Can underwater data centers power the AI age?

Artificial intelligence is fueling an unprecedented expansion in digital infrastructure, and with it a surge in demand for electricity, water, and computing capacity. In 2025, the world's data centers consumed an estimated 448 terawatt-hours of electricity, according to a 2026 report by the United Nations University Institute for Water, Environment and Health. If data centers were treated as a country, they would rank as the world's eleventh-largest consumer of electricity.

The growth trajectory is even more striking. Global demand could exceed 945 terawatt-hours by 2030, enough to supply residential electricity to all 1.3 billion people in Sub-Saharan Africa for about five and a half years. Artificial intelligence is expected to account for a growing share of that increase, with AI workloads projected to rise from roughly one-fifth of data center electricity use in 2025 to 40 percent by the end of the decade. As the race to build AI infrastructure accelerates, the challenge is no longer simply how to provide more computing power, but where to find the electricity, cooling, water, and physical space needed to sustain it.

That challenge is already becoming visible. Data centers face mounting pressure from strained electricity grids, water shortages, and growing opposition from communities concerned about their environmental footprint. As companies race to support increasingly powerful AI models, some are beginning to reconsider one of the digital age's most basic assumptions: that data centers belong on land.

One answer may lie beneath the ocean's surface.

China takes the lead underwater

China has now become the first country to pursue underwater computing at commercial scale.

In May, the Shanghai Lingang undersea data center demonstration project entered operation, marking the latest stage in China's effort to expand the infrastructure needed for the AI era. Located more than 10 kilometers off Shanghai and submerged roughly 10 meters below the surface, the 24-megawatt facility was developed by HiCloud Technology and state-owned China Communications Construction with an investment of 1.6 billion yuan, or roughly $223 million.

The project is powered by a nearby offshore wind farm and relies on the surrounding seawater for cooling. According to Chinese authorities, that combination reduces power consumption by more than 20 percent compared with conventional land-based facilities. The Lingang project builds on HiCloud's earlier underwater data center near Hainan Island, which entered operation in 2023 as the world's first commercial underwater data center. But the Shanghai facility represents a further step by directly integrating submerged computing infrastructure with offshore renewable energy.

Its location is also revealing. Lingang, a high-tech free-trade zone east of Shanghai that is home to Tesla's Gigafactory and other advanced manufacturing facilities, was designed as a hub for emerging industries. The underwater data center points toward a model in which computing infrastructure is located not where people live, but where abundant energy resources are available.

Yet China did not pioneer the idea. Microsoft deployed its Project Natick prototype off the coast of Scotland in 2018 and, two years later, reported that servers operating underwater experienced failure rates roughly one-eighth those of comparable land-based systems. Researchers attributed part of that improvement to stable temperatures and the absence of oxygen and humidity inside the sealed vessel. Despite the encouraging results, however, Project Natick never moved beyond the experimental phase, leaving China to take the concept further with commercial-scale deployments.

Why AI is driving data centers toward the ocean

The appeal of underwater facilities stems from a growing mismatch between AI's demands and the resources required to support them.

A March 2026 study led by researchers at Harvard University estimated that data centers account for more than 4 percent of electricity consumption in the United States and generate more than 105 million metric tons of carbon dioxide emissions annually. Meanwhile,the United Nations University report estimated that global data centers consumed 4.5 trillion liters of water in 2025, enough to fill 1.8 million Olympic-sized pools. By 2030, that figure could reach 9.3 trillion liters.

Cooling systems are a major reason.

Large AI clusters generate enormous amounts of heat, forcing operators to devote substantial energy and water to temperature control. Seawater offers an attractive alternative. The relatively stable temperatures beneath the surface allow facilities to dissipate heat naturally, reducing the need for energy-intensive cooling equipment and lowering freshwater consumption.

The benefits extend beyond water and electricity. Offshore facilities avoid many of the land constraints facing conventional data centers, particularly in densely populated regions. They can also be located close to offshore wind farms, allowing computing infrastructure to take advantage of renewable energy that might otherwise be difficult to transmit over long distances.

That logic is beginning to influence projects beyond China. Samsung Heavy Industries, Supermicro, and Greek shipping company Alpha Ocean Resources recently announced plans for floating AI data centers with capacities of up to 50 megawatts. Rather than placing servers on the seabed, the concept would house computing infrastructure aboard specialized vessels positioned near renewable energy resources.

From proof of concept to reality

Despite promising technical results, underwater data centers face formidable obstacles.

Maintenance remains one of the biggest obstacles. Unlike conventional facilities, submerged systems cannot easily be upgraded or repaired. Major repairs may require bringing large sections of the facility back to the surface, adding complexity and expense.

Microsoft had already demonstrated that underwater data centers could operate reliably and efficiently. Yet after reporting encouraging results, the company did not expand Project Natick beyond its initial pilot. The experiment showed that the technology was technically viable, but turning a successful prototype into commercially deployable infrastructure proved far more difficult. Maintenance, upgrades, and economics presented challenges that outweighed the benefits for a private company.

China appears to be making a different calculation. Rather than evaluating underwater data centers solely through the lens of commercial returns, the Shanghai project forms part of a broader effort to secure the energy and computing infrastructure needed for the AI era. As artificial intelligence becomes a source of economic and technological competition, governments may be more willing than private companies to invest in projects whose strategic value extends beyond short-term profitability.

New risks beneath the surface

Even if underwater data centers prove technically and economically viable, important environmental and security questions remain.

Researchers are still studying the effects of heat discharge on marine ecosystems, as well as long-term concerns surrounding corrosion and maintenance. Some scientists have suggested that underwater structures could function as artificial reefs, attracting marine life, though the consequences of large-scale deployment remain unclear.

Security poses another set of challenges. Researchers from the University of Florida and their collaborators showed in a 2024 study that carefully generated acoustic signals could interfere with underwater data center operations and resource management. Their findings pointed to vulnerabilities unique to submerged facilities and suggested that entirely new forms of cyber and physical attacks could emerge beneath the waves.

Underwater data centers are unlikely to replace conventional facilities anytime soon. Their costs, operational challenges, and unanswered environmental questions remain considerable.  Yet the pressures created by AI are forcing companies and governments to rethink where digital infrastructure should be built. For decades, the geography of computing followed people and cities. In the AI age, it may follow energy, cooling, and access to resources. In some cases, that search may lead beneath the sea.