Artificial intelligence (AI) is transforming many sectors, but its exponential development is raising growing concerns about its environmental impact. In particular, the energy consumption of data centers, essential for powering AI algorithms, is becoming a major issue. With energy requirements expected to double by 2030, tech companies including Google, Microsoft, Meta and Amazon are looking for alternative solutions to limit their carbon footprint while maintaining growing computing power.
Energy-hungry data centers
Data centers are crucial infrastructures for running AI, but they are also responsible for considerable energy consumption. Currently, these centers use around 1% of the world's electricity, a share that continues to rise with AI demand. A study by the French Economic, Social and Environmental Council (CESE) revealed that a single query to a generative AI, such as ChatGPT, consumes ten times more energy than a standard Google search. As the use of these technologies increases, the report predicts that by the end of the decade, AI and data centers could account for 3% of global CO2 emissions, five times more than in 2020.
To address this issue, Google has taken an unprecedented step: the company has announced that it has ordered six to seven mini-nuclear reactors (MNRs) from Kairos Power, a Californian startup specializing in this technology. The aim is not to compete with EDF, but to guarantee a decarbonized power supply for its data centers used by its AI applications. This initiative, a world first, could inspire others. According to experts, the electricity needs of AI systems could double by 2030, while giants like Google, Meta, Microsoft and Amazon have already absorbed 29% of the world's new wind and solar contracts in 2023.
Nuclear power, a controversial but pragmatic solution
Google's announcement sparks debate. Nuclear power, although decarbonized, remains controversial due to the risks associated with radioactive waste and safety. However, it offers a solution for meeting the growing demand for electricity in data centers without resorting to fossil fuels. By turning to mini-nuclear reactors, Google hopes not only to stabilize its energy supply, but also to significantly reduce its carbon emissions.
Other companies such as Microsoft and Amazon are also exploring nuclear partnerships to power their data centers. These initiatives demonstrate the extent to which technology companies are aware of their environmental responsibilities and are seeking to innovate in order to minimize their ecological footprint. However, the use of nuclear power remains a controversial option and will need to be framed by strict regulations.
The limits of renewable energies
If nuclear power is one of the avenues being explored, it's partly because renewable energy sources, such as solar and wind power, are not always sufficient to cover the massive and constant energy needs of data centers. In 2023, Google, Meta, Microsoft and Amazon purchased nearly 30% of the world's new renewable energy contracts, but these investments alone will not be enough to power the growing demand for AI. Indeed, weather conditions can affect the availability of these energies, and periods of high demand often outstrip production capacity.
Google's use of nuclear power is therefore a pragmatic response to the current limitations of renewable energies. As a report by NotebookCheck points out, dependence on fossil energy sources has recently increased in some regions, particularly when green energy is not sufficient to power data centers.
A race for energy efficiency
Faced with this situation, technology companies are investing in the research and development of new solutions to improve the energy efficiency of their infrastructures. A recent breakthrough in AI, unveiled by scientists, could reduce data center energy consumption by up to 95%, according to Futura-Sciences. This breakthrough relies on more optimized algorithms and less power-hungry hardware systems, offering a promising path to reducing the environmental footprint of tech companies.
Energy efficiency is becoming a priority for industry players, as they try to reconcile their growth needs with sustainability imperatives. In addition to using alternative energies such as nuclear power, companies are looking to maximize the performance of their infrastructures while minimizing their consumption.
A growing global impact
The environmental impact of AI is not limited to electricity consumption. Artificial intelligence technologies also require significant quantities of rare materials, used in specialized processors and other hardware components. The production of this equipment contributes to the depletion of natural resources and generates additional emissions. What's more, data centers generate considerable heat, necessitating energy-hungry cooling systems, which further accentuate their consumption.
The EESC recently alerted a plenary session to these issues, stressing the importance of regulating the activities of technology companies to ensure that their growth does not come at the expense of the environment . According to the EESC, companies will need to adopt more responsible practices, while investing in cutting-edge technologies to limit the environmental impact of data centers.
Towards sustainable AI
The road to more sustainable AI still seems a long one, but initiatives such as those taken by Google with Kairos Power are positive signals. By combining innovations in energy efficiency with the use of decarbonized energy sources, companies can hope to mitigate their environmental impact while continuing to grow. Nevertheless, ethical and environmental questions remain, particularly around nuclear waste management and infrastructure safety.
.The technology industry will also have to respond to growing pressure from regulators and consumers, who expect concrete solutions to reduce the carbon footprint of digital giants. As AI develops, the quest for clean, efficient energy will become a central issue in ensuring a sustainable future.
Sources:
- CESE, "What impacts will artificial intelligence have on the environment?", October 15, 2024, available at: https://www.lecese.fr/actualites/quels-impacts-de-lintelligence-artificielle-sur-lenvironnement-seance-pleniere-en-direct
- The Guardian, "Google to use nuclear power to power its data centers", October 15, 2024, available at: https://www.theguardian.com/technology/2024/oct/15/google-buy-nuclear-power-ai-datacentres-kairos-power
- NotebookCheck, "Rising AI demand forces tech companies to forgo greenest energy sources to power data centers", October 15, 2024, available at: https://www.notebookcheck.biz/L-augmentation-de-la-demande-en-IA-oblige-les-entreprises-technologiques-a-renoncer-a-utiliser-les-sources-d-energie-les-plus-ecologiques-pour-alimenter-les-centres-de-donnees.901868.0.html
- Futura-Sciences, "Scientists unveil AI energy breakthrough: 95% savings", October 12, 2024, available at: https://www.futura-sciences.com/tech/actualites/technologie-scientifiques-devoilent-percee-energetique-ia-95-economies-116738/
This post was translated from the original French.