AI Data Centers Are Going Underwater To Stay Cool
From underwater restaurants to research labs, the ocean is becoming an unexpected hub for innovation. Now, AI data centers are joining the trend by moving below sea level. As AI demands surge, tech companies are turning to the ocean's natural chill to cool their servers more sustainably.
China Leads the Charge in Underwater Data Centers
China recently began constructing an AI data center off the coast of Shanghai. This facility aims to cut freshwater use by leveraging seawater for cooling. Beyond environmental benefits, this move reflects Beijing’s ambition to advance its AI capabilities. The center will operate with near-zero carbon emissions by tapping into an offshore wind farm for 97% of its energy needs.
Why Move Data Centers Underwater?
AI models require massive computational power, which generates substantial heat. Cooling these servers traditionally consumes vast amounts of freshwater and electricity. With freshwater shortages becoming more critical globally, especially in arid regions, seawater provides an abundant and efficient alternative.
Details of China’s Ocean-Based Initiative
- Location: 6 miles off Shanghai’s coast
- Operator: Hailanyun
- Power Source: Offshore wind farm supplying 97% of electricity
- Capacity: Nearly 800 AI-ready servers
- Launch: Expected operational by September
- Performance: Capable of training large AI models like GPT-3.5 within a day
Energy analyst Shabrina Nadhila noted that China’s approach marks a significant shift towards sustainable digital infrastructure with potential global influence.
How Underwater Cooling Works
Unlike traditional data centers relying on air conditioning and freshwater cooling, underwater centers use pipes that channel cold seawater directly through radiators behind server racks. This method reduces electricity consumption by approximately 30%, according to Hailanyun.
Microsoft’s Early Experiment
Microsoft pioneered underwater data centers with Project Natick in 2018, deploying a server container on the ocean floor near Scotland. The experiment demonstrated increased reliability and energy efficiency. Although Microsoft has paused further development, they continue to study the technology’s potential.
OpenAI’s Growing Infrastructure Needs
OpenAI, known for ChatGPT, faces increasing compute demands as it trains larger models like GPT-4. Microsoft supports OpenAI with specialized AI data centers integrated into its Azure cloud. While OpenAI hasn't confirmed underwater facilities, the industry’s rising energy and water costs are driving interest in such solutions.
Global Interest Beyond China
Other countries are exploring ocean-based data centers. South Korea is advancing similar projects, while Japan and Singapore explore floating data centers as alternatives. However, scaling these initiatives globally requires addressing regulatory, environmental, and logistical challenges.
Environmental Considerations
Concerns exist about potential impacts on marine ecosystems, such as localized warming or oxygen depletion. Early research suggests temperature increases near underwater data centers are minimal—often less than 1°C—and unlikely to cause significant harm. Hailanyun reports negligible environmental impact around their test pods.
Looking Ahead
Underwater AI data centers provide a promising path to reduce freshwater usage and energy consumption in an energy-hungry sector. While China currently leads commercial deployment, other tech hubs are actively researching and testing similar technologies.
For those interested in expanding their AI knowledge and staying updated on infrastructure trends, Complete AI Training offers courses covering AI development and deployment strategies.
Your membership also unlocks:
