New section Objectif Métropoles "Artificial intelligence facing the challenge of ecological intelligence"

This is the case, for example, with the indicators provided by The Climate Company, which supports the redevelopment of territories such as Lamentin in Martinique through a climate-focused lens. "Smart" electrical grids themselves use AI-boosted algorithms. Increasingly, cities are balancing renewable energy supply and demand in real-time, thereby reducing carbon footprints while limiting consumption peaks. Embix, a partner in urban projects such as Paris Batignolles and Lyon Sollys, is a seasoned user of such technology for designing and operating their networks. Public transport adjusted to real-time flows, congestion reduction through connected sensors—here too, AI is reinventing how our cities "breathe." Since 2019, Singapore has been testing AI to optimize port operations. It allocates anchorage spaces to avoid unnecessary accelerations and prolonged waits, thereby reducing the carbon footprint of ships. As a major hub in Asia, the port uses AI and big data to streamline container flows and improve logistics services. These initiatives are part of Singapore's national AI strategy aiming to position the country as a global leader by 2030.

At the beginning of the year, the dramatic wildfires in Los Angeles sparked a media debate about the role of data centers, which are highly concentrated in California. Central to the discussions were the significant water consumption required to cool these facilities, especially in drought-prone areas. Globally, digital technologies are already estimated to account for 4% of greenhouse gas emissions—a figure that the French Environment and Energy Management Agency (ADEME) predicts could double by the end of 2025.

Behind every algorithm lies data centers consuming vast amounts of electricity and relying on water-intensive cooling systems. These infrastructures also depend on rare materials, whose extraction and processing impose heavy environmental costs. For instance, training an AI model capable of generating complex predictions can emit as much CO2 as a transatlantic flight. A 2019 study by researchers at the University of Massachusetts Amherst, reported by MIT Technology Review, revealed that training the GPT-3 language model produced carbon emissions equivalent to those of five cars over their entire lifetimes.

Each query on ChatGPT requires approximately 2.9 Wh, which is 10 to 25 times higher than the energy consumption of a Google search, due to the complexity and computational power required by the AI model. This translates to a daily electricity consumption of 2.9 GWh or 2.9 Wh per prompt on average, depending on the complexity of the responses generated, amounting to approximately 1,058 GWh annually. For comparison, the average annual energy consumption of a French resident is around 2,223 kWh. This means that one year of ChatGPT usage already equals the annual energy consumption of a city with 476,000 inhabitants, comparable to Lyon or Toulouse.

This result, all other factors being equal, does not account for the exponential growth in usage, the environmental footprint of developing AI models, or that of all other existing or future AI applications. The contribution of AI is undeniable and is not in question. The technological revolution must be an ally. However, it seems essential to consider how it can align with the imperative to reduce our energy consumption. How can our territories succeed in this challenge where innovation demonstrates responsibility? How can we make sustainable room for a digital "brain" on a planet with finite resources? This is indeed the question that must be addressed today.