AI’s Growing Energy Footprint: Individual Actions Can Make a Difference
In the bustling landscape of modern technology, artificial intelligence (AI) tools have become ubiquitous, from crafting social media imagery and offering relationship advice to analyzing medical scans and detecting financial fraud. This rapid proliferation, however, comes at a significant environmental cost, with global data centers now consuming an alarming 414 terawatt-hours per year, representing approximately 1.5% of the world’s electricity usage. This figure, according to the International Energy Agency, surged by 12% annually for five years before a steep 17% jump in 2025. Projections indicate a doubling of electricity demand from data centers by 2030, a demand largely met by fossil fuels, raising concerns about both carbon emissions and the depletion of local water resources in drought-stricken areas used for cooling these energy-intensive facilities.
While the individual environmental footprint of generating AI-powered text or imagery may seem minimal, experts emphasize that the primary responsibility for reducing AI’s resource consumption lies with tech companies. This includes developing smarter, energy-efficient algorithms and more robust hardware. Nevertheless, individual users are not without agency, with simple yet impactful actions that can collectively mitigate the environmental toll of AI usage.
“Individual choices are not meaningless, and some are more powerful than people realize,” states computer scientist Ivana Drobnjak of University College London.
The Energy-Intensive Nature of AI Bots
Quantifying the energy expenditure of a single chatbot query is notoriously complex. Google estimates that its Gemini chatbot consumes approximately 0.24 watt-hours to respond to a median-length text query, equivalent to less than nine seconds of television viewing. This also entails about 0.26 milliliters of water and 0.03 grams of carbon dioxide emissions. While seemingly small, these individual expenditures accumulate rapidly for frequent AI users and businesses.
The substantial energy consumption of AI models stems partly from their reliance on powerful processors like Graphics Processing Units (GPUs), which demand significantly more energy than the Central Processing Units (CPUs) used for simpler tasks. Furthermore, the underlying architecture of most popular generative AI tools, particularly Large Language Models (LLMs) that power chatbots, contributes significantly. These LLMs are built on a “transformer architecture” that enables them to train on vast linguistic patterns and compute billions or trillions of parameters. Each new word generated in response to a user query requires the model to re-run the query and previously generated words, performing billions of calculations every time.
Latest Videos FromLive Science
Despite this, tech companies are making strides in energy efficiency. Google’s 2025 calculations show Gemini’s energy consumption for a median-length text prompt to be 0.24 watt-hours, a 33-fold decrease from the previous year. However, the sheer scale of AI usage amplifies even small amounts of energy. Ivana Drobnjak, referencing 2025 OpenAI figures, estimated that around 3.2 billion queries were sent to ChatGPT daily in May. Users are generating vast quantities of text, images, and video, engaging in lengthy chatbot conversations, and increasingly creating “AI agents” that autonomously send queries to other AI chatbots. Addressing this burgeoning energy demand, experts offer practical advice for users.
Get the world’s most fascinating discoveries delivered straight to your inbox.
Practical Steps for Greener AI Usage:
1. Reconsider AI Necessity: Users should first assess whether AI is truly required for a given task. As computer scientist Günter Klambauer puts it, “Asking ChatGPT ‘What should I wear today?’ or ‘How is the weather?’ is like taking a Concorde to travel to your supermarket.” For web searches, opting for “Web results only” or including “-ai” in queries can bypass AI-generated summaries and conserve energy. “If you’re just looking for a particular article, turning that off could be powerful from a saving-energy perspective,” advises Udit Gupta, an expert in electrical and computer engineering at Cornell Tech.
2. Embrace Smaller, Specialized Models:
For specific tasks like translation or summarization, individuals and businesses can significantly reduce energy consumption by utilizing smaller, specialized language models. These models, trained more narrowly, perform fewer computations than massive, general-purpose LLMs. A 2025 UNESCO study by Drobnjak demonstrated that smaller models, such as “opus-mt-en-es” for English-Spanish translations, consumed 15 to 50 times less energy while delivering comparable or even superior results. These models, though often less user-friendly, are freely available on platforms like Hugging Face. The overall shift to smaller models in the study resulted in a remarkable 90% decrease in energy use, highlighting its potency as an energy-saving strategy. “You don’t need to use a trillion-parameter model for editing an email,” Gupta aptly states.
Using small, specialized models for particular tasks consumes a fraction of the energy guzzled by large, all-purpose models, with similar if even slightly better accuracy.
(Image credit: Knowable Magazine)
3. Encourage Conciseness from Chatbots:
Given that LLMs perform numerous computations for each successive word they generate, opting for models that produce less text is beneficial. Mosharaf Chowdhury of the University of Michigan, who measures LLM electricity usage, notes that “chattier” models tend to consume more energy. For instance, a “problem solving with reasoning mode” version of Alibaba Cloud’s Qwen model consumed significantly more energy due to producing roughly ten times more words than its “text conversation” mode. Experts recommend using reasoning mode only for complex questions and otherwise adhering to a chatbot’s standard mode.
Simply instructing AI chatbots to “be brief” or setting a word limit can also save energy. Drobnjak and her colleagues found that instructing the Llama model to halve its output reduced energy consumption by 50%. In contrast, shortening the prompt itself offered minimal savings (around 5% for a halved prompt length). “The size of the output is what determines and drives the energy expenditure the most,” Drobnjak emphasizes. She has collaborated with the city of San Francisco to develop energy-saving tips for AI users, advocating for specificity and instructions like “five bullets max.”
Keeping chatbot prompts short can conserve some energy, but asking chatbots to keep their responses brief amounts to much bigger savings.
(Image credit: Knowable Magazine)
4. Optimize Image and Video Generation:
Generating images and videos, which involves iterating millions of pixels, consumes significantly more energy than text generation. With nearly 40% of teens aged 13-17 using AI for image and video creation or editing, according to a Pew Research Center study, optimization is crucial. Drobnjak advises generating images or videos only when necessary and at the lowest possible resolution initially. “One option is to just start in low resolution,” she says, “and if the algorithm is in the right direction, you then start increasing resolution.” Editing existing images is also less computationally intensive than generating new ones. Furthermore, batching multiple image generation requests into a single session proves more efficient than individual requests.
While these individual actions may seem like a “drop in the bucket,” experts contend that their cumulative impact is significant. As tech companies, scientists, and policymakers work towards broader solutions for AI’s environmental footprint, “the individual who knows to reach for the right tool can make a real difference,” Drobnjak concludes. “AI is [consuming] so much energy that we have to look at it from every angle.”
Editor’s note: This story was updated on July 21, 2026, to clarify that the energy use of individuals who use AI tools a lot is cumulative, not necessarily huge, as was originally stated.
This article originally appeared in Knowable Magazine, a nonprofit publication dedicated to making scientific knowledge accessible to all. Sign up for Knowable Magazine’s newsletter.
