Frugal AI, an essential transition
A resource-hungry technology
The rise of artificial intelligence comes with an exponential increase in electricity and natural resource consumption. Behind the image of an intangible technology lies a much more concrete reality: the infrastructures supporting AI are energy-intensive, water-consuming and significant sources of greenhouse gas emissions.
The International Energy Agency (IEA) warns that data center electricity consumption could double by 2026, exceeding 1,000 TWh—the equivalent of Japan’s annual consumption. Generative AI, in particular, is accelerating this surge. A single query on ChatGPT consumes ten times more electricity than a Google search. While AI currently accounts for only a tiny fraction of global consumption (0.03%), this share could skyrocket.
Beyond being a simple energy issue, water consumption is also becoming a central concern. Cooling servers requires massive amounts of water. A study by the University of California estimates that twenty-five ChatGPT queries use half a liter of freshwater. On a large scale, this becomes a major drain. Meanwhile, greenhouse gas emissions from tech giants are soaring: +13% for Google between 2022 and 2023, +48% over five years. Microsoft follows the same trend, with a 40% increase in emissions over four years.
Faced with this reality, the concept of frugal AI is emerging as an alternative.
Frugal AI: using AI as much as necessary, but as little as possible
In a world where the dominant trend is to automate everything indiscriminately, frugal AI takes a radically different approach: reducing resource consumption and questioning the very necessity of using AI.
The CNRS defines frugality as a minimal use of resources, adapted to real needs. Unlike the idea of omnipresent AI, this approach is based on a fundamental question: is AI truly necessary? In businesses, administrations, local governments, and even private use, automation has become the norm. This habit fuels a technological race where the environmental impact is rarely considered.
However, it is possible to design more energy-efficient AI models. While generative AI is an immense energy consumer, traditional models offer similar performance at a much lower cost. Even better, compressing AI models can reduce their footprint without significantly affecting their effectiveness. A 45 MB model can be reduced to just 6 MB, while maintaining a satisfactory level of accuracy. Optimizing instead of overproducing—this is the challenge of this new approach.
A vision driven by France
While the United States and China dominate the race for generative AI, France is choosing a different path. Rather than competing on raw power, it aims to position itself as a pioneer in more sustainable and responsible AI.
In Lyon, the IA.rbre program embodies this vision. Far from consumer-oriented applications, this initiative leverages artificial intelligence to identify the best locations for planting trees, with the goal of creating cooling islands in the summer. A concrete application of AI in addressing climate challenges.
Beyond local experiments, France is establishing a comprehensive framework. In January 2024, a working group dedicated to frugal AI was created, bringing together scientists, government bodies, businesses, and experts. From these discussions emerged, in June 2024, the world’s first general framework for frugal AI. While not legally binding, this document represents a decisive step: from now on, digital actors have a common framework to rethink AI usage.
A model for Europe and beyond
This French initiative is not an isolated effort. Upon its publication, the framework was presented to European institutions, where it received a positive response. The debate on digital sustainability is now entering the regulatory sphere, complementing the AI Act, the European Union’s landmark legislation adopted in May 2024 to regulate artificial intelligence.
By 2026, the enforcement of this regulation will impose new obligations on companies in the sector. In this context, France is advocating for the creation of an international coalition for sustainable AI, bringing together governments, industries, and researchers. The goal: to embed this reflection on a global scale and translate the principle of frugality into concrete actions.
Breaking free from technological blindness
Far from being a mere counter-narrative, frugal AI raises a fundamental question: how far should we go in automating the world? While artificial intelligence itself is not a threat, its unchecked and thoughtless use certainly is.
Failing to address this question means accepting a model where innovation comes at the expense of resources and the climate. Regulating does not mean hindering progress but giving it direction. A direction that moves away from reckless escalation, questions each application, and prioritizes efficiency over excess.
Frugal AI is not a step backward. It is a choice: the choice to take control of technological development rather than passively undergo it.
