Artificial intelligence has become one of the defining technologies of this decade. It writes content, powers customer support, predicts equipment failures, accelerates scientific research, and helps businesses automate repetitive tasks. With generative AI tools becoming part of everyday life, organizations are investing billions of dollars into expanding AI capabilities.

However, behind every AI-generated response lies a growing environmental challenge.

Training and operating Large Language Models (LLMs) like ChatGPT, Gemini, Claude, and Llama requires enormous computing power. That computing power comes from data centers filled with high-performance processors running around the clock, consuming electricity, water, and other resources. A single AI text prompt is estimated to use somewhere between roughly 0.03 and 1.14 grams of CO2-equivalent, depending on the model and query complexity — a wide range that reflects just how differently AI systems are built.

The discussion is no longer whether AI is transforming industries — it clearly is. The more important question is whether this transformation can happen sustainably.

Quick Answer

The carbon footprint of AI comes from the energy-intensive computing infrastructure required to train and run large language models. A single query is estimated to emit roughly 0.03 to 1.14 grams of CO2e depending on the model, while global AI systems combined could produce between 32.6 and 79.7 million tonnes of CO2 in 2025 — comparable to New York City’s total annual emissions, according to a 2025 VU Amsterdam study. AI can also help reduce emissions elsewhere by improving renewable energy forecasting, optimizing industrial operations, and enabling smarter resource management. The future impact of AI will depend on cleaner electricity, more efficient hardware, and responsible AI development.

Key Takeaways

• Artificial intelligence relies on energy-intensive data centers and specialized computing hardware.
• A single AI query emits an estimated 0.03–1.14 grams of CO2e; more efficient models (like Google’s Gemini, at roughly 0.24 Wh per query) use meaningfully less energy than denser ones.
• A 2025 study from VU Amsterdam estimates global AI systems could emit 32.6–79.7 million tonnes of CO2 in 2025 — comparable to New York City’s total emissions.
• Water consumption for cooling data centers is an emerging sustainability concern.
• Efficient model architectures — such as DeepSeek’s — are proving that smaller, well-optimized models can rival larger ones on performance while using a fraction of the energy.

AI’s Environmental Impact by the Numbers

The rapid growth of artificial intelligence is reshaping global energy demand. Here are some of the most significant figures driving the conversation:

• Global data centers consumed approximately 415 terawatt-hours (TWh) of electricity in 2024, roughly 1.5% of global electricity demand, according to the International Energy Agency (IEA).
• The IEA projects electricity demand from data centers could reach around 945 TWh by 2030, with AI as one of the primary growth drivers.
• A 2025 study by researcher Alex de Vries-Gao (VU Amsterdam), published in *Patterns*, estimates AI systems alone could produce 32.6 to 79.7 million tonnes of CO2 in 2025 and consume 312.5–764.6 billion liters of water — a carbon footprint in the same range as New York City’s.
• Per-query estimates vary widely by model: Google’s Gemini uses an estimated 0.24 Wh per query, while OpenAI’s ChatGPT is estimated at roughly 0.34 Wh, and denser models like Mistral can run closer to 1.14 grams CO2e per query.
• The Allen Institute for AI, accounting for the full lifecycle of its own smaller model series (hardware manufacturing, development, and training), calculated total emissions of 493 metric tons of CO2 — equivalent to powering roughly 98 US homes for a year.
• Leading technology companies — including Microsoft, Google, Amazon, and Meta — are investing hundreds of billions of dollars in AI infrastructure, including new hyperscale data centers and GPU clusters.

These figures highlight an important reality: AI is no longer simply a software innovation — it is becoming an energy infrastructure challenge.

AI Is Fueling a New Energy Boom

Artificial intelligence is changing the way electricity is consumed across the world. Unlike traditional software applications, generative AI performs billions of mathematical calculations every time a user submits a prompt. As millions of people interact with AI assistants daily, computing demand continues to rise.

This has created what many researchers describe as an “AI energy boom.” Governments, utilities, and technology companies are racing to expand electricity generation while simultaneously increasing investments in renewable energy to support growing digital infrastructure.

Interestingly, improvements in AI efficiency have not necessarily reduced total energy consumption. This phenomenon is known as the rebound effect. As AI models become faster, cheaper, and more accessible, businesses deploy them across more applications and consumers use them more frequently — so overall electricity demand keeps rising despite hardware efficiency gains. Rather than slowing energy growth, better AI often encourages broader adoption.

Why AI Requires So Much Energy

Artificial intelligence relies on massive computational workloads. Before an AI assistant can answer questions or generate content, it must first be trained using enormous datasets containing books, websites, scientific papers, images, code, and other information.

During training, thousands of specialized Graphics Processing Units (GPUs) work continuously for weeks or even months, processing trillions of calculations. After deployment, AI continues consuming energy through inference — the process of generating responses for users.

Every prompt sent to an AI chatbot activates servers inside large-scale data centers equipped with high-performance processors, networking equipment, storage systems, and sophisticated cooling infrastructure. Unlike conventional web searches, generative AI produces entirely new responses rather than retrieving existing information, making each interaction computationally more demanding — one widely cited estimate puts a single AI query at several times the energy cost of a standard web search, though exact multiples vary significantly by model and query type.

Training vs. Inference: Understanding AI’s Two Energy Costs

When discussing AI’s environmental impact, it’s important to distinguish between training and inference.

	Training	Inference
Definition	Building or updating a model on massive datasets	Every user interaction with an already-trained model
Frequency	Periodic (major cycles every months)	Continuous — billions of times per day globally
Energy per event	Very high (weeks of continuous GPU compute)	Low per query (fractions of a watt-hour)
Example figure	Allen Institute for AI’s model series: 493 metric tons CO2 for full lifecycle	Google Gemini ~0.24 Wh; ChatGPT ~0.34 Wh per query
Long-term trend	Emissions concentrated in discrete cycles	Now estimated to account for up to 90% of a model’s total lifecycle energy use as adoption scales

While a single AI prompt consumes far less energy than training, billions of daily interactions across search engines, enterprise software, and digital assistants collectively create a substantial environmental footprint. Many researchers now believe inference — not training — is becoming the largest source of AI-related electricity demand as adoption continues to grow. This is why the sustainability of AI depends not only on building efficient models, but on reducing the energy cost of every query processed.

The Hidden Carbon Footprint of AI

Artificial intelligence doesn’t directly produce greenhouse gases. Instead, its environmental impact comes from the infrastructure that powers it.

Every AI query travels through servers housed in large-scale data centers, where powerful GPUs process billions of calculations in seconds. These facilities require continuous electricity for computing, networking, storage, and cooling. If that electricity comes from fossil fuels, AI’s carbon footprint increases significantly.

The environmental impact of AI depends on several factors:

• The size and complexity of the AI model
• The number of users and daily queries
• The efficiency of GPUs and processors
• The energy source powering the data center
• Cooling technologies and infrastructure design

A data center powered primarily by wind, solar, or hydropower produces significantly lower emissions than one relying on coal or natural gas. This is one reason DeepSeek attracted global attention in the AI efficiency conversation: its architecture demonstrated that a smaller, well-optimized model can deliver competitive performance while requiring meaningfully less compute than larger dense models — a real-world proof point for the “smaller, more efficient models” trend discussed later in this article.

AI’s Hidden Water Footprint

Carbon emissions often dominate discussions about AI sustainability, but water consumption is becoming an equally important issue.

High-performance servers generate enormous amounts of heat. Many facilities use water-based cooling systems that evaporate or circulate freshwater to regulate temperatures. Researchers from the United Nations University Institute for Water, Environment and Health (UNU-INWEH) describe AI as having three interconnected environmental footprints:

• Carbon footprint
• Water footprint
• Land footprint

The VU Amsterdam study cited earlier estimated AI systems’ 2025 water footprint at 312.5–764.6 billion liters — in the range of the global bottled water industry’s annual output. In regions already experiencing water scarcity, expanding AI infrastructure raises important questions about resource allocation and long-term sustainability.

To address these concerns, technology companies are investing in:

• Closed-loop water recycling systems
• Air cooling technologies
• Liquid immersion cooling
• AI-powered cooling optimization
• Heat recovery systems that reuse waste heat

The Rise of AI Data Centers

Artificial intelligence is driving one of the largest waves of digital infrastructure investment in history. Technology giants including Microsoft, Google, Amazon, Meta, and Oracle are expanding hyperscale data centers to meet soaring demand for AI services.

According to the International Energy Agency, electricity demand from AI-focused data centers is expected to grow rapidly throughout the remainder of this decade. This growth presents both opportunities and challenges: expanding infrastructure supports innovation and economic growth, but it also increases pressure on electricity grids and highlights the importance of accelerating renewable energy deployment.

For more insight into how clean energy infrastructure is evolving to meet rising electricity demand, explore our analysis of the Middle East’s accelerating renewable energy transition

Can AI Help Fight Climate Change?

While AI consumes significant resources, it is also becoming one of the most powerful tools available for tackling environmental challenges.

Renewable Energy Forecasting

AI improves predictions for solar and wind generation, enabling electricity grids to balance supply and demand more effectively.

Smart Energy Management

Buildings equipped with AI-powered systems automatically optimize lighting, heating, cooling, and ventilation, reducing unnecessary energy consumption.

Manufacturing Optimization

Factories use predictive maintenance to identify equipment failures before they occur, minimizing downtime while reducing energy waste.

You can explore how digital technologies are transforming industrial sustainability in our article on advanced manufacturing innovations improving environmental performance

Precision Agriculture

AI analyzes satellite imagery, weather data, and soil conditions to optimize irrigation, fertilizer application, and crop management, reducing resource use while improving productivity.

Climate Research

Scientists increasingly rely on AI to model climate scenarios, monitor biodiversity, predict extreme weather events, and improve environmental decision-making.

For a deeper look at real-world applications, read our feature on how artificial intelligence is accelerating climate solutions, where we examine the technologies helping businesses reduce their environmental impact.

Sustainability Trends Shaping the Future of AI

Renewable-Powered Data Centers

Technology companies are signing long-term renewable energy agreements to reduce operational emissions.

Smaller, More Efficient Models

Developers are creating compact AI models capable of delivering high performance while requiring fewer computing resources — DeepSeek is the clearest recent example of this approach reaching mainstream attention.

Energy-Efficient AI Chips

New generations of processors are significantly improving performance per watt, reducing electricity consumption for AI workloads.

Carbon-Aware Computing

Some cloud providers are beginning to schedule computational workloads for when renewable electricity availability is highest.

Circular Economy Practices

Manufacturers are extending hardware lifecycles through refurbishment, recycling, and component reuse to reduce electronic waste.

Expert Perspective

Artificial intelligence should not be viewed solely as either an environmental problem or an environmental solution. Its long-term impact will depend on the choices organizations make today. The gap between the most efficient models (Gemini’s estimated 0.24 Wh per query) and the least efficient (dense models running closer to 1.14 grams CO2e) shows that architecture choices matter as much as electricity source. Businesses that invest in renewable-powered infrastructure, transparent emissions reporting, efficient model design, and responsible data-center operations — and that disclose real figures the way the Allen Institute for AI and VU Amsterdam researchers have — will be better positioned to balance innovation with sustainability.

Final Thoughts

Artificial intelligence is reshaping industries at an unprecedented pace, but its rapid expansion also brings new environmental responsibilities. The carbon footprint of AI extends beyond electricity use to include water consumption, infrastructure development, and the growing demand for critical resources that power digital technologies — with global estimates from researchers like Alex de Vries-Gao putting 2025’s AI-driven emissions in the same range as a major city’s total footprint.

At the same time, AI is proving to be an invaluable ally in the fight against climate change. From improving renewable energy forecasting and optimizing industrial efficiency to supporting biodiversity monitoring and climate research, AI has the potential to accelerate the transition to a more sustainable economy.

The challenge is ensuring that AI’s environmental benefits outweigh its operational costs. Achieving this balance will require cleaner electricity, more efficient computing architectures — as DeepSeek and the smaller-model trend illustrate — responsible infrastructure investment, and greater transparency from technology companies.

Ultimately, the future of AI will be measured not only by the intelligence of its algorithms but also by the sustainability of the systems that support them.