AI revolution could generate emissions hundreds of times larger than previously thought

UK government quietly revises data centre emissions projections

The UK government has quietly revised its estimate of carbon emissions created by AI data centres by a factor of more than a hundred. In updated documents published last week as part of the Government’s ‘Compute Roadmap’, it was claimed that greenhouse gas emissions from AI could reach between 34 and 123 million metric tons of CO2 (MtCO2), or as much as 3.4 percent of the UK’s total emissions by 2035. Notably, 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.

This contrasts starkly with previous estimates released in July 2025 by the Department for Science, Innovation and Technology (DSIT) which suggested that emissions from future data centres would be minimal. Their analysis found that emissions would reach a maximum of 0.142m tonnes of CO2 in 2035, equivalent to less than 0.05% of national emissions[i].

In addition, projections for water consumption from AI data centres have also been increased, growing from 1 trillion cubic meters through 2035, to an annual range between 0.1-0.5 trillion cubic metres (depending on demand and decarbonisation assumptions)[ii]. Government documentation states that this is primarily driven by direct water consumption from cooling, compared to indirect water consumption used in electricity generation to power data centres. For context, total UK water abstraction is around 6.5–7 billion cubic metres per year[iii].

Analysis warns that AI data centre growth risks climate targets

It appears that the newly revised figures have been released following independent analysis which flagged the significant underestimate in AI emissions. Last month a study by Carbon Brief and NGO Foxglove warned that emissions from new AI data centres could be hundreds of times higher than government estimates.

The UK currently ranks second in the world behind the USA in terms of the number of data centres per country[iv], with roughly 1.8 gigawatts (GW) of facilities consuming more than 2% of national electricity[v] as of 2024. According to Ofgem there is already a financial commitment from companies to invest in 71 new data centres which if built, would require around 20GW of electricity.

Carbon Brief argued that emissions from powering data centres could be far higher than the government figures from 2025 suggested if even a small amount of the electricity required is generated by burning gas. They noted that while data centres could run entirely on low-carbon electricity, some in the sector have argued that the government’s AI ambitions would require the UK to use more gas power, especially in the near-term.

Head of Strategy at Foxglove, Dr Tim Squirrel told Carbon Brief that the emissions figures released by DSIT last July were “nonsense”. He said: “The figures that DSIT projects wildly downplay data-centre emissions, even by the standards of the most optimistic energy transition scenario. There is no way that the amount of compute they anticipate can be built and produce the miniscule emissions they’re calculating.”[vi]

The updated figures have raised concerns that the UK’s climate goals could be under threat, with green groups questioning how the current Labour government’s desire to boost AI rollout aligns with its target of reaching clean power by 2030 and net zero by 2050.

Tim Bierley, campaign manager at campaign group Global Justice Now, told Politico: “With this quiet release the government has finally admitted that AI data centres are a climate catastrophe, sucking up unbelievable quantities of electricity which would force us to drive new fossil fuel use at the worst possible time. Expansion on this scale will drive a coach and horses through the UK’s climate goals and plans for the energy transition."[vii]

Speaking in the Guardian about the Government’s revised data, 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.”[viii]