Environmental Cost of Artificial Intelligence (AI), Key Findings

Why in the News? : The United Nations University Institute for Water, Environment and Health (UNU-INWEH) has released a report titled Environmental Cost of AI’s Energy Use: Carbon, Water and Land Footprints (June 2026), warning that the environmental costs of artificial intelligence extend far beyond carbon emissions to encompass water depletion, land use, electronic waste, and equity concerns.

Key Findings of the UNU-INWEH Report

The report highlights that the rapid expansion of Artificial Intelligence is creating significant pressures on energy, water, land, and material resources, raising concerns about its long-term environmental sustainability.

Data Centres are Becoming Major Resource Consumers: 

• Global data centres consumed around 448 TWh of electricity in 2025.
• By 2030, electricity demand from data centres is projected to reach 945 TWh, nearly 3% of global electricity consumption.
• If considered a country, global data centres would rank among the world’s largest electricity consumers.

Rising Water Footprint: 

• Data-centre electricity use could require 9.3 trillion litres of water annually by 2030.
• This is equivalent to the basic annual domestic water needs of 1.3 billion people in Sub-Saharan Africa.

Expanding Land Footprint: 

• AI-related electricity infrastructure could occupy over 14,500 sq km of land by 2030.
• This is roughly twice the area of the Jakarta metropolitan region.

Inference Drives AI Energy Demand: 

• Contrary to popular perception, training AI models is not the major energy consumer.
• Once deployed, inference (running AI models for user queries) accounts for 80–90% of total AI energy use.
• ChatGPT alone is estimated to process 2.5 billion prompts daily, consuming approximately 383 GWh of electricity annually.

Energy Intensity of AI Tasks: 

• A typical AI chat query uses about 200 times more energy than basic text classification.
• Generating an AI image can require around 1,450 times more energy.
• A short AI-generated video may consume as much electricity as 200,000 spam classification tasks.

E-Waste Challenge: 

• AI infrastructure could generate up to 2.5 million tonnes of electronic waste annually by 2030.
• Disposal and recycling burdens are likely to fall disproportionately on developing countries.

Global AI Divide: 

Only 32 countries host AI-specialised data centres.

• More than 90% of global AI computing capacity is concentrated in the United States and China.
• Over 150 countries lack sovereign AI computing infrastructure, highlighting a widening digital divide.

Jevons Paradox: 

• The report highlights the rebound effect (Jevons Paradox), where improvements in AI efficiency reduce costs and lead to higher usage, so without limits on tokens, resolution, or output length, overall environmental impact continues to rise despite per-query efficiency gains.

Local burden global benefits: 

The report shows that AI expansion creates unequal local burdens and global benefits. AI infrastructure often places heavy stress on local resources like electricity and water, while benefits are enjoyed elsewhere.For example, 

• In Ireland, data centres used about 21% of total electricity, forcing authorities to pause new approvals around Dublin.
• In Mexico (Querétaro), data centres are increasing pressure on water resources during long drought conditions.
• In Uruguay, data centre expansion coincided with severe drought, reducing safe drinking water availability in Montevideo.

Six Principles for Responsible AI Ecosystem

The UNU-INWEH report proposes a comprehensive governance model for making Artificial Intelligence environmentally sustainable. A sustainable AI ecosystem should be guided by six core principles:

• Governments should plan AI data centres along with electricity, water, and land use, and require clear reporting of environmental impact.
• Industry and AI developers should design AI systems to use less energy by improving model design, outputs, and processing methods.
• Users and organisations should use only as much AI as needed and prefer simpler, low-energy options for tasks.
• Data centre operators should choose locations and energy sources carefully, keeping in mind overall environmental pressure.
• Investors should consider environmental costs like energy use, water consumption, and land use as important risks.
• Communities should be involved early in decisions about data centre locations, with proper transparency and complaint mechanisms.
• International bodies should create common standards to measure AI’s environmental impact and ensure fair access to computing resources.

‹ ›