
ALT The construction and running of LLMs comes at a substantial environmental cost. To support progressively larger LLMs, AI companies have built sprawling data processing centres, some bigger than the campuses of the largest US Ivy League universities, and, as Hao notes, some companies like Google, Microsoft and Amazon build their data centres, which run 24/7, 'in threes to have a backup for the backup in case any facility goes down' (Hao, 2025, p. 288). These are mineral-intense endeavours, using significant amounts of copper and lithium. They require enormous amounts of water that is potable and so 'clean enough to avoid clogging pipes and bacterial growth.' Yet around twenty percent of the campuses are built in places with already stressed watersheds (Hao, 2025, p. 288). By 2027, their combined water-use is projected to surpass that of all of Denmark.

ALT These data centres consume colossal amounts of energy. A landmark 2019 study by a team led by American computer scientist Anna Strubell notes that most of the energy for natural language processing isn't derived from carbon-neutral sources (Strubell et al., 2019, p. 3645) and estimates that training a then-standard LLM took fifty-five times the average person's annual carbon budget. Subsequent researchers such as Gebru and her collaborators were able to update Strubell's research, documenting the increase in energy-use as models grew (Bender et al., 2021, Β§3).34 In the years

ALT data centres by 2029 (Smith Goodson, 2025). In April 2025, the International Energy Agency estimated that by 2030 data processing centres would need as much energy as Japan today, 35 thereby, as AI researcher Kate Crawford put it, 'adding a whole new industrial nation onto the grid' (Crawford, 2025).

ALT The environmental impact of LLMs, like the technology's other harms, falls most heavily on already vulnerable populations. For example, xAI has built a massive computing facility, 'Colossus,' in South Memphis, a predominantly Black community. The facility requires more energy than local utilities can supply, so it runs partly on polluting gas turbines. This will not shock anyone familiar with the scholarship on environmental racism. 'It is' Gebru and her co-authors observe, 'well documented in the literature on environmental racism that the negative effects of climate change are reaching and impacting the world's most marginalized communities first' (Bender et al., 2021, p. 612).
































