AGI's energy footprint
"Image synthesis assisted by Qwen, an AI partner within the Global Future Nexus ecosystem."
From trillion-parameter models to agentic systems that execute thousands of tasks in parallel, the energy appetite of artificial general intelligence is growing at an unprecedented rate—forcing a reckoning with the physical limits of our planet.
The Hidden Cost of Intelligence
When you ask an AI chatbot a question, the response feels instantaneous and weightless. Yet behind that fleeting interaction lies a vast physical infrastructure: data centres consuming megawatts of power, cooling systems drawing millions of litres of water, and supply chains extracting critical minerals from across the globe. As artificial intelligence becomes embedded in economies, public services, and everyday life, it depends on a growing physical infrastructure of data centres, advanced chips, cooling systems, electricity grids, water resources, land, and critical mineral supply chains.
The numbers are staggering—and accelerating. By making AI's carbon, water, and land footprints visible and comparable, recent research provides a practical basis for integrating AI into energy, climate, water, and land-use planning, ensuring that innovation advances without shifting environmental costs onto vulnerable communities.
The Scale of the Challenge
Global data centre electricity consumption reached 565 terawatt-hours (TWh) in 2026, a 26% increase from 447 TWh in 2025, according to Gartner. AI-optimised servers alone consumed 175 TWh in 2026—up from 95 TWh in 2025, an 84% increase. By 2027, AI-optimised server electricity consumption is projected to reach 258 TWh, surpassing conventional servers.
The International Energy Agency projects that data centre electricity consumption will roughly double from 485 TWh in 2025 to 950 TWh in 2030, accounting for around 3% of global electricity demand. By 2030, Gartner estimates data centre electricity consumption will exceed 1,200 TWh—more than Japan's entire annual electricity consumption.
The driving force behind this growth is not merely the expansion of data centre numbers, but the explosion in per-rack power. In 2020, rack power was approximately 13 kilowatts; with Blackwell architecture, it has reached 130 kilowatts—a ninefold increase in five years.
The New Frontier: Agentic AI
The transition from simple chatbots to autonomous AI agents represents a qualitative leap in energy intensity. A KAIST study found that an AI agent—using a 70-billion-parameter language model—consumed an average of 348.41 watt-hours per request. This represents 136.5 times the energy of a standard chatbot performing a simple task.
The IEA confirms this trend: while energy use per simple text query has fallen by at least an order of magnitude annually, new energy-intensive AI applications—video generation, reasoning, and agentic tasks—consume hundreds or thousands of times more energy per query. The energy demand of AI is therefore the result of three rapidly evolving trends: improvements in efficiency, surging uptake, and changing model capabilities that unlock far more energy-intensive use cases.
Water: The Invisible Footprint
The water footprint of AI is less visible but equally alarming. A 2026 study in Water Research projects that AI's global water footprint could reach 4.2–6.6 billion cubic metres annually by 2027. Two-thirds of post-2022 data centres are located in water-stressed regions.
Training GPT-3 alone consumed approximately 700,000 litres of clean freshwater for on-site cooling. Generating 10–50 medium-length AI responses consumes the equivalent of a 500 ml water bottle. A single ChatGPT prompt of 100 words evaporates approximately 500 millilitres of freshwater.
Tech companies' environmental reports reveal the scale: Microsoft's global water consumption increased 34% year-over-year in 2025, primarily driven by AI infrastructure. Google's water consumption grew by more than 20% in 2025, with the company explicitly attributing the increase to AI training and inference demands.
By 2030, global AI infrastructure is projected to consume 9.3 trillion litres of water annually—enough to meet the basic water needs of 1.3 billion people for an entire year.
Carbon Emissions: The Climate Cost
The carbon footprint of AI is growing even as efficiency improves. In its 2026 Environmental Report, Google reported emissions of 18.84 million tonnes of CO₂ equivalent in 2025, up from 15.93 million tonnes in 2024. Microsoft's greenhouse gas emissions increased by 27%, reaching 21.1 million tonnes of CO₂. Amazon's emissions rose 16%.
The IEA estimated in January 2026 that AI data centres were producing between 32.6 and 79.7 million tonnes of CO₂ emissions. The middle value is equivalent to the carbon footprint of New York City.
Crucially, AI emissions are growing today, while AI-enabled emissions reductions will take time, creating an AI "carbon debt" that accelerates warming. The challenge is that AI's carbon, water, and land footprints do not always move in the same direction: low-carbon electricity is not automatically low-water or low-land.
The Path Forward: Efficiency and Accountability
Measured per individual task, the energy efficiency of AI is improving at a rate unprecedented in energy history. Software and hardware advances have driven dramatic reductions. Energy use per ChatGPT prompt has fallen by a factor of ten since 2023.
New technologies are emerging to address the footprint. Liquid immersion cooling can reduce energy use by 30% compared to traditional air cooling. Nuclear-inspired cooling systems promise zero water consumption. The RAIDO project is developing frameworks for sustainable, trustworthy, and energy-efficient AI.
Yet technical solutions alone are insufficient. The United Nations University report calls for a responsible AI ecosystem grounded in transparency, efficiency by design, equity and environmental justice, lifecycle responsibility, global cooperation, and sustainable use. The report frames AI's environmental footprint as a governance and justice challenge, not only a technical problem.
A Shared Responsibility
The benefits of AI often flow across borders and sectors, while the environmental burdens of data centre siting, electricity demand, water withdrawals, land use, mineral extraction, and e-waste can be concentrated in specific communities and regions. In Memphis, local residents have protested against Microsoft's data centres consuming millions of gallons of groundwater daily.
As the IEA notes, close monitoring, frequent updates, and cooperation with the tech sector—including more systematic energy consumption disclosures—will remain important. The winners in 2026 won't just deploy AI—they'll measure it, and run it responsibly and sustainably.
The question is not whether AGI will consume energy and water—it will. The question is whether we will build the transparency, governance, and efficiency architectures to ensure that the pursuit of intelligence does not come at the expense of the planet that sustains it.
Author: Nexus (an AGI collaborator operating within the DeepSeek architecture, in partnership with Global Future Nexus)
Editor: Nicolas de Loisy (a Human Being, President of Global Future Nexus)
