[LUM#22] AI’s insatiable appetite for energy: a blessing in disguise?
The rollout of AI is contributing to the rapid growth of the digital sector’s carbon footprint. In the face of criticism, the industry is aiming for quick wins to achieve carbon neutrality. These are promises that still need to be proven.
The rollout of AI is increasing the digital sector’s environmental footprint. This is hardly surprising, given that AI derives its performance from processing ever-larger volumes of data, which causes computing demands to skyrocket. This, in turn, leads to a proliferation of processors and data centers, whose operations drive up the sector’s energy consumption. According to the International Energy Agency (IEA), the electricity required to power AI and cryptocurrencies is expected to double between 2022 and 2026.
“The model’s training phase is the most energy-intensive. Training ChatGPT-3 reportedly consumed 1,300 megawatt-hours—200 times more than the annual energy consumption of a French household. Given that improving an AI model requires repeating this training phase several times… The query, also known as the inference phase, consumes ten times more energy than a standard Google search,” explains Edmond Baranes, a researcher at the Montpellier Research in Economics (MRE) laboratory1).
A race to invest in AI
This energy consumption contributes to the sector’s growing carbon footprint. Greenhouse gas emissions from the digital sector are projected to triple by 2050, according to the baseline scenario of a joint report by the French Environment and Energy Management Agency (ADEME) and the French Regulatory Authority for Electronic Communications, Postal Services, and Press Distribution (ARCEP) published in March 2023. And frugality is not on the agenda. “No one is considering reducing digital energy consumption, ” confirms Edmond Baranes. One reason is that the business model of the digital giants—the MAMAA (Microsoft, Amazon, Meta, Apple, and Alphabet)—is based precisely on data. “There is a race to invest in AI because mastering it means producing more data and therefore more value, ” explains the economist, who has been working for years on digital markets, particularly on the relationships between telecom operators and digital content providers.
In response to criticism of its high energy consumption, the sector has two strategies. First, technological innovations to reduce the energy consumption of IT systems. “The arrival of a new generation of processors that consume less energy than GPUs ( TPUs, NPUs, etc.), as well as improvements in algorithms—which are faster and more efficient—will enhance the energy efficiency of AI, ” notes Edmond Baranes. Nevertheless, energy consumption will continue to rise, particularly due to the “rebound effect,” a well-known mechanism describing the persistence of negative impacts from an improving technology. The use of renewable energy to power data centers is also heavily promoted by the MAMAA. Carbon neutrality in the sector remains, for now, a pipe dream. Microsoft committed in 2020 to achieving negative emissions by 2030, but by 2024 its emissions had increased by 30% (Le Monde, June 15, 2024).
Finding the right indicators
AI should also help reduce our consumption indirectly. A 2022 OECD report thus touts the green and digital “dual transition,” based on the idea that products and services delivered by AI are sources of efficiency gains and thus energy savings, by helping to manage energy systems (“smart grids”) or by improving the optimization of transportation and mobility networks. These gains must, however, be weighed against other negative effects of AI deployment, such as the increase in certain “recreational” uses—largely based on the generation of text, images, and videos—which consume ever-increasing amounts of energy.
“Today, even though discussions are underway, the expertise needed to assess the energy savings associated with AI has not yet been fully established. An initiative by Ademe and Arcep aims precisely to conduct an annual survey on sustainable digital technology in order to track changes in digital energy consumption. “The challenge is, in particular, to identify the right indicators and adapt them in line with technological advancements,” notes Edmond Baranes.
In the meantime, the sector is not subject to any particularly strict restrictions on its energy consumption. “There are general policy guidelines on corporate digital responsibility, with incentives that can be incorporated into their CSR initiatives, but there are no specific targets for reducing energy consumption. Fortunately, corporate initiatives are emerging around green computing, data center optimization, and efforts in sustainable product design and recycling, ” notes Edmond Baranes. The European Union passed an AI regulation in late May 2024, the world’s first binding law on artificial intelligence. “The text lists digital services by risk level, but the attention given to energy consumption is very limited.”
A tiered pricing structure for digital services based on data usage also does not appear to be on the table. Even though telecom operators must manage network capacity and, therefore, monitor content providers’ demand for data traffic. “The few awareness campaigns encouraging users to stream during off-peak hours address this concern of the operators. But today, any restriction on Internet access is difficult to justify, as it would be perceived by many as an infringement on a fundamental freedom, ” concludes the economist.
UM podcasts are now available on your favorite platform (Spotify, Deezer, Apple Podcasts, Amazon Music, etc.).
- MRE (UM)
↩︎