Research released this week reveals that artificial intelligence generated emissions equivalent to New York City’s annual output during 2025. The study additionally found that water consumption for AI purposes now surpasses total global bottled-water consumption. These findings represent the first comprehensive attempt to isolate AI’s environmental footprint from broader datacentre operations, as usage of systems like ChatGPT and Gemini expanded dramatically throughout the year.
Researcher Alex de Vries-Gao, founder of Digiconomist, compiled the analysis using technology firms’ own environmental data. His calculations indicate that AI greenhouse gas emissions reached approximately 80 million tonnes, while water usage potentially reached 765 billion litres. He emphasized that current environmental costs fall on society rather than the technology companies profiting from these systems.
AI’s environmental impact equals over 8% of global aviation emissions, according to the study published in the journal Patterns. De Vries-Gao called for stricter transparency requirements, questioning why companies should not bear the environmental expenses associated with their profitable technology. His water impact assessment represents the first such measurement and demonstrates AI usage alone exceeds previous total datacentre water estimates by more than one-third.
The International Energy Agency previously noted that AI-focused datacentres consume electricity comparable to aluminium smelters. Datacentre electricity consumption is projected to double by 2030. Individual hyperscale facilities generate emissions equivalent to multiple international airports, with an estimated 100-200 planned facilities in the UK alone. The largest American facilities will consume electricity equal to 2 million households combined, with the US accounting for 45% of global datacentre electricity use, followed by China at 25% and Europe at 15%.
A planned UK datacentre at Blyth, Northumberland will emit over 180,000 tonnes of CO2 annually when operational, equivalent to 24,000 homes’ emissions. India’s $30 billion datacentre investment raises concerns about diesel generator farms compensating for grid unreliability, creating substantial carbon liabilities. Technology companies provide insufficient environmental disclosure for proper assessment, with Google notably excluding water impact from its recent Gemini reporting despite reducing datacentre emissions by 12% through clean energy sources.
