AI Data Centers Could Emit More CO₂ in 2025 Than New York City, Raising Environmental Concerns

New research warns that AI data centers could produce more carbon dioxide emissions in 2025 than New York City, underscoring the growing environmental impact of energy-intensive artificial intelligence systems. The findings, published in the journal Patterns, highlight the significant climate and resource challenges associated with rapidly expanding AI infrastructure.

Data scientist Alex de Vries-Gao, the lead researcher, estimates that AI-related emissions in 2025 could range from 33 to 80 million metric tons of CO₂. If realized, this level of output could surpass the annual emissions of entire countries, including Chile, Czechia, and Romania. The study emphasizes that the exponential growth of AI workloads—particularly large-scale machine learning models—requires massive computational power, which in turn drives electricity demand and greenhouse gas emissions.

Beyond carbon emissions, the research also examines water usage, estimating that AI data centers could consume between 312 and 767 billion litres of water in 2025. This level of consumption would exceed the total annual usage of the global bottled water industry, reflecting the intensive cooling requirements for high-performance computing systems.

De Vries-Gao cautioned that these figures carry significant uncertainty, largely due to the limited transparency provided by major AI companies regarding energy usage, data center efficiency, and environmental practices. The study calls for greater reporting standards, energy-efficient AI design, and sustainable infrastructure strategies to mitigate the environmental footprint of AI technologies as their adoption continues to expand worldwide.

This research adds to growing concerns about the sustainability of AI, suggesting that while the technology promises innovation and economic growth, it also poses serious environmental challenges that require urgent attention from industry leaders and policymakers.