Generative AI is transforming marketing, but some of the cost and impact of this technology is just coming clear. Intensive GenAI tools can use more energy and processing power than we are all led to believe, so it’s vital for businesses to understand the impact of this technology and how to adopt effective practices for a more sustainable use of GenAI.
Why the Carbon Footprint of Generative AI Matters
Since the public release of ChatGPT in late 2022, Generative AI has revolutionised marketing at every stage, from creating text and images to gathering strategic insights from vast amounts of data.
All digital activities have a measurable environmental impact – every loaded page, search query sent, or video streamed consumes energy. But GenAI takes this to a new level – a single ChatGPT query consumes nearly 10x the energy of a Google search, the impact is being compounded as more internet users adopt this new technology into their daily activities.
Many companies have made commitments in an Environmental, Social and Governance (ESG) policy, and in order to achieve their stated targets they will need to begin accounting for the impact of AI, which starts with determining the additional carbon impact of AI by calculating their energy consumption.
Quantifying the Environmental Impact of GenAI
There is a notable lack of transparency from large-language model (LLM) providers as to the full scale of their operations, but there are various calculations we can make. In 2023, data centres 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 centre power requirements by 2030, pushing their share of global electricity consumption to 3-4%.
The United States is at the forefront of this industry, and has the largest consumption figures: US data centres consumed 3% of the nation’s power in 2022. By 2030, this figure is expected to rise to 8%, requiring approximately $50 billion in new electricity generation capacity. These trends represent the most significant increase in electricity demand in a generation.
How GenAI Marketing Impacts Emissions
In a recent study by The Brandtech Group, calculated emissions train an LLM model at between 250 and 2000 tons of CO2e, depending on the number and complexity of the prompts. But the inference analysis that is required for more nuanced marketing prompts generates significantly more AI carbon emissions – a common model, such as LLaMA 3, will generate 90% of its emissions during inference.
An example of a marketing use case could be AI-generated product pages for online retailers. A major retailer might list around 300,000 new items every year, published in dozens of different languages. We estimate that GenAI creation of product pages on this scale could result in over 50 tCO2e, which is the equivalent to 30 round trips from Paris to New York on a commercial airline flight.
Apart from the environmental cost of this data-centre activity, we can be sure that the prices charged by the LLMs will increase in the coming years. At present, costs of using GenAI are low as the AI providers fight for market share, but are operating at a loss. In the near future prices will inevitably increase, which will have a direct impact on the financial feasibility of AI-driven workflows.
Sustainable AI Practices for Marketers
For companies that have a commitment to reduce energy consumption, it’s key to identify mitigation strategies when using GenAI. As teams and users become more knowledgeable on the inputs and processes, this in turn will lead to a more sustainable use of GenAI in marketing.
Among these best practices, marketers should look to select relevant use cases only for AI deployment. Before deploying AI tools and solutions, marketers should assess the value of the project and the value of using generative AI models over less-energy-hungry technologies.
For any chosen model, careful thought should be put into reducing the number of parameters, as using smaller models where possible, such as Google Gemma 2 or GPT-4o mini, will use fewer vital resources. With effective training, marketers can use more informative prompts to avoid sending multiple queries. Educating teams will help to avoid unnecessary iterations in obtaining the desired result.
The effective use of carbon footprinting tools can help marketers see the full picture, leading to more efficient deployment of GenAI to maximise results, driving measurable business impact and minimising energy consumption.
A simple online search will highlight a number of free tools to help marketers determine the carbon footprint of traditional (i.e. non-AI) marketing campaigns, websites, and measurement tools. There are also a number of global marketing organisations working in collaboration with brands and businesses to effectively implement environmentally friendly GenAI marketing solutions.
