2025人工智能的环境影响研究报告

Environmental Impactsof Artificial Intelligence Authors Jens GrögerFelix BehrensPeter GailhoferInga Hilbert Öko-Institut Consult GmbHBorkumstraße 213189 Berlin Kein Geld von Industrie und Staat Greenpeace arbeitet international und kämpft mit gewaltfreien Aktionenfür den Schutz der Lebensgrundlagen. Unser Ziel ist es, Umweltzerstörungzu verhindern, Verhaltensweisen zu ändern und Lösungen durchzusetzen.Greenpeace ist überparteilich und völlig unabhängig von Politik undWirtschaft. Rund 620.000 Fördermitglieder in Deutschland spenden anGreenpeace und gewährleisten damit unsere tägliche Arbeit zum Schutzder Umwelt, der Völkerverständigung und des Friedens. Impressum Greenpeace e.V.Hongkongstraße 10, 20457 Hamburg, T 040 30618-0 PressestelleT 040 30618-340, F 040 30618-340,presse@greenpeace.de, greenpeace.dePolitische Vertretung BerlinMarienstraße 19–20, 10117 Berlin, T 030 308899-0V.i.S.d.P.Jonathan Niesel Titelfoto© Samuel Golay / picture alliance Stand05 / 2025 Foreword Artificial Intelligence (AI) is omnipresent and transforming the world. The increasing use of AI not only bringsprogress—it also introduces new environmental challenges. Several examples illustrate the current dynamic: ▶AI models are becoming larger, more complex, and more energy-hungry.▷The number of parameters has risen from 1.5 billion (GPT-2, 2019) to 2 trillion (LLama 4, 2025).▷The computational effort required for training doubles approximately every five months. [Epoch AI]▶The global electricity demand for AI computing is expected to be about 11 times higher in 2030 than in2023. [Environmental Impacts of AI, Öko-Institut 2025]▶In the U.S., data centers could consume more electricity by 2030 than the entire energy-intensive goodsproduction sector (cement, chemicals, steel) combined. [International Energy Agency (IEA)] Greenpeace has been advocating for a more environmentally friendly internet and data center infrastructure foryears (e.g., benchmarking the energy performance of the IT sector in 2009; Clicking Clean: Who is winning therace to build a green internet in 2017). But those successes are being undone: instead of decreasing emissions from Google, Microsoft, and AWS, thecompanies are reporting increasing emissions or only incomplete data on environmental impact. At the sametime, they are heavily investing in AI infrastructure to secure market share and future profits. It remains unclearwhether and how this aligns with their self-imposed climate goals. AI can be a useful technology, but its negative environmental impacts must be limited. Greenpeace does not fun-damentally reject AI but emphasizes the urgency of taking action now to ensure AI does not further exacerbatethe climate crisis. This report presents, for the first time, a comprehensive overview of AI’s environmental impacts. It is intendedas a fact-based, scientific foundation for discussion—one on which we can collectively search for solutions andscale environmentally sustainable approaches. While AI clearly has many social and societal opportunities andrisks, this report focuses specifically on its ecological impact. The report by Greenpeace East Asia shows that accelerating the deployment of renewable energy is possiblewithin the chip production supply chain. There is also great potential for improved energy efficiency through "Green AI" policies in companies—for exam-ple, by choosing smaller models, using pre-trained models, and providing training in green coding.Even with energy-efficient algorithms, electricity consumption from AI applications will continue to rise asmore people use AI in both their professional and private lives. As of April 2025, ChatGPT usage set a new globalrecord, doubling to 5.1 billion visits within a year. [Statista] Technological advances alone will not eliminate the environmental impacts caused by technology: efficiencygains reduce costs, which then increases usage and offsets the savings. This is known as the rebound effect, orJevons paradox. Therefore, avoiding environmentally harmful AI use cases must also be part of the solution. For instance, it isnot desirable for AI applications to lower the cost of oil extraction, thereby increasing fossil fuel consumption. Greenpeace calls for the following measures to minimize the environmental impacts of Artificial Intelligence: 1.An energy-efficient AI infrastructure powered 100% by renewable energy. This green power must beadditionally generated.2.AI companies must disclose:a.How much electricity is used in operating their AI.b.How much power is consumed by users during their use of AI.c.The goals under which their models were trained, and which environmental parameters wereconsidered.3.AI developers must take responsibility for their supply chains. They must contribute to the expansionof renewable energy in line with their growth and ensure that local communities do not suffer negativeconsequences (e.g., lack of drinking water, higher electricity prices). This last point