Why AI Data Centers Are Not Built Inside Oceans
You must be coming across headlines everyday about how AI data centers are using large amounts of water daily.
A single ChatGPT query consumes about 500 ml of water, roughly one standard bottle, to cool servers. This sounds like a feasible amount, but the countless prompts people generate every second, every day, quickly add up.
But Why Do AI Data Centers Use So Much Water?.
AI data centers use GPUs (Graphics Processing Units) and TPUs (Tensor Processing Units) to process and generate responses.
Every time to type a prompt, intensive computations take place within these GPUs and TPUs to generate and deliver a response. Such large-scale and continuous processing generates a significant amount of heat.
Data centers use vast amounts of water to cool these processors because excessive heat can damage them.
This cooling process ensures that the powerful AI server chips continue to operate in safely and efficiently within their optimal temperature range.
The Real Problem
The major concern about AI data centers is not there daily water usage but the type of water they consume.
Freshwater. Yes, the water used for our daily activities is also used to cool down these hot processors.
As we already know, usable freshwater make up only 0.3% of all water on Earth
At the same time, rising global temperatures, sea levels, climate change and freshwater depletion continue to put increasing pressure on our water resources.
The growing number of AI data centers further increases demand for this precious resource.
While reading this a staggering question might come to your mind.
Why Don’t These Data Centers Use Ocean Water?
These underwater data centers delivered positive results, requiring comparatively less water consumption.
The data centers ultimately didn’t require freshwater, benefited from an immense cooling system due to their placement, and the servers were more reliable than land-based centers.
But then, why isn’t this technique fully operational yet?
The Reasons Behind It
1. Difficult Maintenance:
These centers are placed deep underwater. So, even if a small part inside the capsule gets damaged, the entire capsule has to be brought to the surface, repaired and then placed back underwater. This makes repairs much more costly compared to land-based centers where one can move in directly.
2. Ecological Risk:
We know data centers release heat when processors are in use. This heat is ejected out into the ocean affecting marine life and biodiversity. While studies show this warming is minor under normal conditions, it could compound environmental stress during severe marine heatwaves.
3. Corrosion or Biofouling:
Ocean environments are extremely harsh. Saltwater can easily corrode the outside of the capsule, which may require regular inspections and repairs. Even algae and barnacles may grow on the data centers, creating a buildup layer of marine organisms that can trap heat and reduce heat dissipation.
4. High Initial Capital Costs:
Designing airtight, waterproof vessels that can withstand intense deep-sea pressure requires highly specialized engineering, making initial setup much more expensive than building a standard warehouse on land. Additional maintenance further adds up to the overall cost.
Are There Any Alternatives?
Researchers and engineers are coming up various alternatives that could reduce the dependency on the land-based and water-based data centers. Following are a few proposed solutions:
1. Space-Based Data Centers:
Researchers, scientists and engineers are working together on the idea of off-planet data centers. The concept is to place data centers in space inside satellites. The freezing vacuum of the space can offer instant cooling to these processors, allowing them to continuously dissipate heat.
Startups like Starcloud have launched test satellites equipped with NVIDIA GPUs, and companies like Lonestar are exploring lunar deployments. Furthermore, Google’s “Project Suncatcher” aims to test orbital computing
2. Subterranean Data Centers:
Deep underground environments tend to keep stable ambient temperatures, significantly lowering the heavy HVAC energy demands of conventional facilities. Solid rock formations provide blast resistance, inherent protection from extreme weather like hurricanes and floods, and added security against physical breaches.
Facilities like Norway’s Lefdal Mine Data Center and Sweden’s Pionen bunker use natural rock coolness and adjacent fjord water for highly efficient operations.
3. Immersion Cooling:
Another approach that can be adopted is immersion cooling, which uses specialized, non-conducting fluids to cool down servers. Instead of blowing air across hot processors, the equipment is placed directly into a specialized, sealed tank or chassis filled with dielectric fluid.
Industry leaders such as Castrol and Shell have engineered specialized dielectric fluids to meet the demanding requirements of modern, sustainable data centers.
As AI adoption accelerates, the challenge is no longer just creating faster processors but building data centers that can support the future without exhausting Earth’s limited resources.