Officials Underestimated AI Datacentres’ Carbon Impact by 100 Times 2026

Key Points

• UK government revised AI datacentres’ carbon emissions estimate by over 100 times, from 0.142m tonnes CO₂ in one year to 34m-123m tonnes over 10 years.
• New figure equates to emissions from 2.7 million people over a decade.
• Revision in UK’s “compute roadmap” for AI ecosystem, tied to economic growth hopes.
• AI datacentres demand far more electricity than standard ones, mostly from fossil fuels.
• Range: 0.9% to 3.4% of UK’s projected 2025-2035 emissions; lower end needs efficiency gains and grid decarbonisation.
• First reported by Politico; investigation by Foxglove and Carbon Brief highlighted underestimation.
• Patrick Galey of Global Witness calls it “historic idiocy” amid tight carbon budget.
• Tim Squirrell of Foxglove warns of doubled national electricity use and net zero conflict.
• Government declined on-record comment.
• Rising global alarm over AI’s climate threat.

London (Britain Today News) April 24, 2026 – The UK government has dramatically revised its estimate of carbon emissions from AI datacentres upwards by a factor of more than 100, exposing a major underestimation of the technology’s climate impact.

According to new data published this week in a revision to the government’s “compute roadmap”, energy use by AI datacentres could emit up to 123 million tonnes of carbon dioxide (CO₂) over the next decade – equivalent to the annual output of 2.7 million people. This replaces a now-deleted prior forecast of just 0.142 million tonnes in a single year.

The disclosure has intensified concerns about AI’s role in exacerbating the climate emergency, as datacentres powering the technology guzzle electricity generated largely from fossil fuels.

Why Did Officials Underestimate AI Datacentres’ Carbon Emissions?

As reported by Daniel Boffey of The Guardian, officials at the Department for Science, Innovation and Technology (DSIT) quietly updated the figures in the UK’s compute roadmap, which outlines plans to foster a “world-class compute ecosystem” for AI to drive economic growth.

The original estimate, since scrubbed from public view, projected a mere 0.142 million tonnes of CO₂ in one year. The revised range now spans 34 million to 123 million tonnes from 2025 to 2035, representing 0.9% to 3.4% of the UK’s anticipated total emissions in that period.

This escalation stems from AI datacentres’ voracious energy needs, far exceeding those of conventional data storage facilities. DSIT’s analysis assumes potential mitigations like improved AI model efficiency, advanced hardware, and accelerated grid decarbonisation to hit the lower end.

The revision followed scrutiny from external analyses. Politico first flagged the change, noting the government’s pivot after initial projections proved wildly off-mark.

What Sparked the Government’s Revised Emissions Forecast?

Investigations by watchdog group Foxglove and climate news site Carbon Brief prompted the rethink. As detailed by Tim Squirrell, Foxglove’s head of strategy, in their report, the government’s embrace of massive AI infrastructure clashed with its net zero 2050 pledge.

Tim Squirrell stated:

“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.”

Carbon Brief’s analysis echoed this, estimating CO₂ from UK data centres could be hundreds of times higher than thought, factoring in grid emissions factors.

DSIT officials made the adjustment without fanfare, embedding it in the compute roadmap update. The document underscores AI as a cornerstone for UK prosperity, yet acknowledges the environmental toll.

How Severe Is the Climate Threat from AI Datacentres?

Patrick Galey, head of investigations for Global Witness climate campaign, lambasted the oversight. As quoted by Daniel Boffey of The Guardian, Galey said:

“We have a handful of years until our carbon budget is exhausted. To waste what little bandwidth we have left – when 750 million people worldwide 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 123 million tonne upper estimate rivals emissions from entire sectors. For context, the UK’s total annual CO₂ output hovers around 350-400 million tonnes, per recent Office for National Statistics data. Over 10 years, AI datacentres could thus claim a slice rivaling millions of households.

Global alarm is mounting. The Guardian highlighted “increasing alarm” at AI’s pollution footprint, linking to broader fears of unchecked expansion.

AI’s energy hunger arises from training and running large language models, which process vast computations. A single ChatGPT query reportedly consumes 10 times the electricity of a Google search, scaling massively for datacentre operations.

Can the UK Mitigate AI’s Carbon Footprint Effectively?

The lower emissions scenario – 34 million tonnes – hinges on breakthroughs. DSIT projects hinge on “greater efficiency in AI models and hardware” and faster fossil fuel phase-out in power generation.

Yet challenges abound. The UK grid remains 30-40% reliant on gas, per National Grid reports. Renewables expansion lags, with offshore wind delays and nuclear setbacks.

Foxglove’s probe revealed DSIT initially overlooked full grid carbon intensity. Carbon Brief quantified this: applying realistic emissions factors ballooned projections.

Government silence persists. DSIT declined on-record comment, leaving questions about modelling flaws unanswered.

Industry voices urge balance. TechUK, a trade body, advocates “green AI” via efficient chips and co-location with renewables, but critics like Global Witness see hype over substance.

What Does This Mean for the UK’s Net Zero Ambitions?

The revelation pits economic zeal against climate goals. The compute roadmap, launched under prior administrations, positions AI as a growth engine, potentially adding billions to GDP via sectors like finance and healthcare.

However, unchecked buildout risks grid strain. National Grid ESO warns data centres could double demand by 2030, necessitating £30 billion in infrastructure.

Net zero by 2050 demands 78% emissions cuts from 1990 levels. AI’s projected 3.4% share could derail this if unabated.

Parliamentary scrutiny looms. The Environmental Audit Committee has probed data centre impacts, citing water use and heat waste alongside carbon.

Internationally, parallels emerge. The US and EU face similar AI energy surges, with hyperscalers like Microsoft and Google pledging renewables yet drawing fossil-backed power.

Who Is Accountable for the Underestimation?

DSIT bears primary responsibility, as roadmap steward. Critics point to rushed projections amid AI hype post-ChatGPT boom.

Foxglove’s Tim Squirrell accused basic arithmetic failures. Politico’s reporting exposed the deleted figure, suggesting embarrassment.

Global Witness’s Patrick Galey frames it morally: prioritising elite tech over global energy access.

No resignations or inquiries announced. Opposition MPs, including Labour’s Chi Onwurah, have demanded transparency, per Hansard records.

Will AI Growth Derail the Climate Emergency Response?

Urgency underscores the stakes. IPCC reports stress halving emissions by 2030. UK’s carbon budget – the safe emissions limit – nears exhaustion, per Climate Change Committee.

AI optimism persists. Proponents cite productivity gains funding green tech, but evidence lags.

The Guardian’s coverage ties this to “unbelievable amounts of pollution”, amplifying calls for regulation.

In sum, the 100-fold revision signals a wake-up. Balancing AI’s promise with planetary limits demands rigorous policy, not underestimation.