The energy Gods giveth: the world (excluding China) added 124 gigawatts (GW) of solar power to global capacity in 2025. And the energy Gods taketh away: 100GW of new data centres are planned between 2026 and 2030. In truth, there are no Gods doing this—just us. Dumb old us.
Globally, the amount of electricity consumed by data centres is projected by the International Energy Agency (IEA) to more than double by 2030, adding an amount equivalent to the current total electricity use of Japan. Investment bank Goldman Sachs calls this “an arms race”, forecasting that global power demand from data centres will increase 50 per cent by as soon as 2027.
Data centres are, of course, not bad things. They handle the world’s internet traffic, and are as much critical infrastructure today as sewage plants and electricity pylons. The trouble is that what’s driving the current boom isn’t your average reading habits, email and spreadsheets. It’s AI, which is making demand go off the charts.
“The AI-dedicated data center is an emerging class of infrastructure… designed for the unique properties of AI workloads,” writes Goldman Sachs. The IEA informs us, astonishingly, that an AI data centre is 10 times more capital-intensive than an aluminium smelter (of which there is only one left in the UK by the way, due to the high cost).
Consultancy firm McKinsey projects that by 2030 data centres will require $6.7 trillion worldwide to keep pace with the demand for computing power, of which AI-specific data centres will account for 70 per cent. Are we really going to fry the world for AI slop?
Mihir Nandkeolyar, a specialist in data centres at building services company Johnson Controls, tells me the IT sector is “building artificial intelligence at a massive scale. And the chips that are used, GPUs, in these types of servers and [AI] training functions, are a step change in power consumption.” While a traditional CPU might consume about 200 watts, a rack of GPUs in an AI data centre might consume 2,000 watts or more, a ten-times increase.
This is leading to a rapid buildout of “hyperscale” AI data centres, says Nandkeolyar, largely for “developing AI platforms in large language models (LLMs).” Examples include Amazon’s Project Rainier and Meta’s $10bn AI hub in Louisiana.
Data centres used to be “on the scale of 30 to 100 megawatts” but are now “up to 3GW in size”, says Nandkeolyar. Such energy demand dwarfs existing energy sources: a large power plant might generate one gigawatt; the Hoover Dam produces around 1.3GW of hydropower; one of the largest solar farms in the world, Anhui Fuyang in China, made up of 1.2m solar panels, generates just 650MW in peak sunshine.
This demand from data centres is increasingly eating into the gains made from renewable energy. Planned data centres in Scotland would demand 2GW-3GW of power—Scotland’s entire national peak winter demand is 4GW—effectively using up all of the nation’s offshore wind power. In Ireland, data centres consumed 22 per cent of national electricity in 2024.
In the UK, a significant rise in annual carbon emissions could result from data centres planned or under construction, potentially blowing the country’s carbon budget, according to a report by charity Global Action Plan and Foxglove, a non-profit. If just the ten largest data centre proposals go ahead it would generate up to 2.7m tonnes of CO2 emissions, equivalent to the carbon sequestered by 44m tree saplings grown for 10 years.
Water consumption has also ramped up with AI data centres, which are so big that they require liquid cooling. Large data centres are thought to use five million gallons of water a day (equivalent to a town of 50,000 people), though Nandkeolyar estimates that some could easily use multiples more. He notes that closed-loop systems that don’t evaporate the water are now being established. Even an average data centre that stores your phone pics can consume 300,000 gallons of water each day.
In the 2025 film Eddington, starring Joaquin Phoenix and set in 2020 against the backdrop of tensions over pandemic masks and Black Lives Matter protests, a small town in New Mexico descends into bloody civil war. Meanwhile, a proposed AI data centre calmly goes ahead. After most of the protagonists have been killed or maimed, the end credits roll over the eerie, finished data centre standing in the middle of the desert, lit into the night, an Ozymandias for our times.
As director Ari Aster tried to explain to anyone who cared to listen (the film flopped), the data centre was “absolutely central to the film’s point”. He told Time: “We begin with the promise of it coming, and we end with it being achieved… all of those stories and all of these characters are now just training data.”
Growth in this sector is unrelenting. TikTok is investing in a $10bn data centre in Brazil, despite indigenous people protesting that their right to consultation—guaranteed under an international agreement—was ignored. Another data centre has been proposed by the energy firm Open Origin on an historic burial site sacred to the native peoples of La Junta in West Texas. Productive farmland is being snapped up by developers across the US, writes agriculture journalist Katherine Albertson, with farmers “being told to name their price”.
Such untrammelled growth isn’t just gobbling up renewables gains. It is also reviving previously mothballed fossil fuel energy sites. Data centres have nearly tripled the demand for gas-fired power in the US in the past year alone, according to findings from Global Energy Monitor, a non-profit that tracks energy projects. Elon Musk’s xAI has just won approval to run 41 new methane gas turbines at its Colossus 2 data centre in Mississippi, despite local outrage.
So now, the fightback. People are actually watching Eddington. A poll for Heatmap News in February found that public support for data centres in the US is now minus 24 per cent, down from plus 2 per cent as recently as September. Some have compared the speed of this shift to the public’ turn against nuclear power after the Three Mile Island accident in 1979. At least 25 data centre projects (totalling 4.7GW of demand) were cancelled in 2025 due to opposition.
Meta shelved a data centre in the Netherlands set to be the biggest in the country, following resistance. Last year, a court in Chile suspended a Google data centre after locals discovered it would extract more than 7bn litres of water annually. A city council in Arizona rejected a proposal for an Amazon data centre following public outcry. In little old England, the government last year overruled community opposition to grant permission for a hyperscale datacentre on greenbelt land in Buckinghamshire, in line with Labour’s AI-growth pledge. However, it has since been forced to U-turn, admitting it made an “error” in pushing ahead without an environmental impact assessment.
Perhaps the Eddington vision of the future isn’t inevitable. In February, New Jersey City Council voted to cancel plans to construct an AI data centre and instead build a new public park where the 27,000-square foot facility would have stood. “We say a big ‘fuck you’ to Big Tech!” local organiser Ben Dziobek shouted defiantly to the crowd waiting in the snow outside the council meeting. “We say a big ‘fuck you’ to private equity! It’s time to build communities, not data centres.” Dziobek added that the fight against data centres will continue all over New Jersey—and, realistically, people who want to stop them are going to have to fight all over the world.
