Generative AI is consuming vastly more energy than traditional search engines, with a single ChatGPT query using 2.9 watt-hours of electricity compared to Google's 0.3 watt-hours, according to new analysis. As AI adoption accelerates globally, the environmental cost of these tools is becoming impossible to ignore. The International Energy Agency (IEA) warns that AI data centers alone could demand nearly as much energy by 2030 as Japan currently consumes, yet only half of that demand is likely to come from renewable sources.

How Much Energy Does AI Actually Consume?

The scale of AI's energy appetite is staggering. If ChatGPT replaced the 9 billion Google searches conducted daily, the electricity demand would require almost 10 terawatt-hours yearly, equivalent to the annual electricity consumption of 1.5 million European Union citizens. Goldman Sachs analysts expect AI to represent about 19% of data center power demand by 2028, with the AI revolution causing overall data center power demand to grow by 160% by 2030.

The computing power required for AI is doubling every 100 days and is projected to increase by more than a million times over the next five years, according to recent analysis. This exponential growth means the environmental impact will accelerate rapidly unless significant changes are made to how AI systems are built and deployed.

What Types of AI Tasks Produce the Most Carbon Emissions?

Not all AI queries are created equal when it comes to environmental impact. Different types of AI tasks require vastly different amounts of energy, creating a hierarchy of carbon intensity:

• Image Generation: Generating images is one of the most energy and carbon-intensive AI tasks. A single AI-generated image can use as much energy as half a smartphone charge using the least efficient model, with the most carbon-intensive image generation model producing emissions equivalent to 4.1 miles driven by an average gasoline-powered vehicle for 1,000 image requests.
• Video Generation: Videos are far more intensive than text or images. Every Sora 2 video generated consumes 1 kilowatt-hour of energy, uses 4 liters of water, and emits 466 grams of carbon.
• Text Generation: AI-generated text requires significantly less energy than images. The least carbon-intensive text model generates 6,833 times less carbon than image models, with 1,000 text queries using as little as 9% of a full smartphone charge.
• Complex Reasoning Queries: Queries asking AI chatbots to think logically and reason require more energy than straightforward questions. Queries about philosophy or abstract algebra lead to more carbon emissions than simple factual questions, with some complex prompts generating 50 times the carbon emissions of simpler ones.

Training large language models (LLMs), which are AI systems trained on massive datasets to generate human-like text, is also heavily energy-intensive. Training GPT-3, one of the most popular models, produced 626,000 pounds of carbon dioxide, equivalent to approximately 300 round-trip flights between New York and San Francisco, nearly five times the lifetime emissions of an average car.

However, the ongoing use of these models may be even more damaging than training them. Inference, the process where AI makes predictions and responds to user queries, impacts the environment just as much or more than training. For popular models like ChatGPT, it could take just a couple of weeks or months for usage emissions to exceed training emissions.

What Is the Hidden Water Cost of AI?

Beyond carbon emissions, AI systems consume enormous quantities of fresh water for cooling data centers. A short conversation of 20 to 50 questions and answers with ChatGPT costs half a liter of fresh water. Training GPT-3 in Microsoft's U.S. data centers directly evaporates 700,000 liters of clean fresh water, enough water to produce 370 BMW cars or 320 Tesla electric vehicles.

This water consumption is particularly concerning in regions already facing water scarcity. As AI data centers proliferate globally, competition for fresh water resources will intensify, creating potential conflicts between tech companies and local communities.

What Do Projections Show for AI's Climate Impact by 2030?

The trajectory is alarming. By 2030, data centers are predicted to emit triple the amount of carbon dioxide annually compared to a scenario without the AI boom. The projected emissions of 2.5 billion tonnes of greenhouse gases equate to roughly 40% of the United States' current annual emissions. In the United States alone, researchers from Cornell University projected that by 2030, the current rate of AI growth would release 24 to 44 million metric tons of carbon dioxide into the atmosphere each year, equivalent to dumping 5 to 10 million more cars on U.S. roadways.

The AI boom fueled by mass use of generative AI released roughly as much carbon dioxide into the atmosphere as New York City in 2025, a city with an urban area of over 20 million people and annual emissions exceeding 50 million metric tonnes.

The energy demand from dedicated AI data centers is set to more than quadruple by 2030, according to the IEA. This growth is outpacing renewable energy expansion, creating a widening gap between energy demand and clean energy supply.

How Are Companies Accounting for AI's Environmental Impact?

Transparency remains a significant problem. According to analysis from The Guardian covering 2020 to 2022, real emissions from the data centers of AI pioneers like Meta could be 7.62 times higher, or 662% more, than reported due to creative accounting. This discrepancy suggests that publicly disclosed environmental impacts may significantly underestimate the true carbon footprint of AI operations.

The escalating spread of data centers to support AI and large language models is moving to the forefront of environmental litigation around the world. In an analysis of over 3,600 climate-related lawsuits, the latest annual review of climate litigation by the London School of Economics found a soaring number of cases challenging the energy sources, emissions, water consumption, and air pollution of data centers.

While traditional AI has demonstrated potential to positively impact the planet with tools that can predict weather, identify pollution, improve waste management, and clean up marine plastic, generative AI presents a different challenge. A study by the Boston Consulting Group stated that if AI is used wisely, it could help mitigate 5 to 10% of greenhouse gas emissions by 2030, but the current trajectory of generative AI deployment suggests this potential is being squandered.