Tech • AI • Robotics • Game
Should AI Be Slowed Down to Save the Climate? The Shift Project Responds
5/10TL;DR
Digital technology already accounts for about 5% of France’s carbon footprint, and the rapid expansion of AI is intensifying pressure on electricity, materials, water and public policy.
KEY POINTS
Why AI is being singled out
AI is not treated as an isolated issue because it sits inside a broader digital system made up of terminals, networks and data centers. Analysts at The Shift Project, which has worked on digital impacts since 2017, argue that AI deserves special attention because it acts as an accelerator of existing trends, especially the growth of computing infrastructure and electricity demand.
What digital infrastructure actually includes
Digital activity depends on personal devices such as phones, computers and televisions, the cables and telecom systems linking them, and the servers that process and store data. These servers consume electricity directly and also require cooling, which adds further power demand. The environmental burden therefore extends well beyond the visible act of sending a query or streaming a file.
Carbon accounting goes far beyond electricity use
The environmental footprint of digital technology is assessed across the full life cycle, from mineral extraction and manufacturing to use and end-of-life treatment. For data centers, global estimates cited in the discussion suggest roughly 75% of the carbon footprint comes from use and 25% from manufacturing. That embedded footprint is considered essential, because focusing only on operational electricity can produce misleading conclusions.
France’s digital footprint is already significant
In France, the annual carbon footprint of digital goods and services is estimated at around 30 million tonnes of CO2 equivalent, or about 5% of the national total. That places digital emissions slightly above those of the country’s heavy goods vehicle sector. Carbon is only one dimension, as digital systems also rely heavily on metals, minerals, water and electricity.
The biggest emitters inside digital
The carbon split presented for digital technology places data centers at 53%, terminals at 44%, and networks at 3%. That means the debate cannot be reduced to smartphones alone or to AI alone. Still, the fast rise of AI workloads is increasing pressure precisely in the segment that already carries the largest share of emissions.
A lack of transparency clouds AI impact estimates
Precise figures for the impact of an individual AI query remain elusive because companies do not disclose enough standardized data across the value chain. The result is that claims about the water or electricity cost of a single prompt should be treated cautiously. French public authorities, including Colab under the Ministry of Ecological Transition, have pushed for a mandatory international reporting framework so AI systems can be compared on a consistent basis.
Efficiency gains are being overwhelmed by growth
The central warning is not that efficiency is useless, but that it is being overtaken by rising demand. Global data-center electricity use is described as having risen from about 160 TWh in 2014 to roughly 420 TWh in 2024, with annual growth accelerating from around 7% to 13% and, for some AI-related activity, even higher. The pattern illustrates the rebound effect: systems become more efficient per task, but total consumption still surges because usage multiplies.
AI may triple data-center demand by 2030
In trend scenarios that assume current expansion continues, worldwide data-center electricity use could triple between 2023 and 2030, rising toward 1,500 TWh. Separate estimates for cryptocurrencies place their 2024 electricity use at roughly 150 TWh, showing that AI is not the only fast-growing digital load, but one of several forces intensifying infrastructure demand.
Frugality and sobriety are not the same
The debate distinguishes between efficiency, sobriety and frugality. Efficiency means lowering the impact of a given task. Sobriety means reducing the number of uses or avoiding unnecessary ones. Frugality goes further by asking whether an AI system is needed at all before trying to optimize it. That approach is increasingly presented not only as an environmental imperative but also as a potential competitive advantage, especially in a context of sovereignty and public demand for transparency.
Policy choices now matter more than technical progress alone
The broader concern is that AI can be deployed to optimize carbon-intensive sectors rather than replace them. One cited example is the use of AI to improve oil and gas exploration, which speeds up the wrong trajectory instead of supporting decarbonization. That is why the debate is shifting from simple performance gains to questions of governance, sovereignty, open-source digital commons and the political courage needed to align digital expansion with climate goals.
CONCLUSION
The emerging consensus is that AI’s environmental challenge is not just a matter of cleaner chips or better models. It is a question of whether governments, companies and users can impose limits, transparency and purpose before digital growth outpaces every efficiency gain.
Ask a question