AI Companies Eye the Ocean for Data Center Expansion
Artificial intelligence companies are increasingly looking to the ocean as a location for data centers, driven by the explosive growth in computing demand that AI workloads require. While the idea of submerging server infrastructure beneath the waves may seem like a practical solution to space and cooling constraints on land, researchers caution that relocating hardware to the sea does not eliminate the underlying environmental challenges — and may introduce new ones.
Data centers are among the most resource-intensive facilities in the modern economy. They consume vast amounts of electricity and require continuous cooling to prevent hardware from overheating. On land, this cooling burden typically means enormous quantities of water or energy-hungry air conditioning systems. Proponents of underwater data centers argue that the ocean provides a natural, near-unlimited heat sink, reducing the need for active cooling infrastructure and potentially lowering operational energy costs.
The concept is not entirely new. Microsoft conducted its Project Natick trials, sinking a prototype data center off the coast of Scotland, and reported promising results in terms of reliability and cooling efficiency. The renewed interest from AI companies reflects how dramatically compute demands have grown since those early experiments, with large language models and other AI systems requiring far more processing power than conventional workloads.
However, experts highlight that shifting data centers to the ocean does not make the fundamental problems disappear. Power still has to come from somewhere, and unless that electricity is generated from renewable sources, the carbon footprint of underwater facilities remains significant. There are also concerns about the impact on marine ecosystems — heat discharge into surrounding water, electromagnetic fields from power cables, and the physical presence of equipment on or near the seabed could all affect local marine life. Maintenance and repair operations in underwater environments are also considerably more complex and costly than on land, raising questions about the long-term viability and safety of such installations.
Regulatory frameworks for underwater data infrastructure remain underdeveloped in many jurisdictions, adding another layer of uncertainty. As AI companies scale up their ambitions, the conversation around where and how to build the physical backbone of artificial intelligence is likely to intensify, with the ocean representing just one front in a broader search for sustainable compute capacity.
