As more people begin to use and experiment with generative AI, there is a growing awareness of its impact on the environment and sustainability. Every aspect of the AI lifecycle consumes natural resources, including energy, water and minerals, and as a result, releases greenhouse gases. There is an increasing emphasis on the need to use it responsibly, and we think that having an awareness of the environmental impacts can help with this. We last published a blog around AI’s environmental impact in 2022, but it’s fair to say that things have changed since then. Generative AI is becoming ever more integrated into tertiary education, and can be found in administration, professional services, research, and teaching and learning. Although we strongly believe that AI can aid education, we would also urge people to be aware of and consider the impact this influx is having on the environment.
Of course, it is important to note that different types of artificial intelligence vary in their environmental footprint. The AI that powers text-to-speech consumes significantly less energy than large language models, and there is also a difference in the energy consumption of generative AI depending on the tasks it is used for. For example, the power required for image generation is considerably higher than that needed for question answering. This blog will look to provide updated information and explore the ongoing issues, and although we won’t be able to cover every aspect, we hope it will help you to make informed decisions about your own and your institutional use of AI.
A positive environmental impact
We’ll begin by looking at how AI can have a positive impact on the environment. It has been argued that AI will, in the long term, help to hit climate targets. Bill Gates noted that the extra demand that AI is having on data centres is likely to be matched by new investments in green electricity. He said that tech companies are willing to pay extra to use clean electricity sources, so that they can say they’re using green energy. Amazon is currently the largest corporate purchaser of renewable energy in the world, while Microsoft agreed to support the development of 10.5 gigawatts of new renewable energy capacity globally, and Google aims to run data centres entirely on carbon-free energy by 2030.
Along with this, there are several ongoing initiatives that are using AI to tackle climate change. These include measuring changes in icebergs 10,000 times faster than a human could, mapping deforestation, as well as analysing data to help companies track, trace and reduce their emissions by 20-30%. More can be read about these and further examples, on the World Economic Forum website. You may have also seen at COP28, a high-level event on Artificial Intelligence for Climate Action took place to raise awareness and share knowledge of the application of AI to help accelerate the implementation of the Paris Agreement. A recording of this event can be viewed on YouTube.
Interestingly, a recent study looked into the carbon emissions of writing and illustrating, and concluded that overall, generative AI produces less carbon emissions compared to a human performing the same task. They focused specifically on ChatGPT, BLOOM, DALL-E 2, and Midjourney, examining the hardware and energy used to provide the AI services, as well as the total annual carbon footprint of humans. More detail can be found in the paper, but the overall conclusion was that while generative AI currently has a lower carbon footprint than humans for writing and illustrating, its growing efficiency and demand could lead to increases in environmental costs (an example of the Jevons Paradox). This therefore suggests that the cooperation between humans and AI might be the most sustainable path forward.
It’s also important to note that the complexity and size of AI models directly correlate to energy consumption. Many people tend to view all AI negatively because they associate it with generative AI, LLMs and data centres, however, the AI that powers text-to-speech technology consumes significantly less energy. Therefore, there is a great deal of AI that is contributing positively and is not as harmful to the environment as people may initially assume.
A negative environmental impact
We’ll now look at how AI can have a negative impact on the environment. It is fair to say that as generative AI tools have become more popular, the increase in computing power, electricity and water consumption have contributed to higher carbon emissions. However, it can be difficult to get accurate or complete data on environmental impacts as precise numbers are closely guarded by tech companies and are often under-documented. We’ll highlight a few of the key points from Microsoft, Google and Amazon later on. However, for now we’ll focus on the main issues surrounding energy and water usage, which are the biggest contributors to the environmental impact.
Energy
There have been several studies and articles published around generative AI’s energy usage, with likely more to come as the available data improves. Although it can be difficult to find out the usage of specific models like OpenAI’s GPT-4, researchers have been able to estimate with others. In this particular study, researchers calculated the carbon emissions caused by using an AI model for different tasks. They found that generating images with a powerful AI model consumed a significant amount of energy, with the least efficient model using around half a smartphone charge per image generation. In contrast, when used to generate text, it was much less energy-intensive, using only 16% of a full smartphone charge. From their findings, it may come as no surprise that image generation was the most energy and carbon-intensive task, with 1000 images producing roughly as much carbon dioxide as driving 4.1 miles in ‘an average gasoline-powered car’.
Research has found that AI is currently estimated to account for less than one-fifth of the overall energy demand of data centres, but this share is likely to significantly increase over the next few years. The International Energy Agency (IEA) expects that global electricity consumption from data centres, artificial intelligence and the cryptocurrency sector could double between 2022 and 2026, which is roughly equivalent to the electricity consumption of Japan. In addition, a graphical processing unit (GPU) is a type of computing hardware that is commonly used to train and deploy AI models. An individual AI model may require multiple GPUs and have been found to use four times as much energy as those serving conventional cloud applications. Despite the large purchasing of renewable energy by large tech companies, the growing energy demand for AI is significantly outpacing the increase in renewable energy. This could therefore suggest that we will continue to see higher greenhouse gas emissions until this levels out.
Water
Generative AI systems need a significant amount of fresh water to cool processors and generate electricity, with sustainability reports over the last couple of years highlighting an increase in water use. An average hyperscale data centre, such as those used for training AI models, uses around 550,000 gallons (2.1 million litres) of water daily. This is especially concerning considering that these data centres are often located in areas with limited water supply. You may have read about the group of Microsoft data centres in West Des Moines, Iowa that required 11.5 million gallons (i.e. 6% of all the water in the whole district) in the final months of training GPT-4. There were also similar concerning figures in Arizona and in Oregon.
Finally, a study reported that the global AI demand may be accountable for 4.2 – 6.6 billion cubic meters of water withdrawal in 2027, which is more than the total annual water withdrawal of half of the United Kingdom. It’s clear from the current data and research that a lot needs to be done to tackle the large consumption. We’ll look now at some figures from big tech’s sustainability reports.
What are the companies reporting?
We thought it would be useful to provide an overview of the key points related to generative AI that have recently been shared in the sustainability reports from Microsoft, Google, and Amazon. The overall trend is rather negative, with Microsoft and Google both reporting increased emissions and delays in climate targets. Data centres seem to be the primary factor in the increase, as they are essential for training and operating the models behind generative AI such as Google’s Gemini and OpenAI’s GPT-4. There has been criticism surrounding the reports, with Amazon being accused of greenwashing, Microsoft promoting AI for fossil fuel extraction and Google failing to disclose how much AI training and inference have contributed to its energy costs.
Microsoft
In their 2024 sustainability report, Microsoft said that their Scope 3 emissions increased by 30.9% and emissions across all scopes (1 – 3) were up 29.1% from the 2020 baseline. Their rise in Scope 3 emissions primarily comes from the construction of more data centres and associated embodied carbon in building materials, hardware components, servers and racks. Along with the rise in emissions, Microsoft also reported that their water consumption has increased. Overall, they’ve admitted to missing their climate goal by a large margin in 2023, with Microsoft’s vice chair and president, Brad Smith admitting that because of their AI strategy, their 2030 net zero goals may not be met.
Similarly to Microsoft, Google also reported an increase in their emissions. Their total greenhouse gas emissions represented a 13% year-over-year increase and a 48% increase compared to their 2019 target base year for their goal of reaching net zero. They concluded that this was primarily due to increases in data centre energy consumption and supply chain emissions. As more AI is being integrated into their products and with an expected increase in technical infrastructure investment, the reduction of emissions may prove to be a challenge. Google also revealed that the expansion of AI contributed to an increase in data centre workloads and therefore, the associated water footprint required to cool them efficiently. This led to a 17% increase in its water footprint in 2023 compared to the previous year, which mirrored the growth in electricity use.
Amazon
Amazon’s sustainability report highlighted that they met their goal of becoming 100% renewably powered seven years ahead of schedule, meaning all the electricity the company uses to power its operations was balanced out by investments in 513 renewable energy projects worldwide. However, they noted the challenges they will face moving forward, given their continued growth and development in new technologies such as AI. The construction, operation and decommissioning of Amazon’s building portfolio, which includes data centres, accounted for one fifth of Amazon’s total carbon emissions in 2023.
The report also stated that water is an essential resource for AWS, with it being primarily used to cool their global data centres. Although they didn’t disclose their water footprint, they did say that at the end of 2023, AWS was 41% of the way toward achieving its water positive goal. They also noted figures including 0.18 Litres of water per kilowatt-hour (L/kWh) water use effectiveness (WUE) for AWS data centres, which was a 5% improvement from 2022.
What are they doing about the increases?
All three sustainability reports mention ways that they are tackling the ongoing climate challenges. These can be read about in more detail in the respective reports, but to give a few examples, Microsoft are collaborating with the UN on the Early Warning for All Initiative and their new data centres are being designed and optimised to support AI workloads and will consume zero water for cooling. Google mention that they’re using their water risk framework to identify climate conscious cooling solutions that consider carbon-free energy availability, watershed health, and future water needs. Meanwhile, Amazon are scaling renewable energy and lower-carbon approaches to heating and cooling. These include rooftop solar installations and transitioning to hydrotreated vegetable oil to power backup generators at AWS data centres in the US and Europe.
How can you be more responsible with your AI use?
Hopefully from the information you’ve read so far, you’ll be starting to consider your own use of AI. As we have stated before, we see AI as a valuable tool in education and do believe in the benefits that it can bring to students, teachers and staff alike. Nevertheless, we all share a responsibility to reduce our carbon footprints and remain conscious of the climate emergency.
When you come to do a task, we’d recommend you consider whether generative AI is the right tool for the job. Could your query or prompt be a simple Google search instead? Better yet, could it be an Ecosia search? Ecosia is a search engine that uses its profits to fund tree-planting projects in 35+ countries. It also runs on renewable energy and supports local communities and biodiversity. Choosing between a search engine and generative AI is particularly relevant in the context of image generation, where you can easily check whether the image already exists. It is estimated that using generative AI tools can use 60 times more energy than a search engine. Consider implementing this into your AI literacy courses and sessions at your institution and include both staff and students in the process.
Another idea is to look at examples that have already been tried and tested. There are a lot of great use cases available now that detail how you can successfully utilise generative AI in your work, with some even including useful prompts. You could begin with our Generative AI in Practice page, or take a look at AI for Education’s Prompt Library.
In summary
We hope that this blog has provided you with some guidance around the environmental issues related to AI tools and encouraged you to reflect on and assess your own usage. The more awareness and understanding that we all have as individuals will help us make more environmentally conscious decisions regarding AI. Although AI is contributing to environmental issues, it is the human use that is driving this. We have the option to choose what we use. Perhaps you don’t need to use the biggest AI models to solve every problem. Hopefully, one day it will be possible to know exact numbers, allowing users to be more aware of the amount of energy and water they are using. Until then, it is our responsibility to adopt a mindful approach.
Are you currently promoting sustainable and responsible use of AI in your institutions? We’d love to hear from you. Feel free to email us at AI@jisc.ac.uk.
If you work in the FE and Skills sector, you may be interested in joining Jisc’s FE and Skills Digital Sustainability Community. If so, please fill out the expression of interest form via this link and we will be happy to email you the invite to join the Teams channel. This online community space will be dedicated to fostering a collaborative environment where members can share ideas, resources, and best practices related to digital sustainability.
Find out more by visiting our Artificial Intelligence page to view publications and resources, join us for events and discover what AI has to offer through our range of interactive online demos.
For regular updates from the team sign up to our mailing list.
Get in touch with the team directly at AI@jisc.ac.uk