The notion that the solution to one of the dirtiest issues in technology may be hidden at the bottom of the ocean has an almost poetic quality. Between 240 and 340 terawatt hours of electricity are used annually by data centers, those enormous, humming warehouses that power everything from hospital diagnostic software to your morning Spotify playlist. It’s not a rounding error. That’s about the yearly energy consumption of a number of mid-sized nations, and a large portion of it is used for one routine but costly task: keeping machines cool.
Engineers were therefore keeping a close eye on Microsoft when it lowered a white, cylindrical pod, roughly the size of a shipping container, into the frigid waters off the Orkney Islands in 2018. The experiment known as Project Natick sounds ridiculous until you look at the numbers. Only six of the 855 servers failed during the course of two years of operation at a depth of 36 meters. That is a failure rate of 0.7%, as opposed to almost 6% in a similar land-based facility. The cold, steady, nitrogen-rich atmosphere seemed to do something that is difficult to achieve on land with millions of dollars’ worth of air conditioning infrastructure: it maintained stability.
It’s difficult to ignore the implications of that outcome. The indifferent, ancient, and genuinely cold ocean may be a better server than any structure created by humans. Approximately 45% of the energy used in a data center can be attributed to conventional cooling systems. The carbon math changes significantly when you remove that almost completely, as underwater placement permits. Highlander Digital Technology in China has gone so far as to report cooling power reductions of up to 90% in its Shanghai and Hainan Island facilities. Even a small portion of that savings applied at scale would be significant, though it’s still unclear if those figures hold up over time.
And yet. The ocean is more than just a free refrigerator. Concerns that weren’t included in the initial pitch decks have been brought up by marine scientists. Temperature stability is essential to the survival of many coastal species, including fish, invertebrates, and marine mammal breeding populations. Food chains may be disrupted by localized thermal pollution from dense underwater server farms in ways that are challenging to predict and more difficult to undo.
Concerns have also been raised about the acoustic disruption during installation, the electromagnetic fields from power cables, and the permanent physical damage to seabed habitats. Regulators rejected NetworkOcean’s proposal for a pilot facility in San Francisco Bay almost immediately. Some Silicon Valley residents were probably taken aback by that response. It shouldn’t have.
Then there is the harsh reality of saltwater. It corrodes. It builds up. Through a process known as biofouling, marine organisms such as barnacles, algae, and microscopic creatures cling to any submerged surface, gradually reducing cooling efficiency and making hardware access more difficult. Retrieving a malfunctioning deep-sea pod necessitates ships, lifting equipment, and the kind of offshore logistics budget that unnerves CFOs, in contrast to a land-based server room where a technician can replace a failed drive in twenty minutes. To its credit, Microsoft was truthful about this. In 2024, the company closed Project Natick, citing commercial realities rather than technical failure. It would be far more expensive to build these things at scale with the marine engineering, grid connections, and offshore installation vessels than it would be to just build another data hall on dry land.
The underwater idea seems to be on hold rather than dead. In order to simultaneously address the energy and cooling issues, Northern European feasibility studies are covertly investigating ways to integrate offshore wind infrastructure with submerged data storage. Although projections for emerging technology sectors tend to be overly optimistic, market analysts predict that the industry could grow from $1.5 billion today to over $6 billion by 2033. It appears more certain that the industry will continue to face pressure to lessen its environmental impact. AI is driving up energy demand more quickly than most energy transition plans had predicted. There must be a compromise. It’s still genuinely unclear whether the next wave of digital infrastructure will be absorbed by the seabed or if the costs and ecological risks will be too great. The icy chasm is patient. The sector might not be.
