Underwater Data Centers: How Offshore Wind Is Transforming Digital Infrastructure

What a Project Off the Coast of Shanghai Tells Us About Where This Industry Is Headed

Key Numbers Worth Knowing

• Global electricity consumption tied to data centers is on track to hit roughly 945 TWh by 2030, driven mainly by AI workloads, cloud computing, and hyperscale buildouts, according to the International Energy Agency.
• China's Lin-gang offshore underwater data center has already achieved cooling energy savings of up to 22.8%, setting a new benchmark for efficient thermal management.
• Underwater facilities can cut physical land use by more than 90% compared with a traditional terrestrial data center, which matters enormously in places where land is scarce or expensive.
• Cooling alone typically eats up 30% to 40% of total energy consumption in a modern data center, making thermal management one of the biggest levers on both cost and emissions.
• AI servers now need 5 to 10 times the power density per rack compared with standard enterprise IT racks, and that's pushing the industry hard toward liquid cooling, immersion cooling, and now underwater deployment.
• 2024 marked the first time global cumulative offshore wind capacity crossed 1 terawatt, opening a real path toward data centers that draw directly from large-scale renewable generation.
• Microsoft's Project Natick found that submerged servers were eight times more reliable than their land-based counterparts, largely thanks to a controlled environment with far less human handling.
• Microsoft, Google, Amazon, and Meta have together committed more than $300 billion toward AI and data center infrastructure for 2025 and 2026, which only raises the stakes for finding energy-efficient, sustainable ways to build.
• Global data center capacity is expected to more than double within the next decade as AI adoption accelerates, meaning power demand for this sector will outpace overall electricity demand growth by a wide margin.

Why the Industry Can't Keep Building the Way It Has Been

Data centers rank among the most resource-hungry structures humans have ever built. They need power around the clock, enormous volumes of cooling water, sizable land footprints, and direct access to high-capacity fiber. As generative AI, cloud services, and edge computing keep pushing demand higher, the way this industry has traditionally been built is starting to run into hard physical limits.

Global data center electricity use sat somewhere between 415 and 460 TWh in 2023, roughly 1.5% to 2% of total global electricity demand, based on IEA figures. If AI adoption keeps growing at its current pace, that number could top 945 TWh by 2030, putting it in the same range as Japan's entire national electricity consumption today.

Water use tells a similar story. A typical hyperscale facility burns through 3 to 5 million gallons of freshwater a day for cooling, roughly what a small or mid-sized city consumes. Google's data centers used more than 5.6 billion gallons in 2022, up 20% from the year before, while Microsoft's usage hit 1.7 billion gallons over the same period, a 34% jump. According to the World Resources Institute's Aqueduct tool, more than a quarter of all current global data center capacity sits in regions already classified as high or extremely high water stress.

Land is its own separate headache. Prices in the Ashburn, Virginia corridor, home to the world's largest concentration of data centers, have climbed more than 300% over the past decade. Singapore's own moratorium between 2019 and 2022 was driven directly by grid capacity and land limitations. A 100 MW facility, once you factor in support infrastructure and required setbacks, needs somewhere between 10 and 20 acres. An underwater facility, by contrast, can shrink that footprint by more than 90%.

What This Actually Means for the Industry

At CMI Consulting LLC, our team has been watching this space closely, and the direction is fairly clear. If global data center energy demand does reach 945 TWh by 2030 and AI workloads keep growing at their current pace, operators will need to think differently, not just about what data centers are built with, but about where and how they're built into the physical environment itself. Offshore wind and ocean energy powering submerged facilities may need to move out of the pilot stage and into practical, repeatable infrastructure within the next ten years. Utilities, investors, cloud providers, and governments will all need to find ways to build fast without sacrificing sustainability, and underwater data centers look like one of the more credible paths toward doing both at once.

Inside the Technology: How an Underwater Data Center Actually Works

An underwater data center is essentially a modular system engineered to handle marine conditions, thermal loads, server hardware, and power delivery all at once, sitting on or near the seabed. Each module sits inside a pressure vessel made from marine-grade steel or high-tensile composite materials, shaped either cylindrical or rectangular to withstand the surrounding hydrostatic pressure. At an operating depth of 100 to 400 meters, that pressure runs somewhere between 10 and 40 atmospheres, and the housing also has to resist biofouling and provide a degree of electromagnetic shielding.

Inside the module, the environment is filled with nitrogen gas kept slightly above atmospheric pressure, which slows metal oxidation and reduces convective heat transfer within the casing. Power reaches the module through a submarine cable, either routed from land or from an offshore power source, most commonly a wind farm. Data moves through submarine optical fiber, part of a global network stretching more than 1.3 million kilometers and carrying over 95% of intercontinental internet traffic.

The Lin-gang Project: Proof This Can Work Commercially

In late May 2026, the world's first wind-powered underwater data center officially went live off the coast of Shanghai, in China's Lin-gang Special Area. The facility was built by a subsidiary of China Communications Construction Company, but what really stands out is the timeline. Phase one construction wrapped in October 2025, and the whole project was fully commissioned roughly seven months later, a pace that would be nearly impossible for a conventional land-based facility of comparable scale.

The site currently runs at 2.3 MW, with a planned total capacity of 24 MW spread across 192 server racks over four levels. It uses a closed-loop cooling system with copper pipes circulating seawater as the heat exchanger, cutting cooling-related electricity use by 22.8% and drawing nearly 95% of its total power from offshore wind. It uses no freshwater at all.

The gap between the planned 24 MW capacity and the 2.3 MW currently running reflects a phased hardware rollout, the same approach any hyperscale operator takes when managing capital spend against uncertain demand. In practice, this means the facility is essentially future-proofed, with more compute hardware likely to come online as AI inference demand grows.

That said, real challenges remain, particularly around how heat gets released back into the surrounding marine environment. A 2023 peer-reviewed study published in Energy Reports looked at heat plume dispersal from a 10 MW submarine facility at 30 meters depth and found sustained temperature increases of 0.5°C to 2.2°C within a 150 to 250 meter radius, a range that overlaps with zones where benthic invertebrates and some coral populations are vulnerable to thermal stress.

There's also no established regulatory framework yet for hardware lifecycle management or disposal in these environments. And maintenance access relies on remotely operated vehicles or dive teams, a very different model from the always-available access land-based facilities take for granted.

How Governments Are Responding

• China: The Lin-gang project sits inside a policy framework laid out in the 14th and 15th Five-Year Plans, both of which name digital infrastructure a strategic priority. The National Development and Reform Commission has classified offshore data infrastructure as eligible for preferential financing under the Green Development Fund, with interest rates reportedly 80 to 120 basis points below market. China's offshore wind buildout, 42.5 GW installed as of 2025 and a target of 150 GW by 2030, gives it an energy base that few other countries can currently match at this scale.
• EU: The recast Energy Efficiency Directive (2023/1791) now requires data centers above 500 kW to report energy use, water consumption, and waste heat recovery to national authorities. Horizon Europe has put €120 million behind the OceanData consortium, a twelve-nation research effort combining submarine data infrastructure with environmental sensing. Denmark and Germany have already issued proposals for energy islands that link data center capacity with offshore wind and green hydrogen production.
• United States: The Inflation Reduction Act offers investment tax credits of up to 30% for offshore wind and related transmission infrastructure, and there's active debate over whether co-located compute facilities qualify. The Department of Energy's Office of Electricity issued a request for information in early 2025 seeking ideas on novel offshore compute projects, and Massachusetts, New York, and California have all added offshore digital infrastructure to their coastal development planning categories.
• Japan: METI released a feasibility study in early 2025 identifying submarine data infrastructure as a realistic way to expand compute capacity within Japan's exclusive economic zone without needing additional land.
• Singapore: The country's long-running data center moratorium created real pressure to find alternatives, and its IMDA is now developing a regulatory sandbox for underwater data center licensing, following a feasibility study that found waters below the tropical thermocline, roughly 200 to 400 meters deep, offer a workable solution to the limits of using warm tropical surface water for cooling.

The Economics Behind This Shift

Our analysis at CMI Consulting LLC puts the global data center infrastructure market at roughly $77.21 billion in 2025, with growth expected to push it toward $264.97 billion by 2035, a compound annual growth rate of about 13.12% across that decade.

Within that broader growth, we estimate underwater data center capacity could pass 500 MW by 2028 and reach somewhere between 3,000 and 5,000 MW by 2032, as manufacturing becomes more modular and construction costs drop by 40% to 50% compared with current levels. As the technology matures, wind-integrated underwater facilities are likely to reach deployment costs of $15 to $20 million per MW, putting them roughly in line with high-end land-based construction in coastal markets already squeezed by land and water constraints.

Underwater facilities still carry a capital cost premium of around 70% to 95% compared with standard land-based builds. But looked at over a ten-year total cost of ownership, the savings on land and water start to close that gap in markets facing real water stress and high land costs. Reduced land acquisition costs (75% to 90% lower), lower cooling-related power costs (65% to 75% lower), zero water expenses, and smaller carbon and ESG compliance costs together tend to offset that upfront construction premium.

Where This Is Likely Headed

The most probable outcome, with roughly a 55% likelihood by our estimate, is a niche but meaningful rollout across 8 to 12 specific markets where land and water constraints, offshore wind resources, and favorable regulation all line up. In that scenario, underwater data centers could account for 3% to 5% of new data center investment by 2030, becoming a standard offering among leading offshore developers, some of whom will likely acquire data center operators outright or form joint ventures with them. Hyperscalers, meanwhile, are expected to use underwater facilities as dedicated capacity in locations where land-based development simply isn't an option.

The companies that end up shaping this industry over the next decade will be the ones that can operate comfortably across three very different disciplines at once: digital infrastructure, offshore energy, and marine engineering. That's a rare combination of skills, and the organizations that build it early are likely to hold a meaningful first-mover advantage in a space where the stakes and the barriers to entry are both unusually high.