The rapid expansion of artificial intelligence in 2025 has generated carbon dioxide emissions comparable to those of New York City, according to new research. The study, published in the journal Patterns, estimates that AI systems could emit up to 80 million tonnes of CO2 this year, while water consumption may reach 765 billion litres—exceeding global bottled-water demand.
Alex de Vries-Gao, founder of Digiconomist and author of the study, claims this is the first attempt to isolate AI's environmental impact from general datacentre activity. The findings indicate that AI-related emissions now represent over 8% of global aviation emissions. De Vries-Gao called for stricter transparency requirements for tech companies, stating: 'Society is paying for these costs, not the tech companies.'
The International Energy Agency has noted that AI-focused datacentres consume electricity comparable to aluminium smelters, with consumption expected to double by 2030. In the UK, a planned datacentre in Blyth, Northumberland, could emit over 180,000 tonnes of CO2 annually once operational, equivalent to more than 24,000 homes.
Donald Campbell, director of advocacy at Foxglove, warned that the findings are 'just the tip of the iceberg,' highlighting that the UK has an estimated 100-200 hyperscale datacentres in planning. In India, concerns have been raised about reliance on diesel generators for backup power, which KPMG described as a 'massive carbon liability.'
De Vries-Gao criticised tech companies for insufficient environmental disclosures, noting that Google's recent report on its Gemini AI did not account for water used in electricity generation. Google acknowledged that achieving climate goals is 'more complex and challenging' due to slow deployment of carbon-free energy technologies.
