Why Tech Companies Are Moving AI Data Centres Into the Ocean

What if the future of AI infrastructure lies beneath the ocean? What sounds like science fiction is already becoming reality, as companies explore underwater data centres to meet AI’s growing demand for power, cooling, and space. The ocean offers a potential solution because seawater can help cool the computers inside these facilities, reducing the need for energy-intensive cooling systems.

What Data Centres Do for AI?

Data centres are facilities that house powerful computers and store large amounts of information. They provide the computing power needed to train AI models and run tools such as chatbots and image generators. When someone asks an AI chatbot a question, the data centre processes the request and generates a response. This everyday use of AI is called inference. A UN-backed study estimates that inference accounts for 80% to 90% of AI models’ total energy consumption.

The Environmental Costs of AI Data Centres

The first challenge is electricity. Data centres need large amounts of power to run their servers and AI systems. The International Energy Agency (IEA) expects global electricity use by data centres to become more than double by 2030, reaching around 945 terawatt-hours a year. This is slightly more than Japan’s annual electricity use. Electricity demand from AI-focused data centres is expected to grow even faster, increasing more than fourfold. However, the IEA notes that the increase in emissions from data centres would be relatively small compared with emissions from the entire energy sector. Using AI in other industries can also help reduce emissions, depending on how the technology is applied.

The second challenge is water. Servers produce heat while processing information, so many data centres need cooling systems to prevent overheating. Some of these systems use large amounts of water. Estimates of future water use vary because they measure different things. One energy research firm projects that use water directly to cool data centres can rise from 222 billion litres in 2025 to 644 billion litres by 2030 if cooling efficiency does not improve. A UN-backed study estimates a much larger water footprint of 9.3 trillion litres by 2030 because it also includes water used to generate electricity for data centres.

Electricity and water are not the only challenges. Data centres also require land, minerals, and electronic equipment. Mining for these materials can damage the environment, while discarded servers and other equipment add to electronic waste. The UN-backed report warns that electronic waste linked to digital technologies can reach 2.5 million tonnes a year by 2030. Poor countries may face a greater share of the environmental costs, even when they receive few of the benefits.

How the Ocean Can Help Cool AI Data Centres?

The ocean offers two potential advantages for data centres: natural cooling and less need for land. Cold seawater can help remove heat from servers, reducing the need for conventional cooling systems. Off the coast of Hainan, China, the company Highlander has installed underwater data centre modules at a depth of 35 metres. The surrounding seawater helps keep the computers cool. The company plans to build a total of 100 modules.

According to Highlander, a completed site could save 122 million kilowatt-hours of electricity, 68,000 square metres of land, and 105,000 metric tons of freshwater each year compared with a land-based data centre of the same size. These figures are the company’s projections for the full project, not independently measured savings from an operational site.

How Floating Data Centres Can Use Ocean Wave Energy?

US startup Panthalassa is developing a different type of ocean-based data centre. Instead of relying on underwater modules, it plans to use floating units called nodes. These units would convert the energy of ocean waves into electricity to power AI chips.

The nodes would process AI tasks at sea and send the results back to land by satellite. Panthalassa raised $140 million to develop this technology. The company planned pilot nodes for 2026, followed by commercial deployments in 2027. However, these are company plans, not confirmed results. The technology has yet to prove that it can operate reliably and efficiently at a commercial scale.

What Microsoft Learned From Underwater Data Centres?

Microsoft tested the idea through Project Natick, which placed a data centre off the coast of Scotland. The experiment found that the underwater servers had a failure rate one-eighth that of servers in a land-based data centre. Microsoft later ended the project but said it would use the findings to guide future research. The company has shared limited public information about why the project stopped.

One challenge is maintenance. Underwater data centres are sealed to protect the equipment, so accessing servers for repairs is much harder than it is on land. Companies must still show that these facilities can be repaired, expanded, and operated at a reasonable cost.

Making AI Data Centres More Sustainable

Underwater data centres are only one possible solution to AI’s growing resource demands. Researchers also recommend improving cooling systems, using more efficient technologies, and choosing locations based on local water availability. Underwater and floating data centres need further independent study, too. Researchers must assess their effects on marine life and determine how easily they can be maintained over time.

To summarize, AI may seem to exist in a virtual world, but it depends on physical facilities filled with powerful computers that generate heat and consume resources. Moving these facilities into the ocean could help with cooling and land use, but it is not a complete solution. More efficient technology, suitable locations, and cleaner electricity will also be important. The challenge is not simply finding a place to run AI. It is finding ways to power it without placing unnecessary pressure on the planet.

Image Source