Recognize the Environmental Impact of AI
Learning Objectives
After completing this unit, you’ll be able to:
- Identify factors that contribute to AI’s environmental footprint.
- Explain how model attributes and tokens affect AI energy use.
- Recognize how AI use affects energy, carbon, and water.
AI’s Environmental Footprint
The potential of artificial intelligence (AI) to transform our world is profound, but it also presents critical challenges. As AI scales, so has its environmental footprint, with increasing demand for energy, carbon, and water across the systems that power AI. Let’s take a look at how some of this demand translates into resource use.
Energy
Data centers house the servers that supply the computing power to train and deploy AI models. The servers need energy to run and so do the cooling systems that keep them from overheating. As AI use grows, data centers deploy more high-performance servers, and power density increases.
From 2024 to 2030, global data center electricity consumption is projected to grow by around 15% per year, more than four times faster than electricity consumption across other sectors.
Carbon
Every AI task runs on energy, and that energy carries a carbon cost—the more tokens a task requires, the more computing resources it consumes and the more emissions it generates. This adds up quickly at scale: Data centers draw massive amounts of electricity, and because many still rely on fossil-fuel-powered grids, that demand translates directly into rising carbon emissions.
Water
Data centers use water in cooling systems to prevent servers and other computing equipment from overheating. As data center demand increases, particularly in water-stressed regions, water use is becoming an increasingly important concern for local ecosystems and communities.
The scale can be significant. Global AI demand is projected to account for 4.2–6.6 billion cubic meters of water withdrawal in 2027—equivalent to four to six times Denmark’s annual water withdrawal or half of the United Kingdom.
These projections show why more sustainable AI use matters as adoption continues to grow. The next step is turning that awareness into action.
Every Prompt Has a Footprint
Unchecked AI use is a growing social and regulatory risk, making sustainable AI both an ethical and business imperative. To address that risk, everyone across the AI value chain must act, from data center operators and energy providers to model providers and users.
The good news is that everyday AI-use habits can be part of that progress. Simple choices—like which model and tool you reach for, how you write a prompt, and how many times you retry—directly impact the inference phase, which is when AI processes live prompts and generates responses. Because inference happens continuously as people use AI and can account for up to 65% of a model’s total lifecycle energy use, individual actions have significant potential to reduce overall impact.
Sustainable and effective AI use can and should go hand in hand. The practices that reduce impacts on the planet also make you more productive by getting you more accurate answers, faster. This badge provides tips that can help you achieve both.
Individual habits are just one part of AI sustainability. The AI Sustainability Outlook explores a broader, system-level approach built around three pillars: smart demand, efficiency, and clean supply.
Core Factors of Sustainable AI
Every interaction with a language model depends on two core factors that affect its environmental footprint: model attributes and the tokens it processes and generates.
Model Attributes
Language models are trained on human language and generate responses by predicting text. Not all models are the same size—and size directly affects environmental impact. Larger models require more computing resources to process and generate responses, consuming significantly more energy than smaller models for the same task.
Model function matters too. Reasoning models “think” by using multistep chains before replying, while non-reasoning models generate answers through pattern recognition. Because reasoning adds extra computing steps, these models use more energy than a non-reasoning model.
Tokens
Tokens are the basic units of AI processing and generation. Everything you send to an AI tool and everything it sends back is measured in tokens. Processing those tokens requires compute, energy, water, and money.
As a rough estimate for English text, one token is about four characters, or about three-quarters of a word. Actual token counts vary by model, encoding, and language.
The number of tokens a task requires depends on factors such as response length, output type (for instance, media versus text), and model size. Not all tokens have the same environmental impact.
Token Type |
Relative Energy Use |
Description |
|---|---|---|
Reasoning |
Highest |
Tokens that a reasoning model uses internally before producing its visible answer. Reasoning tokens count toward output usage. |
Output |
Highest |
Tokens the model generates as its response. Output tokens are a major driver of energy consumption, so reducing unnecessary output supports sustainable AI use. |
Input |
Lower |
Tokens the model uses to process what you send, including text, files, and media. |
Cached input |
Lowest |
Previously processed input tokens that the model can reuse. |
Summary
In this unit, you explored why everyday AI use has an environmental footprint and how factors such as model attributes and token use affect the resources AI requires. These factors provide a foundation for making more intentional choices about how you use AI.
In the next unit, you build practical habits for making those choices more sustainable and effective.