China has begun building a data centre underwater off the coast of Shanghai as part of an attempt to curb the high water and energy usage often required to cool AI servers.
The servers, which handle vast amounts of information and are continually operating, produce tremendous heat, and so they must be constantly cooled to avoid damage to equipment and ensure functionality.
The project formally kicked off in June as China launched efforts to become the leader in the global artificial intelligence race. Nevertheless, this quick growth in AI infrastructure has been greeted with alarm about the huge amounts of water needed to cool data centres.
As Shabrina Nadhila, an Ember energy think tank analyst, notes, “China’s ambitious strategy heralds a drastic move towards low-carbon digital infrastructure, and it can shape the world’s norms for sustainable computing.” She spoke to The Guardian and SourceMaterial, a nonprofit journalism publication.
Why AI Data Centres Overheat
AI data centers are centers of high-capacity data storage and processing, particularly critical for companies more and more reliant on machine learning and automation. Servers are generally jammed into tight spaces and run continuously, which produces waste heat. If not adequately cooled, the heat has the potential to create failures or even data loss.
Cooling uses approximately 40% of a data centre’s overall energy consumption. Most of this is chilling and moving water that is then either sprayed into the air that circulates around the servers or permitted to evaporate, cooling the air. This water is usually drawn from underground aquifers, rivers, streams, or recycled wastewater.
How the Undersea Facility Will Operate
The new sea-level facility is being constructed by the firm Hailanyun and will be powered through renewable energy resources. According to a representative of the firm, the centre will be linked to an existing nearby offshore wind farm, which will be used to provide 97% of the energy needs.
In its first phase, the data centre will consist of 198 server racks, each capable of holding two to four AI-focused servers. Operations are scheduled to commence in September. The system is projected to deliver computing power sufficient to train a large-scale AI model, comparable to OpenAI’s GPT-3.5, in just one day.
To provide the servers with cooling, the facility will utilize seawater pulled through pipes and passed over radiators installed behind the server racks. The water will pick up the heat and transfer it, eliminating the space and energy demands of conventional cooling systems.
Environmental Risks and Concerns
While its pioneering design has pushed boundaries, the submarine route has posed environmental concerns. Scientists warn that during marine heatwaves, sometimes extended periods where sea temperatures rise unusually high, releasing heated water into the ocean might actually exacerbate conditions for marine life.
According to research published in 2022, such release can be less oxygenated and too hot for much of the marine life to thrive.
Additional issues were highlighted in a 2024 report, which found that certain underwater acoustic emissions such as speaker systems’ signals, potentially pose a threat of damaging submerged data centers, raising concerns about structural threats and data protection.
