UK government admits AI data centres could emit 100 times more carbon than previously forecast

AI data centre emissions vastly underestimated, UK admits

The UK's AI data centres could be more than a hundred times more damaging to the climate than previously estimated, the government has announced. The Department for Science, Innovation and Technology this week corrected its forecast for greenhouse gas emissions from AI compute for the next decade, which had been based partly on work by the consultancy McKinsey. Last year the government said annual greenhouse gas emissions from AI compute would be at least 0.025 MtCO₂ -- equating to 0.25 MtCO₂ over the decade to 2035 -- compared to an estimate this week of at least 34 MtCO₂. AI data centres are now expected to account for between 0.9 and 3.4 per cent of the UK's carbon emissions in the decade to 2035. Last year the government said they would make up less than 0.05 per cent of the total. The forecasts for water consumption from AI have also risen. Tim Squirrell, head of strategy at Foxglove, a tech-focused campaign group, said that the government's legally binding commitment to hit net zero emissions by mid-century sat "awkwardly alongside its hell-for-leather embrace of a hyperscale AI data centre build-out". Data centres need round-the-clock reliable power even when the wind is not blowing and the sun is not shining. AI-enabled data centres globally have been particularly reliant on burning gas for energy, due to power-intensive processors that train and deploy more advanced models. The government has pledged to largely run the UK's power system on clean energy by the end of the decade. But it has also acknowledged that in practice this will mean retaining some gas-fired power. Squirrell accused the government of "nonsense magic beans thinking that you can have massive data centre growth without a corresponding increase in new polluting carbon emissions". Two non-profits, Carbon Brief and Foxglove, last month warned that the government was underplaying the climate fallout of data centre growth by underestimating the need for gas power. The government did not explain why earlier estimates needed such a significant correction. A person close to the DSIT attributed the changes to "routine review". For the original estimates, the government commissioned McKinsey to provide a 10-year projection of future demand and supply for AI computing. Cambridge Econometrics, a consultancy, then helped do the environmental analysis. The estimates were drawn from "a combination of academic literature, grey literature, and government projections", as well as assumptions about future scenarios where needed, a report said. It also warned of a potential shortage in AI compute over the next decade and made the case for government intervention to support AI growth. OpenAI put its flagship UK data centre project on hold earlier this month, blaming high energy costs and regulatory uncertainty. McKinsey and Cambridge Econometrics did not immediately respond to a request for comment.

AI data center emissions could be much higher than anticipated, new numbers show

Countries continue to invest in AI infrastructure, often misaligned with previous climate goals. Credit: Mario Tama / Staff / Getty Images News via Getty Images AI data centers across the pond could emit nearly one million more tons of carbon dioxide into the atmosphere than previously estimated, according to newly inflated numbers released by the UK's governing body. First spotted by Politico, UK officials quietly revised an initial range of estimates included in the country's "Compute Roadmap," a policy document outlining the UK's plan to scale its AI infrastructure and build a "world-class compute ecosystem." Previous versions of the document suggested the amount of energy used under current AI data center plans would result in a peak of 0.142 million metric tons (or 142,000 metric tons) of carbon emissions between 2025-2035. But 2025's numbers, and the country's AI goals, appear to have altered that projection, with the document now listing the top value at 123 million metric tons. That's 100 times more than before, the Guardian reported, and is equivalent to the emissions typically generated by 2.7 million people. Chi Onwurah, parliament member and chair of the body's science, innovation and technology committee, told Politico that the new numbers were concerning, demanding more transparency from UK leaders about the environmental impact of its AI plans. Multiple parliamentary groups are investigating AI's energy demands and fossil fuel consumption, as the country has pledged to reach net zero emissions by 2050. Increased energy use by AI data centers in the U.S. have also led to a rise in global emissions, according to recent environmental reports. Domestically, U.S. greenhouse gas emissions went up for the first time in two years following a steady decline. Concerns over the environmental impact of mass generative AI adoption aren't slowing down. Coalitions of environmental advocacy groups have called for federal moratoriums on data center builds. Communities nationwide have been fighting against local billion-dollar data center projects, even proposing state-level bans. Meanwhile, Big Tech's leaders and major AI investors have attempted to assuage climate concerns with promises to reduce and recycle water, use natural gas as an energy source, and offset energy costs.

UK departments at odds over energy demands of AI datacentres

Discrepancy in forecasts raises questions over government planning for net zero One vision of the UK's future involves a decarbonised economy powered by clean, renewable energy. Another involves making the UK an AI superpower. The government departments responsible for these two visions do not appear to have agreed on their numbers. The Department of Science, Innovation and Technology (DSIT) thinks AI datacentres will consume 6GW of electricity by 2030. The Department of Energy Security and Net Zero (DESNZ) appears to think they will use less than a tenth of that. Tim Squirrell, the head of strategy for the NGO Foxglove, said: "The government's cluelessness over the environmental impact of datacentres would be laughable, if it weren't so alarming." Cecilia Rikap, a researcher at University College London, said: "There are two ways to interpret this 'misalignment': either DESNZ and DSIT are incompetent, or there's some kind of magical thinking about AI and big tech. Either way, the episode uncovers how these corporations control not only the AI value chain, but also the UK government." DESNZ is responsible for the UK's carbon budget growth and delivery plan, which sets out how the government will reach its international climate targets. In January, Foxglove filed an environmental impact assessment request with the department, asking how it had incorporated AI datacentres into its projections for Britain's emissions. In response, DESNZ referred researchers to its broader forecasts for the energy use of Britain's "commercial services" sector, and said it did not hold separate projections for datacentre growth. The forecasts appear to project that the energy use of the entire sector will grow by 528MW between 2025 and 2030 - equivalent to adding the consumption of 1.7m homes by the end of the decade. This projection is 10 times lower than the amount of electricity the government has committed to AI datacentres as part of its UK compute roadmap. That policy paper, put forward by DSIT in 2025, sets out a "bold, long-term plan to transform our national compute ecosystem" by building AI datacentres. It adds: "We forecast that the UK will need at least 6GW of AI-capable datacentre capacity by 2030." This will come from multiple AI growth zones - hubs across the country where the government is attempting to attract investment into datacentres. Each would require at least 500MW of electricity - an amount only slightly less than DESNZ's forecast for the increase in energy usage of the entire commercial services sector. It is unclear how the discrepancy between the two departments' forecasts arose. But one day after the Guardian requested comment from DSIT and DESNZ, DSIT appears to have revised its figures published on its website for the total emissions of the AI datacentre sector, raising them more than a hundredfold. Originally, DSIT's projections for the carbon emissions of additional AI computing capacity were between 0.025m and 0.142m tonnes of carbon equivalent (MtCO₂) - below 0.05% of Britain's projected emissions. These figures were in an annexe to the compute roadmap. Earlier this year, that document was removed from the government website after Carbon Brief raised questions about the plausibility of the figures. On Thursday, after the Guardian asked about them, DSIT updated its numbers. In a statement posted online, it said: "The UK's cumulative 10-year greenhouse gas emissions from AI compute could range from 34 to 123 MtCO₂ - this is around 0.9-3.4% of the UK's projected total emissions over the 10-year period. "If successful, the UK's grid decarbonisation plans would help to reduce emissions from datacentres towards the bottom end of this range," it added. A spokesperson for DESNZ said: "datacentre emissions are factored into our modelling, including for carbon budget 7," adding: "The AI Energy Council is exploring opportunities to attract investment and support the development of clean power for datacentres." Carbon budget 7 is to be released this summer. A spokesperson for DSIT referred the Guardian to DESNZ.

Officials hugely underestimated impact of AI datacentres on UK carbon emissions

Revised figures increase fears about how the energy-intensive sites could worsen the climate emergency The UK government vastly underestimated the climate impact of artificial intelligence, it has emerged, after officials raised their estimate of carbon emissions from the technology by a factor of more than 100. According to new data quietly published this week, energy use by AI datacentres in the UK could cause the emission of up to 123m tonnes of carbon dioxide (MtCO₂) - about as much as generated by 2.7 million people - over the next 10 years. That latest figure replaces a previous estimate - since deleted - that claimed emissions would reach a maximum of 0.142 MtCO₂ in a single year. The revision comes amid increasing alarm at the carbon impact of AI and when calls for the world to reduce emissions to mitigate the climate emergency are becoming increasingly urgent. "We have a handful of years until our carbon budget is exhausted," said Patrick Galey, head of investigations for the Global Witness climate campaign. "To waste what little bandwidth we have left - when 750 million people worldwide still lack access to electricity - assisting some of the richest men ever to hone their plagiarism bots would be a historic idiocy that future generations are unlikely to forgive today's leaders for." The latest estimates were revealed in a revision to the UK "compute roadmap", which sets out the government's plan "to build a world-class compute ecosystem" for delivering artificial intelligence in the UK - a goal on which the government has staked its hopes for economic growth. However, AI datacentres require huge amounts of electricity to operate - much more than the usual datacentres that store online data - and most of that continues to be generated by fossil fuels. According to the Department for Science, Innovation and Technology's latest estimates, the carbon impact of the planned AI buildout could range from 34 to 123 MtCO₂ - about 0.9% to 3.4% of the UK's projected total emissions between 2025 and 2035. The lower range of the estimate would depend on greater efficiency in AI models and hardware, and faster decarbonisation of the UK's energy grid. Officials from the DSIT appear to have made the revision, first reported by Politico, after an investigation by Foxglove, an independent watchdog, and the Carbon Brief news site, pointed out they appeared to be a significant underestimate. Foxglove's head of strategy, Tim Squirrell, said: "The government has a legally binding commitment to reach net zero by 2050. This already sat awkwardly alongside its hell-for-leather embrace of a hyperscale AI datacentre buildout, which unchecked could double the electricity consumption of the entire country. "The situation has now been revealed to be much, much worse, given the fact the government doesn't seem to have done even the most basic arithmetic needed to measure the potential new carbon emissions of these datacentres." The government declined to comment on the record.