Scotland. Each site was verified by hand and its capacity estimated from published figures or floor area. The pipeline comes from Barbour ABI construction data, filtered to projects that have entered planning but are not yet built.
Operational carbon is capacity x 8,760 hours x utilisation time x PUE x grid carbon intensity. The grid intensity is the National Energy System Operator’s half-hourly actual figure for 2024, which averaged 125g of CO2e per kilowatt-hour, with regional values used for the location-weighted estimates. The marginal figure uses a gas-fired intensity of 400g per kilowatt-hour.
Embodied carbon applies order-of-magnitude estimates – roughly 300kg to 500kg of CO2e per square metre for the building shell (LETI best-practice benchmarks), 800 to 2,500 tonnes per megawatt for IT equipment (derived from Dell, HPE and Lenovo product carbon footprints), and 200 to 600 tonnes per megawatt for electrical and mechanical plant – annualised over 40, four and 15 years. The absolute figures move with those factors, but the ratio of embodied to operational, roughly 40% to 50%, does not.
Every capacity figure is an estimate of potential demand, not metered load. The 14.6GW pipeline counts only projects with a published capacity (or data from which it is possible to derive capacity), so it is a lower bound. Regional grid intensities are forecasts rather than metered actuals, because the operator does not publish regional actuals. The sector comparison uses the Department for Energy Security and Net Zero’s final 2024 territorial emissions.
