AI could soon emit as much CO₂ as New York

New research indicates that by 2025, AI systems could emit as much CO₂ as New York and consume water at a rate comparable to global bottled water consumption.

The study, conducted by Alex de Vries-Gao, a researcher at Vrije Universiteit Amsterdam (VU Amsterdam) Institute for Environmental Studies and founder of Digiconomist, highlights the environmental impact of AI systems. According to his estimates, AI could account for between 32.6 and 79.7 million of CO₂ emissions. To put this into perspective, this level of emissions is comparable to the entire carbon footprint of New York City in 2023, which stood at 52.2 million tons.

“AI could therefore end up with a climate footprint comparable to that of a major world city or other energy-intensive sectors,” he said.

Furthermore, these systems are expected to consume between 312.5 and 764.6 billion liters of water, an amount that rivals the total global annual consumption of bottled water.

AI's growing demand for energy

De Vries-Gao points out that AI's growing energy demand is primarily driven by data centers, which are rapidly increasing their electricity consumption due to AI workloads. Previous reports suggested that AI currently accounts for approximately 20% of global electricity consumption in data centers, and this figure is projected to rise to nearly 50% by the end of 2025.

Training a single, large AI model can consume as much electricity as several hundred homes use in a year. By 2025, AI could consume up to 82 terawatt-hours, equivalent to Switzerland's annual power consumption. AI models' energy consumption is primarily driven by inference—the real-time process of responding to user queries—rather than by their initial training.

Call for transparency

Despite the increasing environmental concerns, major tech companies often lack transparency regarding the energy and water consumption of their AI systems and data centers. This lack of disclosure makes it difficult to accurately assess the full environmental impact of AI.

De Vries-Gao and other experts advocate stricter reporting requirements, urging companies to provide detailed data on energy consumption, CO₂ emissions, and water use per data center, distinguishing between AI and non-AI applications.

“If we want AI to contribute to a sustainable future genuinely, we first need a clear picture of its environmental cost to society. That starts with transparency,” he added.

The paper was published in the scientific journal Patterns.