UK Government Departments Clash Over AI Datacentre Energy Projections
Summary
- • DSIT projects AI datacentres will consume 6GW of electricity by 2030, while DESNZ forecasts only 528MW for the entire commercial services sector
- • DSIT revised its carbon emissions estimates more than a hundredfold after press inquiries, from under 0.142 MtCO₂ to 34–123 MtCO₂ over ten years
- • A prior annexe with implausibly low emissions figures was quietly removed from the government website after Carbon Brief raised questions
- • Critics say the episode exposes either departmental incompetence or uncritical deference to Big Tech narratives on AI infrastructure
Details
DSIT and DESNZ hold wildly divergent projections for AI datacentre energy demand by 2030
DSIT's compute roadmap targets at least 6GW of AI-capable datacentre capacity by 2030. DESNZ's commercial services sector forecast shows only 528MW of total growth over the same period — roughly ten times less. The two departments have not reconciled their figures, exposing a coordination failure at the heart of the government's dual ambitions on AI and net zero.
Each planned AI growth zone would alone require 500MW — nearly matching DESNZ's entire sector forecast
The UK compute roadmap envisions multiple AI growth zones across the country, each needing at least 500MW of electricity supply. DESNZ's projection for growth across the entire commercial services sector between 2025 and 2030 is 528MW. A single growth zone would therefore consume nearly as much new power as the climate department has budgeted for all commercial services combined.
DSIT raised its AI compute emissions estimate more than a hundredfold after press scrutiny
Original figures put cumulative ten-year emissions at 0.025–0.142 MtCO₂, less than 0.05% of projected UK emissions. After Carbon Brief questioned those numbers, the annexe was removed from the government website. Following the Guardian's press inquiry, DSIT published revised figures of 34–123 MtCO₂ — representing 0.9–3.4% of UK projected emissions over ten years. The revision was made without a formal correction notice.
Researchers argue the discrepancy reveals either departmental incompetence or undue Big Tech influence on UK AI policy
UCL researcher Cecilia Rikap framed the gap as evidence that corporations control not only the AI value chain but also government planning processes. NGO Foxglove's Tim Squirrell called the situation 'alarming.' Both characterizations reflect a broader concern that industrial AI policy is being set without adequate scrutiny of its climate and energy implications.
DESNZ says datacentre emissions are included in carbon budget 7, due for release this summer
The department stated that datacentre emissions are factored into its modelling for carbon budget 7 and pointed to the AI Energy Council as a vehicle for attracting clean-power investment for datacentres. This suggests some integration is occurring, but it does not explain the gap between the two departments' published projections.
Policy = government plans and regulatory actions; Infrastructure = physical buildout and capacity; Industry Update = institutional developments and corrections; Insight = attributed analytical or critical perspectives
What This Means
For AI practitioners and policymakers, this episode is a concrete warning that national AI infrastructure strategies can be built on energy and emissions assumptions that have not been stress-tested against climate departments' own models — with the gap only surfacing through NGO requests and press inquiries rather than routine interdepartmental coordination. The more-than-hundredfold upward revision to DSIT's emissions figures, made quietly after outside pressure, undermines confidence in the reliability of the numbers underpinning the UK's AI compute roadmap. Until governments establish formal mechanisms requiring AI industrial policy to be costed against verified climate projections, similar misalignments are likely to emerge in other jurisdictions pursuing both net-zero targets and AI superpower ambitions.