The UK government has revised its AI data centre carbon estimate by a factor of more than 100, projecting up to 123 million tonnes of CO2 over the next decade,a correction that blows a hole in climate targets and energy planning assumptions alike.
The original figure, quietly removed from the Department for Science, Innovation and Technology's Compute Evidence Annex after Carbon Brief and Foxglove flagged it, projected emissions peaking at 0.142 million tonnes of CO2 in a single year. The revised range: 34 to 123 million tonnes across the 2025-2035 decade. At the upper end, that's equivalent to the annual output of 2.7 million people, and represents 0.9% to 3.4% of the UK's total projected emissions for the period. The Telegraph put the revision more bluntly: 136,000% off. The government has not commented.
Foxglove's strategy director Tim Squirrell told the Guardian that the situation has proved far graver than initially apparent, noting that the government appears not to have conducted even the most fundamental calculations before committing to rapid hyperscale data centre expansion. The Compute Roadmap, which outlines the UK's strategy to build a premier AI computing ecosystem, had been presented as an economic opportunity. The revised emissions figures reveal the environmental accounting was missing entirely.
The miscalculation cascades into grid infrastructure, renewable procurement, and planning approvals. Ofgem's connections pipeline already shows 71 mature data centre projects totalling roughly 20GW of capacity. The original DSIT figure assumed 11.2GW of AI compute demand by 2035, which was already contested. Energy planners, local councils issuing planning permissions, and grid operators making long-term investment decisions all worked from the lower emissions baseline. Those assumptions now need rebuilding from scratch.
The lower end of the new estimate hinges on two variables that are far from guaranteed: improved efficiency in AI workloads and a faster than expected grid decarbonisation. As Data Centre Review noted in March, the situation is messier than a simple fossil-fuelled boom scenario, but the direction of risk is unambiguously upward. Hyperscale campuses coming online in the next three years were contracted before the emissions revision. Their carbon footprint is locked in by the energy mix they'll draw on when they come live.
What This Means for AI Infrastructure Builders
For startups building or co-locating AI infrastructure in the UK, the regulatory environment has materially shifted in the last week. Planning approvals for energy-intensive facilities will face greater scrutiny. Carbon offset requirements, currently light-touch, are likely to tighten as DSIT works on its promised more accurate assessment. Investors evaluating data centre assets need to model the cost of compliance with a climate framework that is now recalibrating against a much larger emissions number.
The procurement decision that matters most right now is energy sourcing. Facilities that can demonstrate power purchase agreements with renewable generators, or that are co-located with nuclear or offshore wind assets, will face less regulatory headwind than those drawing from the standard grid mix. The UK's net zero 2050 obligation is legally binding. Doubling electricity consumption via unchecked AI expansion, without corresponding renewable buildout, forces a political confrontation between the government's AI ambitions and its climate commitments. That confrontation is arriving faster than anyone planned for. Build your energy stack accordingly.
Also read: Jesse Pollak bets AI agents will drive the next wave of crypto payments • Developers flee Claude Code rate limits for OpenAI Codex as throughput war heats up • Met Police Palantir AI flags hundreds of officers, sparking privacy probe
