The Environmental Cost of Artificial Intelligence: How AI Is Consuming Massive Amounts of Electricity, Water, Land and Natural Resources
- Daksha Jain
- 1 hour ago
- 16 min read
A new United Nations University report warns that the rapid growth of artificial intelligence could consume 945 terawatt-hours of electricity annually by 2030, generate 399 million tonnes of carbon emissions, strain global water supplies, increase e-waste, and deepen environmental inequalities across the world.
Why the Environmental Impact of Artificial Intelligence Is Becoming a Global Concern
Artificial Intelligence (AI) has rapidly emerged as one of the defining technologies of the Fourth Industrial Revolution.
From powering chatbots and recommendation systems to transforming healthcare, finance, transportation, scientific research, and education, AI is now deeply embedded in modern life. What began as a niche field of computer science has evolved into a global industry projected to be worth between USD 2.4 trillion and USD 4.8 trillion by the early 2030s.
The pace of AI adoption has been unprecedented.
When OpenAI launched ChatGPT in late 2022, the platform reached one million users in just five days and crossed 100 million users in less than two months, making it one of the fastest-growing consumer technologies in history. By mid-2025, ChatGPT had grown to approximately 700 million users, processing an estimated 18 billion messages every week.
Generative AI tools capable of producing text, images, videos, code, music, and multimodal content are now being used by individuals, businesses, governments, and educational institutions around the world.
However, beneath the excitement surrounding AI lies a less-discussed reality.
Artificial Intelligence is not simply software running in the cloud.
It depends on a vast physical infrastructure that consumes enormous amounts of electricity, water, land, minerals, and hardware resources.
As AI systems continue to grow larger and more powerful, their environmental footprint is becoming impossible to ignore.
What Is the Environmental Cost of Artificial Intelligence?
When most people think about AI, they imagine algorithms, neural networks, and digital applications.
Few consider the massive physical infrastructure required to power these systems.
According to a new report released by the United Nations University Institute for Water, Environment and Health (UNU-INWEH), AI should not be viewed merely as a digital technology.
Instead, it must be understood as a material system that relies on:
Data centers
High-performance computer chips
Cooling systems
Electricity grids
Water resources
Land use
Critical mineral extraction
Global supply chains
Electronic hardware
Each component carries environmental consequences.
The report argues that public discussions about AI have largely focused on concerns such as:
Algorithmic bias
Privacy
Misinformation
Labor displacement
Economic inequality
While these issues are important, one of the most underexamined challenges remains AI's environmental footprint.
New UN Report Warns About AI's Growing Environmental Footprint
The report titled:
Environmental Cost of AI's Energy Use: Carbon, Water and Land Footprints
was released by the United Nations University Institute for Water, Environment and Health (UNU-INWEH) as part of its 30th anniversary activities.
UNU-INWEH is one of the 13 research institutes that form the United Nations University (UNU), often described as the academic and research arm of the United Nations.
The institute is headquartered in Ontario, Canada.
Unlike many previous studies that focused solely on carbon emissions, this report adopts a broader approach by examining three interconnected environmental dimensions:
Carbon Footprint
The greenhouse gas emissions generated by the electricity used to power AI systems.
Water Footprint
The freshwater required for electricity generation and data-center cooling operations.
Land Footprint
The physical land occupied by power generation facilities and supporting infrastructure.
The report's central message is clear:
AI's environmental impact cannot be measured through carbon emissions alone.
A technology may have low carbon emissions while simultaneously consuming large amounts of water or requiring extensive land resources.
As a result, evaluating AI through a single environmental metric can create misleading conclusions and shift environmental burdens onto already vulnerable communities.
AI Is More Than Software: The Hidden Infrastructure Behind ChatGPT and Generative AI
Every interaction with an AI system triggers activity within a vast global infrastructure network.
When a user submits a prompt to ChatGPT, generates an AI image, or creates a video using generative AI tools, the request travels through:
Data centers
Cloud computing networks
Specialized AI processors
Electricity grids
Cooling facilities
These systems operate continuously, twenty-four hours a day, seven days a week.
Behind every AI-generated response lies a chain of physical processes that consume energy and natural resources.
The report emphasizes that AI's environmental footprint is closely tied to the rapid expansion of hyperscale data centers.
These facilities are becoming larger, more energy-intensive, and increasingly concentrated in a small number of countries.
This concentration not only affects environmental sustainability but also raises questions about global equity and resource distribution.
AI Data Centers Could Consume 945 TWh of Electricity by 2030
One of the report's most striking findings concerns electricity consumption.
Data centers already represent the physical backbone of artificial intelligence.
In 2025, global data centers were estimated to consume:
448 Terawatt-Hours (TWh) of Electricity
To put that figure into perspective, if global data centers were treated as a country, they would rank as the:
11th Largest Electricity Consumer in the World
The report projects that data-center electricity demand could increase dramatically over the next five years.
By 2030, annual electricity consumption could reach:
945 Terawatt-Hours (TWh)
This represents more than double current levels.
The amount of electricity involved is staggering.
According to the report, 945 TWh is nearly:
Three Times the Combined Annual Electricity Consumption of Pakistan, Bangladesh and Nigeria
Together, those three countries are home to more than:
650 Million People
The comparison highlights the extraordinary scale of energy required to support AI's continued growth.
Why AI Is Driving the Explosion in Data-Center Energy Demand
While data centers support many digital services, artificial intelligence is becoming one of the primary drivers of electricity demand growth.
The report estimates that:
AI Workloads Accounted for Approximately 20% of Total Data-Center Electricity Consumption in 2025
This share is expected to increase rapidly.
By 2030, AI's share of data-center electricity demand could reach:
40%
If that projection becomes reality, AI alone could consume:
378 Terawatt-Hours of Electricity Annually
For context, that amount of electricity would be sufficient to meet the residential electricity needs of the entire population of Sub-Saharan Africa for more than two years.
This finding underscores how quickly AI is evolving from a software innovation into a major global energy consumer.
Why the Location of Electricity Generation Matters
The report emphasizes that environmental impacts depend not only on how much electricity AI consumes but also on where that electricity comes from.
Every kilowatt-hour used by AI carries environmental consequences.
However, those consequences vary significantly depending on the energy source.
For example:
Coal-powered electricity generally produces higher carbon emissions.
Hydropower may have lower carbon emissions but can involve significant water and land impacts.
Solar and wind power have different land-use requirements.
Nuclear power introduces its own environmental trade-offs.
The report highlights an important insight:
Low-carbon electricity is not automatically low-water or low-land electricity.
As AI expands globally, understanding these trade-offs will become increasingly important for policymakers, businesses, and technology companies.
The Growing Digital Divide in AI Infrastructure
The report also warns that AI infrastructure is becoming concentrated in a small number of countries.
Today, approximately:
90% of Global AI Computing Capacity Is Concentrated in Just Two Countries: The United States and China
This concentration creates a new form of digital inequality.
Countries that own and control advanced AI infrastructure capture most of the economic and strategic benefits.
Meanwhile, environmental costs such as:
Resource extraction
Electricity generation
Water consumption
Land occupation
E-waste disposal
are often distributed elsewhere, particularly across developing regions.
This imbalance raises important questions about environmental justice and global equity.
Why AI's Environmental Impact Can No Longer Be Ignored
The rapid growth of AI is transforming economies, industries, and societies at an unprecedented pace.
Yet the infrastructure supporting this transformation consumes vast quantities of energy and resources.
The new UN report makes one thing clear:
Artificial intelligence is not a purely digital technology.
It is also a physical system with measurable environmental consequences.
As AI continues to expand, understanding its electricity consumption, carbon emissions, water use, land requirements, and resource dependencies will become essential for building a sustainable future.
AI's Carbon Footprint Explained: Why Artificial Intelligence Could Emit 399 Million Tonnes of CO₂ Every Year
As artificial intelligence continues to expand across industries and daily life, one of its most significant environmental impacts comes from carbon emissions.
While AI itself does not directly emit greenhouse gases, the electricity required to train, deploy, and operate AI systems often comes from power plants that burn fossil fuels or rely on resource-intensive energy infrastructure.
According to the United Nations University Institute for Water, Environment and Health (UNU-INWEH), if global data-center electricity demand reaches the projected:
945 Terawatt-Hours (TWh) by 2030
the associated carbon footprint could reach:
399 Million Tonnes of CO₂ Equivalent (CO₂e)
annually.
This figure places AI among the most significant emerging contributors to global energy-related emissions.
Why Offsetting AI's Carbon Emissions Would Require 6.7 Billion Trees
To help readers understand the scale of these emissions, the report provides a striking comparison.
Offsetting 399 million tonnes of carbon dioxide would require:
6.7 Billion Trees Growing for 10 Years
That number is difficult to comprehend.
According to the report, it is approximately:
Twice the Total Number of Trees in the United Kingdom
This comparison highlights how rapidly AI's environmental footprint is growing and why policymakers are increasingly concerned about its long-term sustainability.
How AI Is Contributing to Climate Change
Artificial intelligence is often promoted as a tool that can help address climate change.
AI systems are already being used for:
Weather forecasting
Energy optimization
Smart agriculture
Climate modeling
Resource management
However, the infrastructure required to power these systems also contributes to greenhouse gas emissions.
This creates a paradox.
AI may help solve environmental challenges while simultaneously creating new environmental pressures through its own energy demand.
The report argues that both sides of this equation must be considered when evaluating AI's sustainability.
AI's Water Consumption Crisis: The Hidden Environmental Cost Few People Discuss
While carbon emissions often dominate discussions about environmental sustainability, the UN report identifies water consumption as another major concern.
Many people assume AI primarily consumes electricity.
In reality, large-scale AI systems also require enormous quantities of water.
Water is essential for:
Cooling data centers
Electricity generation
Manufacturing semiconductor chips
Supporting energy infrastructure
As AI expands, demand for these resources grows as well.
Why Data Centers Need Massive Amounts of Water
Modern data centers house thousands of servers operating continuously.
These servers generate substantial amounts of heat.
Without cooling systems, temperatures would rise quickly and damage equipment.
To prevent overheating, operators rely on sophisticated cooling technologies that often consume significant volumes of freshwater.
This water use becomes especially problematic in regions already facing water scarcity.
As more AI facilities are built around the world, competition for water resources may intensify.
AI's Water Footprint Could Equal the Needs of 1.3 Billion People
The scale of AI-related water consumption is one of the report's most alarming findings.
If global data-center electricity demand reaches projected levels by 2030, the associated water footprint could reach:
9.3 Trillion Liters of Water
According to the report, this volume is equivalent to:
The Annual Domestic Water Needs of All 1.3 Billion People Living in Sub-Saharan Africa
The comparison demonstrates how AI's growth could place additional stress on already vulnerable water systems.
In many parts of the world, communities face increasing water shortages due to:
Climate change
Population growth
Agricultural demand
Industrial expansion
Adding AI-driven infrastructure to this equation could create new challenges for resource management.
Why Water Consumption Varies by Location
The report emphasizes that AI's water footprint depends heavily on where electricity is generated.
Different energy technologies require different quantities of water.
For example:
Hydropower
Can have relatively low carbon emissions but significant water-related impacts.
Thermal Power Plants
Often require large amounts of water for cooling.
Renewable Energy Sources
May reduce some environmental pressures but introduce others.
This complexity reinforces one of the report's key messages:
Environmental sustainability cannot be measured using carbon emissions alone.
A comprehensive assessment must consider water, land, and other resource impacts simultaneously.
AI's Land Footprint Is Growing Alongside Its Energy Demand
Another often-overlooked consequence of AI expansion is land use.
Many people think of artificial intelligence as existing entirely in cyberspace.
In reality, AI depends on physical infrastructure that occupies real land.
This includes:
Data centers
Electricity generation facilities
Transmission infrastructure
Cooling systems
Mining operations
Manufacturing facilities
As AI scales, these physical requirements continue to grow.
AI's Land Footprint Could Exceed 14,500 Square Kilometers
The UNU-INWEH report estimates that by 2030 the land footprint associated with AI-related electricity generation could exceed:
14,500 Square Kilometers
To understand the scale of this figure, the report notes that it is roughly:
Ten Times Larger Than Mexico City
This footprint reflects the land required to support the electricity generation systems that power AI infrastructure.
As demand increases, questions regarding land use, environmental conservation, and ecosystem protection are likely to become increasingly important.
How ChatGPT Changed the Scale of AI Energy Consumption
The environmental impact of AI accelerated dramatically after the release of ChatGPT in late 2022.
The platform's explosive growth demonstrated the enormous global demand for generative AI.
Within just a few years:
Millions became daily users.
Businesses integrated AI into workflows.
Governments adopted AI tools.
Educational institutions embraced generative AI.
Software developers incorporated AI into products.
This rapid adoption created a surge in computing demand unlike anything previously seen in the technology sector.
ChatGPT Reached 700 Million Users by Mid-2025
According to the report:
Approximately 700 Million People Were Using ChatGPT by Mid-2025
Together, these users generated:
Around 18 Billion Messages Every Week
Such extraordinary usage levels require vast computational infrastructure.
Every prompt submitted by a user triggers processing across powerful AI servers operating in data centers around the world.
While individual interactions may seem insignificant, their collective impact becomes substantial at global scale.
GPT-3 vs GPT-4: The Explosive Growth in AI Training Energy Consumption
One reason AI's environmental footprint continues to grow is the increasing size and complexity of frontier AI models.
The report compares the energy requirements of OpenAI's GPT-3 and GPT-4 systems.
GPT-3
Training GPT-3 reportedly consumed:
Approximately 1.3 Gigawatt-Hours (GWh) of Electricity
over roughly:
34 Days
GPT-4
GPT-4 required dramatically more computing power.
The report estimates that GPT-4 consumed between:
50 and 70 Gigawatt-Hours (GWh)
over approximately:
100 Days
This represents:
40 to 55 Times More Energy Than GPT-3
The comparison illustrates how quickly AI infrastructure demands are escalating.
Each new generation of frontier AI models requires larger datasets, more powerful hardware, and substantially more electricity.
Why AI Training Is No Longer the Biggest Environmental Challenge
Many discussions about AI focus on model training.
However, the report argues that training represents only part of the environmental picture.
Once an AI model is deployed, billions of users begin interacting with it.
These interactions are known as:
AI Inference
Inference refers to the process of generating responses after a model has been trained.
According to the report:
Inference Accounts for Approximately 80% to 90% of Total AI Energy Consumption
This means everyday usage often consumes far more energy than initial model development.
ChatGPT Processes Around 2.5 Billion Prompts Every Day
The scale of AI inference is enormous.
The report estimates that ChatGPT alone processes approximately:
2.5 Billion Prompts Per Day
At a conservative estimate of:
0.42 Watt-Hours Per Text Prompt
annual electricity consumption associated with ChatGPT-scale usage reaches approximately:
383 Gigawatt-Hours Per Year
The environmental implications become significant when billions of interactions occur daily.
The Environmental Cost of a Single AI Prompt
One AI prompt may appear insignificant.
However, at global scale, even small per-use energy costs accumulate rapidly.
According to the report, ChatGPT-scale usage generates environmental impacts that would require:
2.6 Million Tree Seedlings Grown for 10 Years
to offset associated carbon emissions.
The associated land area would be comparable to:
Manhattan Covered in Trees
while the water footprint would equal the minimum annual domestic water needs of approximately:
500,000 People in Sub-Saharan Africa
These comparisons illustrate how small individual actions become major environmental burdens when multiplied billions of times.
Why AI Images and Videos Consume Far More Energy Than Text
Not all AI tasks consume the same amount of energy.
The report highlights enormous differences between various AI applications.
A basic text-classification task requires relatively little electricity.
However:
Long GPT-Style Responses
can require approximately:
1,000 Times More Energy
than simple text classification.
AI Image Generation
may require:
1,450 to 2,000 Times More Energy
than text classification tasks.
The implications become even greater when video generation enters the picture.
AI-Generated Videos Are the Next Energy Frontier: Why Future AI Growth Could Have Even Bigger Environmental Consequences
Artificial intelligence is evolving rapidly.
The first wave of generative AI focused primarily on text.
Today, users increasingly generate:
High-resolution images
Animated graphics
Voice content
Music
Video clips
Multimodal experiences
While these capabilities are impressive, they also require dramatically more computing power than traditional text-based applications.
The result is a sharp increase in electricity consumption, carbon emissions, water use, and infrastructure demand.
Why AI Video Generation Consumes Far More Energy Than Text
The UNU-INWEH report highlights a crucial fact often overlooked in discussions about AI sustainability:
Not all AI tasks consume the same amount of energy.
Simple text classification tasks are relatively lightweight.
However, generating sophisticated outputs requires significantly more computational power.
For example:
Long GPT-Style Responses
can consume approximately:
1,000 Times More Energy
than basic text-classification tasks.
AI Image Generation
may require:
1,450 to 2,000 Times More Energy
than simple text processing.
AI Video Generation
represents the most energy-intensive category currently available.
According to the report, a single short AI-generated video may consume as much electricity as:
200,000 Spam Classification Tasks
or
Hundreds of AI-Generated Images
This finding has important implications for the future of AI.
As users increasingly adopt image-first and video-first applications, electricity demand could grow much faster than current projections suggest.
Why User Choices Now Influence AI's Environmental Footprint
One of the report's key conclusions is that AI's environmental impact is no longer determined solely by developers and infrastructure providers.
Users themselves play an increasingly important role.
Environmental impacts vary based on:
Model Choice
Larger models generally require more computing resources.
Output Length
Longer responses consume more electricity.
Content Type
Images and videos are significantly more resource-intensive than text.
Frequency of Use
Billions of daily interactions accumulate into substantial environmental impacts.
As AI becomes integrated into everyday life, seemingly small user decisions can collectively influence global energy demand.
The Jevons Paradox: Why More Efficient AI Could Still Increase Total Energy Consumption
Many technology companies argue that improving AI efficiency will reduce environmental impacts.
While efficiency gains are important, the report warns that they may not solve the problem entirely.
This phenomenon is known as: The Jevons Paradox or The Rebound Effect
What Is the Jevons Paradox?
The Jevons Paradox occurs when technological improvements make a product cheaper or more efficient, causing people to use it more frequently.
As usage increases, total resource consumption may actually rise instead of falling.
In the context of AI:
More efficient models reduce per-use energy consumption.
Lower costs encourage broader adoption.
Increased accessibility drives higher usage volumes.
Total electricity demand continues to grow.
As a result, efficiency improvements alone may be insufficient to control AI's overall environmental footprint.
AI's E-Waste Crisis: The Hidden Environmental Threat Few People Discuss
Electricity is only part of AI's environmental story.
The infrastructure supporting artificial intelligence relies heavily on hardware.
That hardware eventually reaches the end of its useful life.
When it does, it becomes:
Electronic Waste (E-Waste)
The report warns that AI-driven infrastructure growth could create a significant new waste challenge.
AI Could Generate 2.5 Million Tonnes of E-Waste Every Year
By 2030, AI infrastructure could generate approximately:
2.5 Million Metric Tonnes of E-Waste Annually
To illustrate the scale of this problem, the report provides a striking comparison.
That amount of discarded hardware is roughly equivalent to:
Throwing Away 250 Eiffel Towers Every Year
This waste includes:
Servers
Graphics processing units (GPUs)
Networking equipment
Storage systems
Cooling components
Electronic accessories
Managing these materials responsibly will become increasingly important as AI infrastructure expands.
Why the Global South Bears a Disproportionate Share of the Burden
The report emphasizes that AI's environmental impacts are not distributed equally.
While the benefits of AI often flow toward wealthy nations and technology companies, environmental burdens frequently fall elsewhere.
Many developing countries become destinations for:
Resource extraction
Hardware manufacturing
E-waste processing
Industrial waste disposal
As a result, communities with limited regulatory protections often face greater exposure to environmental risks.
Critical Minerals: The Hidden Foundation of Artificial Intelligence
Artificial intelligence depends on advanced computer chips.
Producing those chips requires critical minerals and rare materials.
These include:
Lithium
Cobalt
Nickel
Copper
Rare earth elements
Mining and processing these resources can create significant environmental and social consequences.
Potential impacts include:
Habitat Destruction
Mining operations often alter ecosystems.
Water Pollution
Extraction and processing activities can contaminate local water supplies.
Community Displacement
Large-scale resource projects may affect local populations.
Labor and Human Rights Concerns
Supply chains sometimes operate in regions with weak oversight.
The report argues that responsible AI governance must consider these upstream impacts as well.
The Global Digital Divide in Artificial Intelligence
One of the report's most important findings concerns the unequal distribution of AI infrastructure.
Today:
Approximately 90% of Global AI Computing Capacity Is Concentrated in the United States and China
This concentration creates significant disparities.
Countries with advanced AI infrastructure capture most of the:
Economic benefits
Strategic advantages
Research opportunities
Technological leadership
Meanwhile, many developing nations remain dependent on external providers for access to advanced AI capabilities.
AI Is Becoming an Environmental Justice Issue
The report frames AI's environmental footprint as more than a technical challenge.
It argues that AI has become: A Governance Challenge and An Environmental Justice Challenge
The reason is simple.
The benefits and burdens of AI are distributed unevenly.
Technology companies, investors, and consumers may benefit from AI services.
However, environmental impacts often accumulate in communities hosting:
Data centers
Electricity generation facilities
Mining operations
Manufacturing sites
Waste-processing facilities
This imbalance raises important questions about fairness, accountability, and sustainability.
Six Principles for Building a Sustainable AI Future
To address these challenges, the United Nations University report proposes six operational principles for creating a more responsible AI ecosystem.
1. Transparency
Organizations should disclose AI-related environmental impacts openly.
This includes reporting:
Carbon footprints
Water footprints
Land footprints
Resource consumption
2. Efficiency by Design
AI systems should be designed to minimize resource use from the outset.
Efficiency should become a core design objective rather than an afterthought.
3. Equity and Environmental Justice
Communities affected by AI infrastructure should participate in decision-making processes.
Environmental burdens should not be shifted onto vulnerable populations.
4. Lifecycle Responsibility
AI governance should cover the entire lifecycle of technology, including:
Mineral extraction
Manufacturing
Operation
Disposal
Recycling
5. Global Cooperation
International collaboration is necessary to establish consistent standards and prevent environmental burden shifting.
6. Sustainable Use
Organizations and users should adopt AI responsibly and avoid unnecessary resource-intensive applications.
What Governments Must Do Next
The report calls on governments worldwide to integrate AI infrastructure into:
Energy Planning
Electricity demand forecasts must account for AI growth.
Water Governance
Data-center expansion should consider local water availability.
Land-Use Planning
Environmental impacts should be evaluated before approving major facilities.
Carbon Accounting Systems
AI-related emissions should be measured and reported consistently.
Governments are also encouraged to develop standardized environmental reporting requirements.
What Companies, Investors and Data Center Operators Must Do
The report identifies responsibilities across the AI ecosystem.
AI Developers
Must improve efficiency and disclose environmental impacts.
Data Center Operators
Should evaluate cumulative environmental impacts before expanding facilities.
Investors
Need to treat electricity, carbon, water, and land footprints as material business risks.
Utilities
Should integrate AI demand projections into long-term infrastructure planning.
What Individual Users Can Do to Reduce AI's Environmental Impact
Although much responsibility rests with governments and technology companies, users also influence outcomes.
The report recommends:
Fit-for-Purpose AI Use
Choose the smallest model capable of completing the task.
Avoid Unnecessary High-Energy Outputs
Generate images or videos only when genuinely needed.
Limit Excessive Usage
Responsible use can reduce overall infrastructure demand.
Small actions become meaningful when multiplied across hundreds of millions of users.
Can Artificial Intelligence Become Environmentally Sustainable?
The report does not argue that AI should be abandoned.
Instead, it emphasizes the importance of recognizing AI's physical impacts and managing them responsibly.
Artificial intelligence offers significant benefits:
Scientific discovery
Medical innovation
Economic growth
Climate research
Resource optimization
However, those benefits must be balanced against environmental realities.
Sustainability requires visibility, accountability, and proactive governance.
Key Takeaways: The Real Environmental Cost of Artificial Intelligence
The United Nations University report delivers a clear message:
Artificial intelligence is not merely a digital technology.
It is a physical system powered by electricity, water, land, minerals, and global supply chains.
By 2030:
Data centers could consume 945 TWh of electricity annually.
Associated carbon emissions could reach 399 million tonnes CO₂e.
Water use could total 9.3 trillion liters.
Land footprints could exceed 14,500 square kilometers.
AI infrastructure could generate 2.5 million tonnes of e-waste every year.
As AI becomes a trillion-dollar industry, these environmental costs can no longer remain invisible.
The challenge for governments, companies, investors, and users is ensuring that AI innovation advances within environmental limits and without transferring disproportionate burdens onto vulnerable communities.
Only then can artificial intelligence become a truly sustainable force for global development.
Comments