China has launched an innovative underwater AI data center off the coast of Shanghai, utilizing ocean water for cooling and offshore wind for power, marking a strategic move to align its AI development with sustainable infrastructure.
Constructed by the tech firm Hailanyun as part of a $223 million initiative, the submerged data center maintains its cool without relying on energy-intensive traditional cooling systems. Instead, it circulates seawater through radiator-equipped server racks. The facility is powered almost entirely by a nearby offshore wind farm, which supplies 97% of its energy, leading to near-zero carbon emissions. This combination of natural ocean cooling and renewable energy significantly minimizes the environmental footprint of large-scale data processing.
Inside, the data center houses 198 server racks engineered to handle demanding AI workloads. Hailanyun reports that the system possesses sufficient computing power to train a large language model in a single day. The stable, thermally regulated environment requires minimal human intervention, contributing to its efficiency.
The Shanghai underwater site highlights the inefficiencies of conventional land-based data facilities. Traditional data centers, like those operated by Google and Meta, primarily function on land. While dry air in arid regions protects hardware, it places a significant strain on limited water supplies for cooling, often requiring hundreds of thousands of gallons daily from shared sources. Furthermore, nearly 40% of their electricity consumption is dedicated solely to cooling systems. As AI workloads continue to expand, the demands on these facilities intensify. A recent report from UNESCO and University College London warns that the energy consumption of artificial intelligence, particularly large language models, is becoming unsustainable.
Submerging data centers offers an alternative by leveraging naturally cold seawater and clean offshore wind, thereby substantially reducing both energy and freshwater usage.
However, subsea systems are not without their trade-offs. While seawater cooling lowers energy consumption, it can lead to environmental impacts. The discharge of warmer water can decrease oxygen levels around the capsule, potentially affecting marine life. Hailanyun spokesperson Li Langping has dismissed this concern, stating that the temperature rise was “under one degree and had no substantial effect.”
Other challenges include potential disruptions from sound frequencies. Researchers at the University of Florida have discovered that certain sound frequencies can interfere with underwater servers and have developed a machine learning tool to detect and contain such interference early. Repairs also pose a significant challenge; what is routine on land becomes costly and slow underwater due to limited access and the difficulty in fixing failures.
The concept of subsea data centers is not entirely new. Since 2021, marine technology firm Beijing Highlander has been testing sealed server capsules off the coasts of Zhuhai and Hainan, aiming to transition subsea computing from experimental to established infrastructure. China’s undeniable ambition in this area indicates its intent to compete in the AI landscape far beyond traditional land-based operations.
