The Hidden Costs of AI
The Energy Cost of a Single Chatbot Query
Every day, we casually ask AI questions like, “What should I have for lunch today?” or “Please polish this email.” The moment we press Enter with a single finger, we are consuming something. According to Sasha Luccioni, a computer scientist who is a leading AI researcher and climate lead, a single ChatGPT query consumes approximately 2.9 Wh of electricity. That’s equivalent to leaving a TV on for about 9 seconds. On its own, that number doesn’t seem like much. But when you multiply it by the hundreds of millions of AI queries made worldwide every day, the story changes completely.
Photograph from the GigaLight, from “Three Illustrative Examples Demonstrating Liquid Cooling as the Optimal Choice for Cutting-Edge Data Centers”
A Thirst for Power: Giant Power Plants That Never Shut Down
Data centers never sleep. As long as the internet keeps running, servers operate day and night, emitting heat. According to the IEA 2026 survey, global data center electricity consumption is expected to exceed 1,000 TWh in 2026. This is equivalent to Japan’s annual total electricity consumption. Apparently, a single industry now consumes as much electricity as a nation.
The impact is even more significant in smaller countries. Furthermore, the IEA 2026 report projects that by 2026, data centers will account for 32% of Ireland’s total national electricity demand. Fearing grid saturation, the Irish government has effectively restricted the construction of new data centers in the Dublin area. The power issue is not merely a matter of utility bills. The more we rely on fossil fuel-based power grids, the more AI growth is accompanied by carbon emissions. The Irony is clear: “Advances” in intelligence lead to a regression in climate action.
Water scarcity: The invisible river used to cool servers
After electricity comes water. When servers run nonstop, they generate enormous amounts of heat. To cool that heat, data centers consume more water than you might imagine. According to Brookings 2025/2026, a single large hyperscale data center consumes approximately 19 million liters, or 5 million gallons, of water per day. This is equivalent to the daily water usage of a city with a population of 50,000. It is no coincidence that areas with high concentrations of data centers are experiencing rising complaints about groundwater depletion, a surge in complaints about municipal water supplies, and conflicts over municipal water supplies.
As criticism grew, major tech companies responded. According to the 2026 Google Water Stewardship Project Portfolio, Google and Microsoft have pledged to become “Water Positive” by 2030, aiming to return more water to nature than they consume. This goes beyond conservation toward active restoration of water resources. However, the gap between ambition and implementation remains wide. This is because, as demand for AI surges, the total amount of water required for cooling is expected to rise accordingly.
The Emergence of Alternatives: Reshaping the Landscape with Nuclear Power and Liquid Cooling
Faced with criticism, the industry is pursuing solutions in two directions: restructuring energy supply and rethinking cooling technologies.
First, nuclear power is regaining attention. According to Amazon News, big tech companies, including NVIDIA, are signing direct power supply contracts with small modular reactors (SMRs) and existing nuclear power plants to secure a stable power supply dedicated to data centers. With its ability to provide large-scale power without carbon emissions, nuclear energy is emerging as a practical solution.
Meanwhile, cooling technology is going through a shift. Instead of traditional air cooling, “Immersion Cooling” technology, where servers are submerged in non-conductive liquid, is gaining traction, significantly improving efficiency by over 27%. While these solutions are not yet fully adopted, the direction is clear. The industry is moving beyond incremental fixes toward a fundamental redesign of its infrastructure.
Toward Sustainable Intelligence
The pace of AI development is remarkable, but it must not come at the expense of strained power grids or drying rivers.“Green AI” is no longer an idealistic concept; it has become a prerequisite for the survival of the AI industry.
The “Circular Data Center” model, which uses heat generated by data centers to heat nearby buildings and connects renewable energy with AI workloads on a local scale, is already being tested in parts of Northern Europe. It represents a shift from energy-consuming machines to an ecosystem that recycles energy.
The fact that each query consumes 2.9 Wh does not mean we should stop using AI. However, recognizing that cost, aligning technology, policy, and user behavior to reduce it, is the first step toward a more sustainable form of intelligence.
Footnotes, MLA citation
Amazon News Staff. “Amazon's Investment in Small Modular Reactors for Carbon-Free Energy.” About Amazon, 16 Oct. 2025, www.aboutamazon.com/news/sustainability/amazon-smr-nuclear-energy. Accessed 29 Mar. 2026.
Gigalight. “The Paradox of Intelligence: AI Data Centers' Energy and Water Consumption Challenges.” Gigalight Insights, 17 April. 2024, www.gigalight.com/news-events/insights-7992.html. Accessed 29 Mar. 2026.
Google. "Google Expands Hamina Data Center and Launches Off-site Heat Recovery Project." Google News, 20 May 2024, https://blog.google/around-the-globe/google-europe/finland-data-center-expansion-heat-recovery/. Accessed 29 Mar. 2026.
Google Sustainability Team. 2026 Water Stewardship Project Portfolio: Progress Toward Our 2030 Goal. Google, Mar. 2026, sustainability.google/reports/water-stewardship-project-portfolio/. Accessed 29 Mar. 2026.
Institute of International and European Affairs (IIEA). “Data Centres in Ireland: The State of Play and Future Challenges.” IIEA Blog, 18 Dec. 2025, www.iiea.com/blog/data-centres-in-ireland-the-state-of-play. Accessed 29 Mar. 2026.
International Energy Agency (IEA). Electricity 2026: Analysis and Forecast to 2026. IEA,2026, www.iea.org/reports/electricity-2026. Accessed 29 Mar. 2026.
Luccioni, Sasha. "AI and the Environment: A Primer." Hugging Face Blog, 1 Nov. 2023, huggingface.co/blog/sasha/ai-environment-primer. Accessed 29 Mar. 2026.
Villasenor, John. “The AI Paradox: Data Centers, Water Consumption, and Environmental Sustainability.”
Brookings Institution, 20 Nov. 2025, www.brookings.edu/articles/ai-data-centers-and-water/. Accessed 29 Mar. 2026.
){kind=small}
){kind=small}
){kind=small}
){kind=small}