AI Systems Now Consume as Much Water as Global Bottled Water Industry
Amsterdam, Wednesday, 17 December 2025.
Dutch researcher Alex de Vries-Gao reveals artificial intelligence consumes between 313-765 billion liters of water annually, matching worldwide bottled water production.
Groundbreaking Research Reveals Staggering Consumption
Alex de Vries-Gao, a data scientist at Vrije Universiteit Amsterdam, published his findings on December 16, 2025, in the journal Patterns, revealing that AI systems consume between 312.5 and 764.6 billion liters of water annually [1]. This consumption range equals the entire global bottled water production consumed worldwide each year [1][2]. The research combines sustainability reports from major technology companies with scientific data on carbon dioxide emissions and water usage per kilowatt-hour, alongside data from the International Energy Agency [1]. De Vries-Gao’s methodology provides the first comprehensive assessment of AI’s hidden environmental footprint, addressing a critical gap in transparency as major tech companies have been reluctant to publish AI-specific figures on energy and water consumption of their systems [1].
Carbon Footprint Rivals Major Metropolitan Areas
The environmental impact extends beyond water consumption to significant carbon emissions. De Vries-Gao’s study estimates that AI systems could be responsible for 32.6 to 79.7 million tons of carbon dioxide emissions annually [1]. To put this in perspective, New York City emitted 52.2 million tons of CO₂ in 2023, meaning AI systems could potentially exceed the carbon footprint of one of the world’s largest cities [1]. The researcher emphasizes that AI can develop a climate footprint comparable to that of a major world city or other energy-intensive sectors [1]. This dramatic environmental cost stems from the massive energy requirements of data centers housing the graphics processing units that power AI models like ChatGPT [3].
Exponential Growth in Data Center Demands
The surge in AI applications following ChatGPT’s launch in late 2022 has dramatically increased demand for computational resources [1]. AI applications now consume an estimated 15-20% of data centers’ global power consumption in 2024, with projections showing this could reach as much as half of global data center electricity consumption by the end of 2025 [3][1]. The power consumption of individual server racks in large data centers has increased from 10-15 kilowatts to 40-60 kilowatts due to AI implementations, excluding the additional energy required for cooling systems [3]. Data centers worldwide consumed 415 terawatt hours of energy in 2024, with projections indicating this could double to 945 terawatt hours by 2030 [3]. In the Netherlands specifically, data centers consumed 5,100 gigawatt hours of electricity in 2024, representing 4.6 percent of the country’s total electricity consumption, up from 3.3% in 2021 [3].
Water Usage Details and Corporate Transparency Concerns
The water consumption occurs through multiple channels: cooling servers in data centers, electricity generation at power plants, and semiconductor chip production [4]. Research from American universities found that even simple interactions with AI systems consume significant water resources - asking ChatGPT 10-50 questions requires approximately 0.5 liters of water [4]. The International Energy Agency estimated that all data centers consumed 560 billion liters of water in 2023, making AI’s portion of 312.5 to 764.6 billion liters particularly striking [1]. De Vries-Gao criticizes the lack of transparency from major technology companies, stating it is “extremely problematic that these companies now share so little information about this” [1]. Google explicitly stated in a recent report about their Gemini model’s environmental impact that they prefer not to report water consumption at power plants because they do not fully control it [1]. This opacity makes De Vries-Gao’s research particularly valuable, as he warns that indirect water consumption by AI systems may be underestimated by a factor of three to four [1].