Al, data, and computing: shaping infrastructures for a decarbonised world

The Shift Project’s Digital team is pleased to publish its final report « Al, data, and computing: shaping infrastructures for a decarbonised world ».

You can now download:

• The summary
• The final report

In the midst of the current “generative AI moment”, The Shift Project has chosen to shed light on the topic through its own lens: a physical perspective. To do so, we examined a key component of digital infrastructures—the data center sector—and the way it is being shaped through its interactions with artificial intelligence, which today is the main driver of its dynamics.

The Shift Project’s Digital Program (see all our work) has for several years been conducting and documenting a reflection on the practices and actions that can limit the direct and indirect environmental impacts of digital technologies, without hindering the net positive contribution that some of their potential levers may offer for the ecological transition.

The digital sector is far from negligible: it already accounted for 3 to 4% of global emissions in 2020 (The Shift Project, 2021), on the same order of magnitude as all heavy-duty trucks worldwide (IEA, 2021), with its footprint growing by an average of 6% per year. At the French scale, it represented 4.4% of the country’s carbon footprint in 2022 (ADEME, 2025).

This new project outlines how the widespread deployment of AI is reshaping these already unsustainable trends. It highlights the pathways to follow in order to redirect our technological choices—true political, economic, and strategic decisions—toward energy and carbon sustainability.

Today, contrary to many forecasts, the electricity consumption of data centers worldwide is far from being contained by energy efficiency: the explosion in computing power supply could lead to a threefold increase between 2023 and 2030.

In 2024, more than half of the global electricity consumed by data centers comes from fossil sources, and their future demand is expected to follow the same pattern: far from being tools for decarbonization, these infrastructures are above all an additional problem to manage.

Taking into account production- and construction-related impacts (25% of the total), current trends could lead the data center sector to double or even quadruple its carbon footprint, reaching up to 920 MtCO2e per year in 2030—up to twice France’s annual emissions.

With the footprint of data centers growing at a rate of +9% per year, every year without action amounts to accepting an additional 50 MtCO2e in annual human emissions—equivalent to the yearly emissions of the French livestock sector.

To meet a decarbonization target of –90%, even with the most ambitious possible decarbonization of electricity and production, data centers cannot exceed an annual consumption threshold of 1,000 TWh. Current deployment trajectories are therefore incompatible with this dual carbon constraint, regardless of technological progress.

In Europe, data centers already account for 2.5% of total electricity consumption, with annual growth of 7%. Their consumption could double between 2023 and 2030, and triple by 2035.

Moreover, the geographical distribution of these infrastructures is uneven: it is at the local level that pressures are felt most strongly. In Ireland, nearly 20% of national electricity consumption is absorbed by data centers—equivalent to the entire industrial sector. There, as in Amsterdam, de facto moratoriums are emerging due to the scale of data center demand, which exceeds the capacity of the electrical grid.

In France, data centers represent around 2% of total electricity consumption. This consumption could quadruple by 2035.

By 2035, data centers could account for up to a quarter of the additional electricity consumed compared with 2020. Because this consumption has not been anticipated, allowing the current trend to continue and recent announcements to materialize poses a real risk to the energy transition—particularly for sectors whose full decarbonization will rely on electrification.

Finally, in addition to the pressure on energy systems, other externalities must be taken into account at the local level, such as competition for electricity and water with other sectors (industry, agriculture, for example), job creation, economic attractiveness, and more.

Artificial intelligence refers to the most advanced automation applications, at any given time, in terms of information processing, task complexity, accuracy, and reliability.

The “generative AI moment” is the main driver of trends in the data-center sector, due to its very high computational intensity. It is indeed computing supply that determines the trajectory currently followed by the sector.

From the training phase to the deployment and usage phase, a generative AI model mobilizes more material and computational resources the larger, more versatile, more accurate it is, and the more use cases it serves.

Despite optimization efforts by AI developers, this footprint has continued to grow over the past 10 years: since technological improvement alone is not enough, it is necessary to prioritize use cases.

AI, like digital technologies more broadly, is a catalyst: it accelerates the system in which it is deployed. Rolling it out widely therefore means building an AI infrastructure that will be just as fossil-dependent as the economy in which it operates. Any deployment must therefore be conditioned on verifying its compatibility with carbon constraints—sectoral or territorial carbon trajectories, an organization’s carbon footprint, and its climate targets.

If a solution cannot be deployed in a way that is compatible with this dual constraint, then it must be abandoned and replaced by lower-impact alternatives, whether non-AI or even non-digital.

Our report proposes a step-by-step methodology to address these questions in a systematic, comprehensive, and appropriately scaled manner.

Three major objectives must be pursued:

• Anticipate: it is essential to identify and monitor data-center sites, while providing the digital sector and the data-center industry with a reference carbon-energy trajectory in the SNBC3 and in other planning exercises at both French and European levels. This inventory must be mandatory, and compliance must be monitored.

• Discern: any deployment of an AI solution must be accompanied by an assessment of its relevance, particularly within organizations, in order to identify and prioritize only those applications compatible with the organizations’ climate trajectories.

• Reorient: condition data-center deployments on their compatibility with the sector’s reference trajectory, and phase out AI solutions that cannot be made compatible with the carbon constraint through design levers (optimization and modification or even removal of features) and deployment levers (restricted and targeted rather than broad and indiscriminate deployment).

Steering the decarbonisation of the data centre sector requires decisions that are not only technical but also societal and political.

• Maxime Efoui-Hess, Industry & Digital Coordinator: maxime.efoui@theshiftproject.org
• Huges Ferreboeuf, Digital Project Manager: hugues.ferreboeuf@theshiftproject.org
• Marlène De Bank, Research Engineer
• Pauline Denis, Research Engineer
• Alexandre Theve, Member of the Digital Expert Group
• Lila Wolgust & Lucas Scaltritti, communication

Press contact: lucas.scaltritti@theshiftproject.org