By Marina Grossi*
Recently, social media has been flooded with AI-generated drawings in the style of the renowned Studio Ghibli, the Japanese animation studio behind films such as "Spirited Away" and "Ponyo." trend The phenomenon – as the instant trends that take over the digital universe are called – was so overwhelming that it brought 1 million new users to OpenAI, owner of the ChatGPT tool. The intense data traffic overloaded the servers, forcing the company's management to limit access to the functionality. The boom in images in the style of anime This has generated ethical discussions about copyright and AI, as well as about the high consumption of natural resources, especially water and energy, demanded by data centers in generative AI.
To generate texts, images, audios, and videos very similar to those created by humans, generative AI processes a very high volume of data, cross-referencing information from extensive databases and applying models to produce answers in a short time. All this data is stored and processed in data centers, composed of servers that, to function, depend on increasingly large amounts of electricity, since cross-referencing a large volume of information in a short time consumes much more energy than a simple internet search. On average, each query to ChatGPT consumes 2.9 Wh (watt-hours) of electricity – in comparison, a regular search uses a fraction of that, about 0.3 Wh.
A study by MIT Technology Review indicated that during the peak week of AI usage, the estimated energy consumption to generate the images was 40 MWh (megawatt-hours), equivalent to the consumption of 7,000 Brazilian households for 24 hours. The expansion of AI use to a scale of millions of users makes the technology's impact on energy consumption, as well as water consumption necessary for cooling data centers, significant. An estimate from the United Nations Environment Programme (UNEP) indicates that the demand for water resulting from AI could reach 4.2 to 6.6 billion cubic meters in 2027 – this figure includes not only data center consumption but also the volumes needed for semiconductor production.
In the case of energy, a report from the International Energy Agency (IEA) reveals the global dimension of the problem: in 2024, data centers accounted for 1.5% of the world's electricity consumption, equivalent to 415 TWh (terawatt-hours), and this demand is expected to more than double by 2030, reaching 945 TWh, equivalent to 3% of global consumption. This increased energy demand is accompanied by a rise in greenhouse gas emissions related to electricity consumption, which are expected to jump from the current 180 million tons of carbon equivalent to 300 million tons in 2035, according to the IEA. The Open Compute Project estimates that the information and communication technologies (ICT) sector is on track to increase its share of global carbon emissions from 4% in 2021 to 8% by the end of this year, as global internet traffic is expected to double to 4.2 trillion gigabytes.
The environmental impacts of AI expansion have already raised alarm bells at the UN, which launched an international coalition for environmentally sustainable AI in February. This initiative involves more than 100 partners – including 37 technology companies, 11 countries, with France at the forefront, and five international organizations. The initiative, spearheaded by UNEP, envisions a collaborative approach to finding standardized methods and metrics for measuring the environmental impacts of AI, as well as for producing knowledge and research on sustainable intelligence. UN Secretary-General António Guterres urged the parties to design algorithms and infrastructures that consume less energy, and to integrate intelligence tools to optimize electricity use and assist in decarbonization processes.
Large technology companies already recognize the challenge of reconciling the exponential energy demand of data centers with their net-zero climate neutrality goals. Google has seen its emissions increase by 48% in the last five years due to the expansion of its data centers supporting artificial intelligence systems, jeopardizing the tech giant's plan to become net-zero by 2030. Meta, which also aims for net-zero emissions in its value chain by 2030, acknowledged an increase in emissions of 46% in 2022 compared to the previous year. As a rule, companies' strategies to reduce emissions involve purchasing energy from renewable sources to power data centers, decarbonizing their value chain, and offsetting with carbon credits.
Generative AI technologies will continue to expand and be incorporated into people's daily lives – whether as a work tool or for leisure. It's crucial not to lose sight of the fact that this technological boom is happening at a critical moment for the planet, when humanity is already feeling the impacts of the climate crisis – 2024 was the hottest year on record, with the global average temperature 1.55ºC higher, surpassing, albeit temporarily, the critical barrier of the Paris Agreement. Channeling technological advances to address climate change, generating knowledge and implementing energy efficiency and clean energy solutions is more than necessary – it's urgent. To deal with a planet in turmoil, we need AI that is not only generative, but above all regenerative.
* Marina Grossi is president of CEBDS (Brazilian Business Council for Sustainable Development), an organization with 28 years of experience and 120 large member companies.
