Jump to content

Environmental and climate-related impacts of AI-searching

From UBC Wiki
According to Reuters Legal in August 2025 Perplexity AI failed to convince a judge to dismiss a lawsuit over its alleged "misuse of articles" to train its AI...

Compiled by

Updated

See also

Introduction

"...Artificial intelligence (AI) systems are thirsty, consuming as much as 500 milliliters of water – a single-serving water bottle – for each short conversation a user has with ...OpenAI’s ChatGPT....the same amount of water to draft a 100-word email message..." – Lo, 2025

Environmental and climate implications of LLMs

Large language models (LLMs) such as ChatGPT by OpenAI have a measurable environmental and climate-related impact, arising from the electricity demands of large-scale data centres and high-performance computing infrastructure. AI systems require considerable computational power / substantial energy both during model training (a one-time but intensive process) and during inference (the ongoing generation of responses to user queries). As AI-powered search tools become embedded in research workflows, cumulative energy use becomes a central concern.

Energy consumption in model training

Training AI models can require thousands of specialized GPUs operating continuously for weeks or months. Depending on the energy mix of the hosting data centre, this process can result in significant greenhouse gas emissions. While widely cited comparisons such as equating model training emissions to long-haul flights vary substantially by model architecture and electricity source, the broader point is that carbon intensity depends heavily on whether energy is derived from fossil fuels or renewable sources.

The rapid growth of AI-driven search technologies also intensifies demand for specialized hardware, including GPUs and data storage systems. This demand carries environmental consequences related to rare earth mineral extraction, manufacturing emissions, and electronic waste disposal. From a lifecycle perspective, AI infrastructure extends beyond electricity consumption to include material resource depletion and waste streams.

Institutional and strategic concerns for libraries

For librarians and research institutions, the environmental footprint of AI-powered searching raises institutional and strategic questions. The scalability of these tools means that even modest per-query energy costs can become significant at scale. However, mitigation strategies are emerging. Advances in model efficiency, smaller domain-specific models, optimized inference techniques, and improvements in data centre energy sourcing can reduce environmental impacts. Some major technology providers are investing in renewable energy procurement and carbon accounting, although global reliance on non-renewable energy remains substantial.

As AI search tools proliferate in scholarly communication and knowledge synthesis, their environmental implications should form part of ethical and institutional decision-making. Sustainable AI infrastructure, transparent reporting of energy use, and responsible procurement policies will be essential if libraries and universities are to align technological innovation with climate commitments.

MIT 2025 Report on AI Energy Footprint

  • AI energy use is growing rapidly: Training and running large AI models requires vast amounts of electricity, and demand is accelerating as AI systems become larger and more widely deployed.
  • ChatGPT is now estimated to be the fifth-most visited website in the world, just after Instagram and ahead of X.
  • Data centers are needed to power AI. In fact, hyperscale data centers powering AI rely heavily on energy-intensive hardware (especially GPUs), significantly increasing electricity and water consumption.
  • Big tech companies often lack transparency about the true carbon footprint of AI, making it difficult to assess real environmental costs.
  • Efficiency gains may not offset growth: While hardware and software efficiencies are improving, these gains are being outpaced by the explosive growth in AI usage (a rebound effect).
  • Policy and accountability lag behind innovation: Governments and regulators have not yet caught up with AI’s environmental implications, leaving sustainability largely to voluntary corporate commitments.

Librarian point of view: This MIT report is a credible and timely overview of AI’s growing energy demands and climate impact, noting that it translates complex technical data into accessible insights for the public, students, and policymakers. The report relies on estimates and corporate disclosures rather than primary datasets, so it should be supplemented with peer-reviewed studies or lifecycle analyses for rigorous research. The report is valuable for understanding trends, highlighting transparency gaps in big tech, and supporting discussions about sustainability, ethics, and policy, making it a useful addition to interdisciplinary collections and research guides.

AI Carbon Footprint

The training of AI models often requires substantial computational power, using significant energy for processing. The deployment and operation of AI systems also contribute to their carbon footprint. AI applications that run on cloud infrastructure or data centres require a lot of power to operate, resulting in considerable energy consumption and carbon emissions. The computations required for deep learning research have increased by 300,000 fold from 2012 to 2018. Manufacturing and disposal of hardware components, AI-specific chips or servers, contribute to carbon footprint. A recent study found that training an off-the-shelf AI language-processing system produced around 600 Kg of emissions, about the amount produced by flying one person roundtrip between New York and San Francisco.

There are initiatives to reduce the AI carbon footprint and improve its sustainability. Efforts include developing more energy-efficient algorithms and optimizing computing infrastructure to minimize energy consumption during training and operation. CodeCarbon is a software package that can be integrated into a Python codebase and estimates the amount of CO2 produced by the cloud or personal computing resources used to execute the code. Furthermore, using renewable energy sources to power data centers and adopting energy-efficient hardware designs can also help reduce the environmental impact of AI.

AI queries' CO2 footprint

While ChatGPT generates more carbon dioxide than many of its competitors, Elon Musk's Grok AI is the most environmentally friendly, according to research published in 2025. Source: https://www.cyberdaily.au/security/11942-chatgpt-produces-an-estimated-4-32-grams-of-co-for-every-query

Here are the differences in AI expressed in grams of CO₂:

ChatGPT produces an estimated 4.32 grams of CO₂ for every query. Perplexity is not far behind with 4 grams of CO₂. Grok AI comes out on top thanks to an architecture that is designed around lower power consumption. Each Grok AI query is roughly on par with a single Google search.

Recent research estimates that each ChatGPT query produces about 4.32 grams of carbon dioxide, highlighting the environmental cost of using large AI models. This figure—based on assumptions about data center energy use and electricity sources—suggests ChatGPT has a larger carbon footprint per query than many competitor models, largely due to its computational intensity. While 4.32 grams may seem small in isolation, the millions or billions of daily queries can add up to significant emissions. The analysis underscores the broader challenge of balancing AI innovation with sustainability and improving energy efficiency in data centers.

Librarian's point of view: if a single Google search requires 0.17 grams of C02, and ChatGPT-4 is 35-40x higher - imagine the power used by tools like Elicit.com or Undermind.ai? RAG tools are particularly computationally-intensive, leading to an environmental impact 25 times (or more) than a single Grok query. One ChatGPT query generates the equivalent amount of CO₂ as sending 21 emails or fully charging one smartphone.

Libraries in the climate change era

Libraries are increasingly vital in addressing climate change issues by developing sustainable collections and engaging communities in discussion. Librarians face direct climate impacts on operations, necessitating sustainable practices like energy-efficient systems and expanded digital resources. Libraries can curate materials promoting environmental stewardship, climate literacy, and local resilience, acting as hubs for climate education and community initiatives. Challenges include limited funding, technological barriers, and the need for greater community involvement, especially in vulnerable regions. Integrating Indigenous knowledge and aligning with the United Nations Sustainable Development Goals (SDGs) can strengthen libraries’ role in fostering climate resilience and empowering communities.

Presentation

Video asks: As tech companies increase their AI production, the environmental costs are coming to light. What are the resources fuelling the AI revolution? And how does AI impact the tech industry’s climate goals? Will AI drive or solve the climate crisis?

References

Disclaimer

  • Note: Please use your critical reading skills while reading entries. No warranties, implied or actual, are granted for any health or medical search or AI information obtained while using these pages. Check with your librarian for more contextual, accurate information.
Retrieved from "https://wiki.ubc.ca/index.php?title=Environmental_and_climate-related_impacts_of_AI-searching&oldid=896389"