This Danish article from the Dansk Data Science Community examines AI's climate impact, highlighting the extreme difficulty of determining accurate energy consumption figures for large language models like ChatGPT due to lack of transparency from companies like Meta and OpenAI, who only report emissions from GPU components rather than full system impacts. The authors note that training GPT-3 consumed an estimated 1,287,000 KWh and produced around 550 tons of CO2 (equivalent to 50 Danes' annual emissions), while usage estimates vary wildly—with the highest estimate being 633 times greater than the lowest. Though individual AI queries appear to consume minimal energy (comparable to a fraction of iPhone charging), the cumulative impact could be massive as billions adopt AI-powered applications daily. The article also discusses other environmental impacts like water consumption for data center cooling, reports significant emission increases from Google (48% since 2019) and Baidu (32.6% since 2021) attributed to AI development, but notes promising efficiency improvements and calls for more transparency from AI companies to enable informed debate about whether AI's climate costs justify its benefits.