The rapid growth of AI is driving huge demand for data-center computing, electricity, cooling and land, prompting companies to explore unconventional infrastructure. One option is moving servers into the ocean, where seawater can provide natural cooling and underwater facilities can reduce dependence on fresh water and scarce land. The idea is not entirely experimental: Microsoft tested a sealed underwater data center near Scotland’s Orkney Islands in 2018, containing 864 servers. After two years, Microsoft reported that its servers failed at roughly one-eighth the rate of comparable land-based servers.
Several countries are now pursuing different versions of the concept. China began full commercial operation of an underwater data center in 2026, using seawater cooling and offshore wind power; the project reportedly uses at least 30% less electricity than conventional data centers. Japan is testing container-based data centers on floating platforms powered partly by solar energy, while Singapore is developing a large floating facility scheduled to open in 2028. South Korea is also planning an underwater facility designed to house more than 100,000 servers. These projects demonstrate that ocean-based computing could become a meaningful alternative as AI infrastructure expands.
However, moving data centers underwater does not automatically make AI infrastructure environmentally sustainable. The biggest concern is heat: servers transfer large amounts of waste heat into surrounding seawater. Even a roughly 1°C temperature increase near individual facilities could become significant if many installations operate in the same marine environment. Warmer water can affect oxygen levels, marine organisms, breeding patterns and migration. The concern is particularly important because marine ecosystems are already under pressure from climate change and other forms of human activity.
There is also a major maintenance and scalability problem. A failed server in a conventional data center can be replaced relatively quickly, whereas repairing underwater equipment may require bringing an entire sealed module back to the surface. Microsoft ultimately ended its underwater project in 2024 and continued focusing on land-based facilities, where equipment can be upgraded and repaired more easily. The broader lesson is that underwater data centers could reduce some of AI's demands on fresh water, land and cooling energy, but they also transfer environmental and operational pressures to the ocean. The technology will only qualify as a genuinely sustainable AI-infrastructure solution if those new ecological, maintenance and regulatory challenges can be solved.
