In the French Mediterranean port of Marseille, three data centers operated by Digital Realty are pioneering water-saving data center cooling techniques by utilizing water from an old coal mine. This initiative comes as artificial intelligence (AI) infrastructure grows increasingly resource-intensive, consuming vast amounts of energy and water. The environmental impact of AI is a major focus of February’s global AI summit in Paris, emphasizing the need for sustainable cooling solutions.
The Marseille data centers extract water at a rate of approximately 3,000 cubic meters per hour. This water, while unsuitable for human consumption due to its high mineral content, is effectively used in a closed-loop system known as “river cooling.” The system leverages the coal mine water, which maintains a steady temperature of 14°C (57°F) year-round, to regulate the temperature of server rooms. This innovative approach eliminates the need for energy-intensive air conditioning, reducing overall power consumption by 20%.
Maintaining optimal conditions for AI-driven operations is crucial. Data centers must ensure that servers remain at a stable 25°C with atmospheric humidity between 60% and 80% to prevent overheating. If cooling systems fail, temperatures can soar to 45°C within ten minutes, potentially disrupting operations. AI chips, particularly those designed by companies like Nvidia, generate significant amounts of heat and require more advanced data center cooling solutions compared to traditional processors.
As AI adoption expands, the demand for high-performance computing continues to rise, necessitating more efficient cooling methods. Many data centers still rely on air conditioning or water-evaporation systems, such as adiabatic cooling, which Microsoft and other major players employ. However, alternative cooling techniques, including ambient air cooling and immersion cooling, are also being explored. Microsoft is testing immersion cooling, where hardware components are submerged in specialized coolants to dissipate heat. Despite its effectiveness, widespread deployment is limited due to environmental concerns associated with PFAS-based chemicals.
Water consumption remains a pressing issue for the AI sector. Google’s data centers used 24 million cubic meters of water in 2023, marking a 14% increase from the previous year, while Microsoft’s consumption surged by 22%. Given these rising figures, sustainable data center cooling methods are becoming more critical. Companies like Digital Realty, Data4, and Microsoft are implementing localized cooling strategies to minimize environmental impact based on regional water availability.
Despite the benefits of river cooling, the approach has faced opposition in Marseille. Local government officials and campaign groups argue that Digital Realty is monopolizing a valuable water resource. While the extracted water is undrinkable, critics suggest it could be repurposed for agriculture, street cleaning, or public parks. Concerns have also been raised about potential “greenwashing,” with activists emphasizing the need for a holistic evaluation of AI infrastructure’s environmental impact, from raw material extraction to final energy consumption.
As AI technology advances, developing sustainable data center cooling solutions is essential to mitigate its environmental footprint. While river cooling presents a promising alternative, broader considerations, including energy use and resource allocation, must be addressed to ensure long-term sustainability.
https://www.geo.tv/latest/588584-data-centres-chase-water-energy-savings-as-ai-race-ramps-up