Carbon Footprint of AI Tools

Artificial intelligence has become a popular and innovative tool which has helped to improve and transform industries across the globe, with millions of individuals benefiting from the solutions it offers. This popularity has spanned into a plethora of AI tools for many kinds of generation, such as text, image, video and more, all of which require energy-intensive processes to produce and maintain their increasingly high output.

From this, concerns have been raised on the sustainability of these tools and the potential outcomes as they continue to grow in utilization and consumption. There has been a focus to bring down energy costs and carbon emissions, to ensure that these tools can continue to develop whilst balancing with environmental responsibility.

Here at KnownHost, a study was conducted to uncover the environmental impact of these tools that are growing in usage, looking at the Carbon Rating of 50 AI tools and comparing how much CO₂ is produced per generator.

AI Tools with the Worst Carbon Footprint

Rytr is the AI Tool with a high output of carbon dioxide, producing 10.1g per page view on their site. During a month, the site will receive 104,700 users on average which means 1,057 kilos of CO₂ will be produced from this tool alone.

Spellbook is second, producing 6.5g of CO₂ per page view, with an average monthly traffic of 3,100, but still producing 20 kilos of carbon dioxide a month overall. Wordtune produces 4.4g of CO₂ per page view, totalling just over 1,200 kilograms a month due to over 250,000 users on average.

In fourth position, Synthesia produces 3.3g of CO₂ per page view, with a large average of monthly traffic at 468k, totalling over 1,530 kilograms of CO₂ a month. Lastly, Deep Dream Generator produces 2.7 grams of CO₂ per page view and produces 750 kilograms a month on average from a traffic volume of 276k.

From the study, it was also found that the popular tool, ChatGPT, with over 164 million users a month, has the highest production of CO₂, yet it has a lower production per page view of 1.59g than the top five we’ve discussed. It produces an average over 260,930 kilograms of carbon dioxide per month.

AI Tools with the Best Carbon Footprint

At the top of the list is MusicGen, which produced 0.07g of CO₂ per page view. It also has a low monthly traffic with the estimated footfall of viewers producing 335 grams of carbon dioxide a month.

Grok is second with 0.17g of CO₂ per page view, however it has a higher amount of traffic with an average of 190,100 users a month. Fliki has a slight increase in both areas 0.19g of CO₂ produced per page view and a total traffic of 193,800 a month.

Artbreeder has another jump to producing 0.23g of CO₂ per page view, but with a low traffic of 27,200 viewers per month which produced just over 6 kilos of carbon overall. Appypie Design has the largest traffic at 661,500 therefore produces 175 kilograms of carbon dioxide per month on average, however it still produces a small amount of CO₂ at 0.26g per page view, which is the fifth lowest across all tested AI tools.

Conclusion

For the future of AI innovation, it is clear to recognise the importance of advocating for energy-efficient practices and gain insight into the changes that can be made so that the contribution to harmful emissions can be reduced. This approach can then lead to a greener future, aligning with global sustainability along with fostering a culture of technological advancement.

Methodology

For this study, a seed list of 50 Generative AI Tools was drawn up and their URLs tested by https://www.websitecarbon.com/, an online calculator that rates websites on a scale from A+ to F. These URLs were then crawled by Screaming Frog SEO Spider to find out the amount of carbon produced per page view, which was then multiplied by estimated site traffic, according to Ahrefs, to establish how much CO₂ is produced.