AI’s environmental impact per task balloons with more complexity
An analysis by Vals AI indicates that as AI usage shifts to more complex tasks, its environmental footprint, including energy, carbon, and water consumption, increases dramatically. Lengthier AI tasks can have an environmental impact 10,000 times greater than simple queries due to increased computational demands. The report also highlights the lack of transparency from AI companies regarding their models and data center infrastructure, despite pressure from new climate disclosure laws.
A new analysis by Vals AI, an independent AI benchmarking company, reveals that the environmental impact of artificial intelligence tasks escalates significantly with increased complexity. The firm's study found that lengthier AI tasks, such as building a software application, can have an environmental footprint up to 10,000 times greater than simple queries, like answering a basic question. This dramatic increase is attributed to the extended computational requirements for "agentic" tasks, which demand more processing steps.
The analysis evaluated 16 models across carbon emissions, water consumption, and electricity use, utilizing baseline assumptions for hardware, energy mix, and data center efficiency. While most models analyzed were from Chinese companies, US-based Thinking Machines Lab also had a model included. Vals noted that most leading US AI models, like those from OpenAI and Anthropic, are "closed-weight," making direct environmental evaluation challenging.
Despite limited public disclosure from major US AI companies regarding the environmental toll of complex tasks, previous estimates by OpenAI's CEO Sam Altman and Google placed the energy cost of an average text query at 0.24 to 0.34 watt-hours. Microsoft's research corroborates Vals AI's findings, suggesting that long-reasoning requests can increase energy consumption by more than an order of magnitude. The Vals report also highlighted that small performance improvements in AI models often come with disproportionately large environmental costs.
The article also points out that while Amazon and Google reported increased greenhouse gas emissions in 2025 due partly to data center expansion, detailed disclosures about AI's specific carbon emissions remain scarce. Vals co-founder and CEO Rayan Krishnan emphasized the need for AI companies to provide more transparency on model architecture, hardware, and data center infrastructure to facilitate accurate environmental impact measurement, especially with a new California climate disclosure law pressuring businesses.