UK authorities underestimated AI data center emissions by more than 100 times

Britain suddenly discovered that artificial intelligence's climate footprint in the country is not just noticeable, but potentially enormous: after revising official calculations, authorities admitted that emissions from AI data centers in the country could have been underestimated by more than a hundredfold. This is an important signal not only for British politics, but for the entire race for computing power: the more actively states and corporations build infrastructure for AI, the harder it becomes to pretend that its energy cost is secondary. The revision was triggered by updated data that British officials published without much fanfare.

According to the new assessment, energy consumption from AI data centers in the UK could result in emissions of 34 million to 123 million tons of CO2 in the period from 2025 to 2035. The upper bound is comparable to emissions from approximately 2.7 million people over the same timeframe.

For comparison, the previous assessment, which was subsequently removed, spoke of a maximum of 0.142 million tons of CO2 in a single year. Formally, this is a different calculation horizon, but even with this caveat, the scale of the discrepancy seems too large to dismiss as methodological nuances.

The new figures appeared in a revised version of Britain's compute roadmap — a state road map for creating a "world-class" computational ecosystem for developing artificial intelligence. This strategy is a priority for authorities as a source of future economic growth. The logic is understandable: more data centers, more computing power, better chances of attracting investment, research projects, and AI business.

The problem is that AI data centers consume significantly more electricity than traditional facilities for storing and processing regular online data, and a substantial portion of this energy remains tied to fossil fuels. Even the lower bound of the new assessment looks sensitive within the scope of the nation's climate agenda. According to calculations by the British Department for Science, Innovation and Technology, AI infrastructure could account for 0.

9% to 3.4% of the country's total emissions in 2025–2035. The lower scenario is only possible with faster improvements in model and hardware efficiency, as well as accelerated energy system decarbonization.

In other words, the optimistic option depends not on a single factor, but on several improvements that still need to be implemented in practice. This turns the conversation about "green AI" from a matter of image into a matter of concrete energy policy. The revision of the assessment came after criticism from the independent organization Foxglove and climate publication Carbon Brief, which pointed out that the previous figures appeared implausibly low.

Critics note that Britain has a legally enshrined goal of achieving net zero by 2050, and it is increasingly incompatible with aggressive expansion of hyperscale data centers. Foxglove believes that without control, such a course could sharply increase electricity demand and add the country a new major source of emissions precisely at a moment when the carbon budget is becoming tighter. The government, meanwhile, did not provide detailed public commentary on the revision.

For the market, this is a story not only about a calculation error, but about the end of a convenient illusion that the AI boom can be discussed separately from energy and climate. If even in a country with a formalized climate agenda and developed state apparatus the effect of AI data centers was initially underestimated to such a degree, then similar blind spots could exist in other jurisdictions as well. The next stage of the discussion will not be about whether AI infrastructure is needed, but about who will pay for its energy consumption, how quickly networks and generation can adapt, and what restrictions states are willing to introduce for their own climate goals.