Since the public release of ChatGPT in late 2022, Generative AI has been revolutionizing marketing at every stage, from creating a new campaign to organizing follow-ups and gathering strategic insights from vast amounts of data. And while the cost of using GenAI is still relatively low – a fact that might change in the near future, as most AI providers are reportedly operating at a loss as of yet, there is another cost to this technology: its impact on the environment.
Less discussed than the effects of transportation, food, or fashion on our environment, the impact of digital technology should not be overlooked, especially considering its rapid growth. Since every loaded page, search query sent, or video streamed consumes energy, digital activities have a measurable environmental impact. And considering that a single ChatGPT query consumes nearly 10x the energy of a Google search, this impact is expanding fast and shows no sign of slowing down as more and more internet users adopt this new technology into their daily activities.
“A single ChatGPT query consumes nearly 10x the energy of a standard Google Search.”
From "A deep dive into the environmental cost of Gen AI" by The Brandtech Group, fifty-five, and Scope3
At fifty-five, we firmly believe that you cannot manage what you don't measure. To help organizations achieve their ESG goals, our experts have already provided free tools to determine the carbon footprint of traditional (i.e., non-AI) marketing campaigns, websites, and measurement tools. And now, as marketing teams integrate generative AI at a rapid pace, we have teamed up with The Brandtech Group and Scope3 to determine the actual environmental impact of GenAI by calculating its energy consumption.
As Rebecca Sykes, Partner at The Brandtech Group, writes in the foreword of this study: "The conversation around AI and sustainability has, for too long, been built on assumptions rather than data. While many of us in the industry acknowledge the theoretical environmental impact of AI, there's been little in the way of concrete measurement or actionable insights. That's why we embarked on this project – to build a foundation of knowledge that will allow us, and the broader marketing industry, to take meaningful steps forward."
Indeed, as of 2025, we still know very little about the carbon footprint of generative AI due to a concerning lack of transparency from LLM providers as to the environmental impact of this technology. Faced with this lack of data, our combined teams of experts decided to conduct their own research to determine how much energy generative AI consumes, thereby promoting better data-driven accountability.
Their primary focus: the data centers that power AI models, from training to actually running LLMs.
“In 2023, data centers accounted for 1-2% of global electricity use, equivalent to powering 17 million homes. But this demand is rising rapidly. Goldman Sachs Research projects a 160% increase in data center power requirements by 2030, pushing their share of global electricity consumption to 3-4%. The United States illustrates this trend. US data centers consumed 3% of the nation's power in 2022. By 2030, this figure is expected to rise to 8%, necessitating approximately $50 billion in new electricity generation capacity. These trends represent the most significant increase in electricity demand in a generation.”
From "A deep dive into the environmental cost of Gen AI" by The Brandtech Group, fifty-five, and Scope3
Our combined teams of experts have developed a methodology to calculate the carbon footprint of generative AI, focusing on the two main components of a large language model's lifetime: training and inference. Note that the next iterations of this study will also include water consumption for cooling purposes in data centers as well as non-renewable materials used in hardware.
Training refers to the stage where a generative AI model is trained to produce content according to instructions. During training, AI models are "fed" large datasets to study patterns and develop their capabilities.
Inference is the next stage, when a trained model actually generates content according to input.
According to our study, while training models requires considerable energy (between 250 and 2000 tons of CO2e are emitted to train an LLM depending on its size), inference is where most AI carbon emissions come from. A common model, such as LLaMA 3, will generate 90% of its emissions during inference.
To bring these statistics into perspective, our study highlights the environmental impact of several typical AI marketing use cases, such as AI-generated product pages for online retailers. For major retailers, around 300,000 new items could be released every year, published in dozens of different languages. Automating the creation of their product pages with GenAI could result in over 50 tCO2e, equivalent to 30 round trips from Paris to New York.
“Automating 300,000 product pages with AI = 50 tCO2e, or the equivalent of 30 round trips from Paris to New York”
From "A deep dive into the environmental cost of Gen AI" by The Brandtech Group, fifty-five, and Scope3
Our research and combined expertise allowed us to identify several mitigation strategies to reduce energy consumption when using GenAI. Among these best practices:
Before deploying AI tools and solutions, marketers should assess the value of the considered use case and the worth of using generative AI models over less-energy-hungry technologies.
The more parameters a model has, the more computational resources it requires and the more energy it consumes. Choosing a "smaller" model, such as Google Gemma 2 or GPT-4o mini, whenever possible, will use fewer resources.
By training your teams to use prompts effectively, you will help them avoid sending multiple queries – with slight modifications each time – to obtain the desired result.
To help you assess your GenAI-related carbon emissions, we developed an open-source calculator, available here.
For us, the transformative potential of generative AI is undeniable, having helped our clients deploy this technology strategically and securely through many successful projects over the past few years. But the environmental impact of AI should not be ignored, says Scope3 CEO Brian O'Kelley. "When marketers can see the full picture – both performance and environmental impact – they can deploy AI strategically to maximize results while minimizing waste. This is how we turn AI's enormous potential into responsible, sustainable growth."
To explore the methodology we relied on, detailed use cases, and real-world examples, download the full 40+ page whitepaper here. And if you would like to know how fifty-five can assist you with your ESG strategy and AI implementation, don't hesitate to contact us.
Discover all the latest news, articles, webinar replays and fifty-five events in our monthly newsletter, Tea O'Clock.
.png)
.svg)
