A recent study by the Green IT association, published on September 30, warns that the environmental and health impact of artificial intelligence (AI) could grow seven times between 2025 and 2030. The research estimates that AI's electricity use and carbon dioxide emissions could increase significantly during this period. The study uses established scientific methods, including multi-criteria life cycle analysis, the European Commission's Product Environmental Footprint, and the Ademe's digital services guidelines. It considers all AI types—such as generative, agentive, and predictive—and their infrastructure, including servers, cooling systems, and data centers. By 2030, AI could consume the sustainable annual budget of 645 million people, which is 1.4 times the population of the European Union. This sustainable annual budget represents the environmental and health impacts an individual can have in a year without exceeding the planet's limits. AI contributes to four major environmental and health issues: global warming, eutrophication of water bodies, depletion of non-renewable resources, and emissions of fine particles. The Green IT association calls for stronger public policies, including legislation to promote responsible AI use, an "AI" label similar to those for genetically modified organisms, and eco-design requirements for AI systems in France. According to the study, AI could account for 15% of global greenhouse gas (GHG) emissions, rising from 99 million to 636 million tons of CO2 equivalent between 2025 and 2030. In 2030, AI would consume 16% of an individual's annual sustainable budget, excluding other digital uses, which could take up 56% of a European's budget. This surge is due to the doubling of graphics processing units (GPUs) and the tripling of power each GPU uses, from about 700 to 2,000 watts. The report also predicts that the surface area of data centers will grow sixfold by 2030, with most located in the United States and China, where electricity is more carbon-intensive. The environmental footprint of AI infrastructure could increase by 6.9 times between 2025 and 2030. The study projects that the number of GPUs used in AI will rise from 13 million in 2025 to 25 million by 2030, with electricity consumption expected to jump from 174 terawatt-hours (TWh) per year in 2025 to 1,169 TWh in 2030. Server density is also increasing, with the number of servers expected to triple from 1.9 to 6.2 million. The report highlights that 92% of environmental impacts in 2025 and 96% in 2030 come from electricity use, with manufacturing contributing less than 5%. The study also notes that AI could lead to 20,639 additional deaths in France by 2030 due to pollution and resource depletion. The Green IT association recommends five measures for public authorities, including promoting AI sobriety through legislation, implementing an "AI" labeling system, creating an excellence track in frugal AI, mandating eco-design for AI systems in France, and introducing a bonus-malus system for AI centers using fossil fuels. In France, a 109 billion euro plan announced in early 2025 includes 35 new AI-dedicated data centers, such as Google's project in Châteauroux. However, there are currently no environmental regulations for AI at the national or European level. In the United States, New York has imposed a temporary halt on new data centers of 50 megawatts or more. The study was conducted independently by five members of the Green IT association, using the NegaOctet and Ecoinvent databases, and includes an internal review but no external peer review.