Digital transformation has turned data centers into the critical infrastructure par excellence. But the growing demand for computing, now driven by artificial intelligence, is leading operators to explore previously unthinkable locations. From the icy regions of the Arctic to the ocean depths, the quest for energy efficiency and sustainable cooling is redefining the cloud map.

Nearly 15 years ago, Facebook (now Meta) opened its data center in Luleå, a Swedish town located in the Arctic Circle. Mark Zuckerberg explained at the time that they took advantage of the low outside temperature to cool the servers naturally. More recently, in 2018, the underground data center Lefdal Mines in Norway, located in an old iron mine, went into operation. This complex boasts being one of the greenest in Europe, powered by renewable energy and cooled with seawater from the adjacent fjord, which remains at about 8 degrees Celsius.

The impact of AI on energy demand

The emergence of artificial intelligence has intensified this race for efficiency. The increase in computing needs associated with language and computer vision models has led to proposals as bold as Google's to put data centers in space. However, besides looking outward, there is still much surface on the planet to explore, and 70% is covered by oceans and seas.

Liquid cooling is becoming an indispensable requirement in the sector, compared to conventional air cooling. With AI workloads, thermal density skyrockets and liquid cooling is more efficient. But opting for this technology has a cost: a 2026 study by the Institute for Water, Environment and Health of the United Nations University (UNU-INWEH) reveals that the water footprint associated with AI is reaching worrying levels. For example, the electricity consumption of data centers in 2025 was equivalent to 4.5 trillion liters of water, enough to fill 1.8 million Olympic swimming pools or cover the basic domestic water needs of about 600 million people in sub-Saharan Africa.

As a result, the demand for water use efficiency is increasing. A recent report by the AFCOM association indicates that 48% of AI data center users include water consumption assessment as one of their essential criteria, when just two years ago the main criteria were power and cooling capacity. Part of this interest also responds to social, regulatory, and shareholder pressure to demonstrate environmental responsibility. This context has opened a fertile ground — an ocean? — to explore data center modalities that offer other possibilities.

Developing underwater AI: from Hainan to Lingang

In May 2026, the Shanghai Lingang underwater data center was presented, located ten kilometers offshore from the coast of this Chinese city and ten meters below the sea surface. With an initial capacity of 24 MW, it is a joint initiative between HiCloud Technology and the state-owned China Communications Construction. It runs on energy supplied by a nearby offshore wind farm. According to Chinese authorities, the infrastructure consumes one-fifth less energy than its land-based counterparts, thanks to the cooling effect of water. The project, announced in 2025, continues the tests carried out by HiCloud on Hainan Island, with submerged facilities at 30 meters that have been operating since 2022.

If Hainan was the first commercial underwater data center, Lingang is the first submerged one powered by offshore wind. But such facilities had been tested before. Microsoft's Project Natick kept the Northern Isles data center at the bottom of the sea in the Orkney Islands (Scotland) for two years. A facility at 36 meters depth, the size of a container, which remained sealed and loaded with 12 racks and 864 servers. To preserve the equipment, the atmosphere was filled with nitrogen; there was no active cooling, but it came from the surrounding waters. The container was extracted at the end of 2020.

“Submerging a data center brings several advantages over the traditional model,” explain from Microsoft, among which they cite a stable and naturally cooled environment, “which can contribute to more efficient operation.” In fact, one of its main conclusions was reliability: these servers recorded a failure rate eight times lower than that of a control group on land. In addition, the infrastructure operated on 100% renewable electricity produced locally. They point out other advantages, such as speed of deployment, with less than 90 days from factory to operation, and its operation without maintenance for long periods, with an expected operation of up to five years. “Project Natick demonstrated that underwater data centers can be a viable alternative from a technical, operational, and environmental point of view,” they summarize.

Asked about the effects on the environment, Microsoft highlights that this was one of the aspects evaluated, monitoring the environmental conditions around the underwater data center to observe marine life and analyze possible changes. In addition, the seabed was restored to the state it was in before deployment, as well as the recovery and recycling of the module and materials. “The project results helped demonstrate that it is possible to design and operate such facilities with an approach aimed at minimizing their environmental impact,” they insist, without specifying beyond this impact.

Towards underwater AI? Challenges and prospects

So, will we see our coastlines turned into a landscape of data centers? It does not seem likely, at least in the short term. From the SpainDC association they clarify that “it is still too early to know if these advantages compensate for the technical and economic challenges.” “On paper, it can offer advantages in certain environments, but it also introduces greater complexity in areas such as maintenance, access, connectivity, and operation.” This does not prevent them from recognizing that they could fit in some cases, given the current increase in computing and cooling needs derived from AI.

“But it does not seem reasonable to consider it a general solution,” they assess. Specifically, regarding its viability in Spain or Europe, they call for analyzing each project case by case. “It is not enough to have a coast. It would be necessary to study the electrical and telecommunications connection, maintenance, permits, environmental impact, and economic viability.” They do recognize its potential to reduce land occupation “and, depending on the design, the use of freshwater for cooling,” but emphasize the need to rigorously evaluate its impact on the marine environment, its maintenance, and its removal at the end of its useful life. “Its development will depend on demonstrating, with real experience and at scale, clear advantages in costs, reliability, maintenance, and environmental impact,” they insist. “For now, it seems more prudent to think about specific applications than a generalized extension of the model.”

In any case, the option of the ocean as a location for data centers will continue to be explored, whether to submerge them completely or in other modalities that take advantage of the environment. Tech mogul Peter Thiel is making million-dollar investments in the startup Panthalassa, which proposes floating data centers. In Japan, it is a matter of time before facilities that use used ships for this purpose are launched. In a context dominated by the growing demand for AI, no one is willing to leave a potential goldmine unexplored; especially when it is so abundant.

This type of infrastructure innovation also has implications for security and risk management. Agentic AI unleashes a new era of cyber risks, and underwater data centers are not immune to it. Furthermore, energy efficiency and sustainability are key in the adoption of advanced cloud solutions, such as those described in Advanced solutions in Microsoft Azure: beyond the basic cloud.

Original source: ComputerWorld. Analysis and adaptation by ForgeNEX.