The Department for Science, Innovation & Technology (DSIT) has been forced to admit a serious miscalculation in its projections for the carbon footprint of AI datacentres in the UK. In a correction published last week, the government department revised its estimates upwards by roughly 100 times, sparking widespread criticism and casting doubt on the nation's ability to reconcile its AI ambitions with legally binding net-zero targets.
Last July, the DSIT published its Compute evidence annex, which outlined future AI compute demand and its implications for carbon emissions. The document stated: 'We estimate that by 2035, the UK's greenhouse gas emissions from AI compute could range from 0.025 to 0.142 MtCO₂ – this is below 0.05% of the UK's projected total emissions.' These figures were presented as annual projections, but the underlying calculations were fundamentally flawed.
In the correction, the DSIT declared: '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.' The stark discrepancy – an increase by a factor of about 100 – has raised questions about the department's competence and the rigour of its modelling.
The scale of the error
To put the miscalculation into perspective, the original estimate of 0.025-0.142 MtCO₂ would equate to roughly 0.05% of total UK emissions – a negligible amount that would barely register in national carbon accounting. The revised figure, however, represents up to 3.4% of projected UK emissions over a decade, a significant contribution that could severely hamper progress towards the government's 2050 net-zero target. The error appears to stem from the department using annual figures where cumulative decade-long numbers were required, effectively misrepresenting the scale of the challenge by two orders of magnitude.
Further analysis by climate science and policy research group Carbon Brief suggests that even the corrected estimates may be optimistic. Central to this concern is the government's target for grid carbon intensity: 50gCO₂/kWh by 2030. This ambitious goal assumes that almost all electricity will come from clean sources such as wind, nuclear, hydro, and solar. However, recent research published by Carbon Brief in collaboration with environmental campaigners Foxglove indicates that this assumption may be wildly unrealistic if gas-fired power generation is required to meet baseload demand.
The role of gas in the energy mix
Gas-powered electricity generation carries a carbon intensity roughly 10 times that of clean sources. According to Carbon Brief, if just 5% of the electricity consumed by AI datacentres came from gas, emissions could reach 3.4 MtCO₂ over a decade. If gas were to account for 95% of the supply – a scenario that may become increasingly likely given the intermittent nature of renewable energy – emissions could skyrocket to 68.1 MtCO₂. This upper figure is not far off the annual carbon emissions of Sweden, a country of 10 million people with a highly industrialised economy.
The basis for these calculations is a recent Ofgem projection that UK datacentres could require up to 20GW of electricity by 2030. To contextualise this demand, the same Ofgem document noted that actual peak electricity demand in February 2026 was 45GW. In other words, datacentres alone could consume nearly half of the country's peak power within a few years – an extraordinary leap from the current 1.6GW of installed capacity identified by Computer Weekly's analysis.
Current datacentre pipeline and capacity
Computer Weekly's own research has found that there is currently approximately 1.6GW of operational datacentre capacity in the UK. However, a staggering 8GW is in the planning or construction phase, with the majority of large projects concentrated in the north of England and Scotland. The M62 corridor and regions beyond London are becoming the new hubs for hyperscale AI datacentres, while the traditional M4 corridor now accounts for only about 25% of projected capacity. This rapid expansion is driven by the insatiable demand for AI compute power, which has grown exponentially since the launch of large language models and generative AI applications.
Tim Squirrell, head of strategy at Foxglove, expressed deep concern over the government's handling of the situation. He 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 urgently needs to confront the reality that it can't rubber stamp hundreds of new datacentres, whilst keeping its manifesto promise to the country – and legal obligation – to combat the climate crisis.'
Historical context and policy implications
This is not the first time that UK government projections on energy and emissions have been called into question. In 2022, the Office for Budget Responsibility criticised the Treasury for relying on overly optimistic assumptions about carbon pricing and technology costs. More recently, the Climate Change Committee has repeatedly warned that the UK is off track to meet its 2030 emissions reduction target of 68% below 1990 levels. The AI datacentre miscalculation adds to a growing body of evidence suggesting that ministers may be underestimating the scale of the transformation needed.
The National Energy System Operator (NESO) has also contributed to the debate. Its research, which surveys customers about future grid connection requirements, provided the basis for Ofgem's 20GW projection. However, critics argue that NESO's methodology may not fully capture the potential for energy efficiency improvements or the impact of government intervention to limit datacentre growth. The Foxglove and Carbon Brief report highlighted that if aggressive energy efficiency measures and strict carbon intensity limits were enforced, the worst-case scenario could be avoided. But without such measures, the UK could find itself locked into a high-carbon energy pathway for decades.
International comparisons
Other countries are also grappling with the tension between AI development and climate goals. Ireland, which hosts the European headquarters of several tech giants, has seen its datacentre electricity consumption more than triple since 2015, reaching 21% of national demand in 2022. The Irish government imposed a de facto moratorium on new datacentre connections to the grid until 2028, citing grid capacity constraints and renewable energy availability. Singapore similarly paused datacentre construction for several years before introducing a green datacentre roadmap that emphasises energy efficiency and carbon offsetting.
The UK appears to be taking a different approach, with the government actively encouraging datacentre investment as part of its AI and digital strategy. In the spring Budget, the chancellor announced new incentives for hyperscale projects and fast-tracked planning approvals. Yet the revised carbon footprint figures suggest that this policy may be inconsistent with the country's legally binding net-zero target. Environmental groups have called for an immediate moratorium on new datacentre approvals until a comprehensive carbon assessment is conducted.
The technical challenge of decarbonising AI compute
Beyond the immediate carbon accounting issues, there are deeper technical challenges to decarbonising AI compute. Training large AI models consumes enormous amounts of energy, and the inference phase – when models are used to generate responses – is also energy-intensive. Improved hardware efficiency, such as the use of more advanced GPUs and specialised AI accelerators, can help but may be offset by the rapid increase in model size and usage. Cooling datacentres also requires substantial energy, though immersion cooling and other innovations are reducing that burden.
The government's own AI Opportunities Action Plan, published earlier this year, set a target of 6GW of AI-capable datacentre capacity by 2030. Achieving that while also decarbonising the grid would require either a massive expansion of clean energy generation or a drastic reduction in the carbon intensity of backup power. The latter seems unlikely given that gas-fired plants are currently the primary source of backup for intermittent renewables. Without a breakthrough in grid-scale battery storage or carbon capture technology, the carbon footprint of AI datacentres could be much higher than the DSIT's revised estimates suggest.
Computer Weekly has also reported that the UK's 2030 datacentre capacity target appears shaky. Data from Barbour ABI shows that the total datacentre pipeline amounts to around 8GW, but only a fraction of that is likely to be operational by the end of the decade. Planning delays, supply chain bottlenecks, and grid connection queues are all slowing progress. The government's reckoning with the carbon implications of this buildout comes at a critical juncture, as policymakers weigh the economic benefits of AI against the environmental costs.
In conclusion, the DSIT's miscalculation of AI datacentre carbon emissions has exposed a significant gap between government rhetoric and the reality of the UK's energy transition. The revised figures, combined with Carbon Brief's even more pessimistic analysis, suggest that the UK's AI ambitions could come at a considerable environmental price. Without immediate action to improve grid decarbonisation, enforce energy efficiency standards on datacentres, and potentially slow the pace of new construction, the country's net-zero target may be placed in jeopardy. The debate now shifts to whether the government can reconcile its dual commitments to AI leadership and climate action before it is too late.
Source: ComputerWeekly.com News