The UK government's Department for Science, Innovation & Technology (DSIT) has been forced to correct its projections for the carbon footprint of artificial intelligence (AI) datacentres after a miscalculation that resulted in figures being revised upwards by around 100 times. The original estimates, published in July 2025 as part of the Compute evidence annex, stated that by 2035 the UK's greenhouse gas emissions from AI compute would range from 0.025 to 0.142 million tonnes of CO₂ (MtCO₂), an amount described as below 0.05% of the UK's projected total emissions. However, last week DSIT quietly updated that document, acknowledging that the correct cumulative 10-year figure would be between 34 and 123 MtCO₂, representing 0.9% to 3.4% of the UK's projected emissions over that period.
The blunder highlights the growing challenge of accurately forecasting the environmental impact of the rapid expansion of AI infrastructure. Datacentres are enormous consumers of electricity, and as the UK pushes to become a global leader in AI, the energy demands are projected to surge. According to Computer Weekly's own tracking, the UK currently has approximately 1.6 gigawatts (GW) of operational datacentre capacity, with over 8 GW in planning or under construction. This is in line with the government's target to achieve 6 GW of AI-capable capacity by 2030.
What went wrong with the calculations?
The initial error appears to have arisen from a simple misreading of timescales. The original figures were presented as annual emissions, whereas the corrected figures represent cumulative emissions over a decade. This mistake was compounded by the fact that the government had not adequately factored in the potential reliance on gas-fired power generation to meet the soaring electricity demand from datacentres. The original document assumed that all electricity would come from clean sources such as wind, nuclear, hydro, and solar, with a carbon intensity target of just 50 grams of CO₂ per kilowatt-hour by 2030. But that assumption is now being widely challenged.
Analysis by Carbon Brief, a climate change research group, suggests that the government's clean energy target is wildly optimistic if any significant portion of datacentre power needs to be supplied by gas. Gas-generated electricity has a carbon intensity roughly ten times that of clean sources. Carbon Brief's calculations indicate that emissions could range from 3.4 MtCO₂ (if just 5% of power came from gas) to a staggering 68.1 MtCO₂ (if 95% of power was gas-generated). The upper estimate is not far off the total annual carbon emissions of Sweden.
The Ofgem projection of 20 GW of future datacentre electricity demand underscores the scale of the challenge. To put that in context, the UK's peak electricity demand in February 2026 was 45 GW. Adding 20 GW of datacentre load would increase the national peak by nearly half, placing enormous strain on the grid and making decarbonisation far more difficult. The National Energy System Operator's research, which collected future grid connection requirements from customers, informs this projection.
Environmental campaigners cry foul
Tim Squirrell, head of strategy at the environmental watchdog Foxglove, which partnered with Carbon Brief on the research, was blunt in his criticism. He said the government has a legally binding commitment to reach net zero by 2050, which sits awkwardly alongside its 'hell-for-leather embrace of a hyperscale AI datacentre buildout'. He warned that unchecked, such a buildout could double the country's electricity consumption, and the latest revelations suggest the government hasn't done even the most basic arithmetic to measure potential new carbon emissions. He urged the government to confront the reality that it cannot rubber-stamp hundreds of new datacentres while keeping its manifesto promise to combat the climate crisis.
The controversy comes at a time when the UK is positioning itself as a global hub for AI. In March 2026, the government announced an 'AI Growth Zone' with streamlined planning approval for new datacentres. Several large projects have already been approved, including a 1.2 GW facility in Buckinghamshire that was the first to be designated as a Nationally Significant Infrastructure Project. Another 800 MW campus was recently approved in the North West, and multiple hyperscale facilities are planned across Scotland and the M62 corridor.
The wider context: Energy and net zero
The miscalculation raises questions about the government's ability to reconcile its AI ambitions with its climate obligations. The Climate Change Committee, which advises the government on emission targets, has warned that even without AI datacentres, the UK will struggle to meet its 2030 clean power target. Adding datacentre demand could make the target unachievable unless massive investments in renewable generation and grid infrastructure are accelerated.
One potential solution is to require datacentres to be powered by on-site renewable generation or to purchase certified clean power. However, critics argue that even with such measures, the embedded emissions from constructing these facilities and the manufacturing of hardware are significant. Moreover, the reliance on backup generators (often diesel-powered) for grid outages remains a concern. Industry figures from the Uptime Institute suggest that many datacentres rely on diesel generators for an average of 50 to 100 hours per year, which can add to overall carbon emissions.
The UK is also exploring the use of small modular reactors (SMRs) to provide carbon-free baseload power for datacentres. In February 2026, the government announced a partnership with Rolls-Royce to develop SMRs for industrial users, with datacentres seen as a prime candidate. However, SMRs are not expected to come online until the early 2030s at the earliest, leaving a gap in the short term.
As the debate continues, the focus remains on transparency and accuracy. The DSIT's initial error has eroded trust in the government's handling of the issue. Environmental campaigners are calling for a mandatory carbon impact assessment for any new datacentre over a certain size, similar to the environmental statements required for other large infrastructure projects. The government has not yet responded formally to the correction, but it is expected to publish an updated Compute evidence annex in the coming weeks.
In the meantime, the UK's datacentre industry is continuing to grow at a breakneck pace. According to data from Barbour ABI, the total pipeline of datacentre projects now exceeds 10 GW, with most of the largest projects located in the North of England and Scotland. London and the M4 corridor still account for about a quarter of projected capacity, but the trend is firmly northwards as land and power become scarcer in the South East. The challenge for the sector, and for policymakers, will be to ensure that this growth does not come at the cost of the country's climate commitments.
Source: ComputerWeekly.com News