The explosion in demand for electricity due to artificial intelligence (AI) is already having local impacts, according to industry experts. Sam Altman, the founder of OpenAI and a prominent figure in the AI sector, has warned that we still don’t fully appreciate the energy requirements of this technology. He made this statement during the Davos summit in Switzerland, where he was not alone in sharing this concern.
A recent report by the International Energy Agency reveals that the electricity consumed by digital data centers worldwide is predicted to double by 2026. This increase is primarily attributed to the growing prominence of AI and cryptocurrencies. The report states that the energy needs could rise from 460 TWh in 2022, which currently accounts for 2% of global demand (with cryptocurrencies making up 25% of that demand), to a staggering 1,050 TWh. This surge is equivalent to adding the energy consumption of a country as large as Germany.
In the United States, data centers could triple their energy consumption by 2030, reaching 390 TWh, according to a study by the Boston Consulting Group. This projection is alarming since it indicates that a data center which used to double its power consumption every five years could soon be doubling it annually. The industry attributes this increase to the introduction of supercomputers equipped with exceptionally powerful graphics cards that are specifically designed for training AI models and handling user requests.
The rise in demand for specialized graphics cards has sparked a competitive rush among industry players. Meta (the parent company of popular social media platforms Facebook and Instagram) plans to acquire 350,000 state-of-the-art cards from Nvidia, the industry leader, by the end of 2024. These cards significantly outstrip the power consumption of normal servers, with a single supercomputer cabinet consuming approximately five times more energy.
The surge in AI-related electricity demand has already had local ramifications. In the United States, it has led to tensions surrounding new data center projects in Northern Virginia’s Data Center Alley. Additionally, an energy supplier in the Kansas City area has postponed the closing of a coal-fired power plant due to increased demand.
In France, where decarbonized nuclear energy is prominent, there have been discussions about the role of data centers. Marseille’s city council voted last October to debate the issue, while deputy mayor and environmentalist Sébastien Barles called for a moratorium. Marseille currently has five data centers and 11 projects under development, as the city is connected to 16 undersea internet cables. Other network hubs such as Amsterdam, London, and Dublin have also sparked debates surrounding the implementation of moratoria on new data center projects. In France, voluntary load shedding during periods of local consumption peaks is already planned to manage electricity demand.
The growing concerns around the surge in AI-related electricity demand highlight the need for a balanced approach. While the expansion of AI presents numerous opportunities, it is crucial to consider the environmental impact and address the energy needs of this technology. Finding sustainable solutions will be crucial to managing power consumption and ensuring the long-term viability of AI development.
