China has opened the world's first commercial undersea data center designed specifically for artificial intelligence workloads, using the ocean itself as a cooling system and offshore wind turbines as its primary power source. The Shanghai Lingang project, which entered operation in May 2026, demonstrates a radically different approach to solving one of tech's most pressing problems: how to power and cool the next generation of AI infrastructure without overwhelming cities and depleting freshwater supplies.

The facility sits about 6 miles offshore in Shanghai's Lingang area and is built by a subsidiary of China Communications Construction. Its planned capacity is 24 megawatts, enough to power roughly 20,000 households, with the first phase operating at 2.3 megawatts and a second phase intended to scale to full capacity.

Why Is AI Pushing Data Centers to the Ocean?

The urgency behind underwater data centers reflects a dramatic shift in computing's energy footprint. Global electricity demand from data centers grew 17% in 2025, but demand from AI-focused data centers climbed 50% that same year. The International Energy Agency expects total data center electricity consumption to roughly double from 485 terawatt-hours in 2025 to 950 terawatt-hours by 2030.

In the United States, the strain is already visible. American data centers consumed 183 terawatt-hours of electricity last year, more than the entire state of Ohio uses in 12 months. The North American Electric Reliability Corporation warns that much of North America is at risk of supply shortages, with the PJM grid, which covers 13 states and Washington, D.C., facing the most exposure. Forecasted data center expansions have pushed its capacity bill to $16 billion.

Cooling is the hidden culprit. In conventional data centers, cooling accounts for about one-third of total electricity use. Tsinghua University Professor Li Zhen noted that an undersea data center of the same scale could bring that share down to about one-tenth.

How Does the Shanghai Undersea Model Work?

The Shanghai Lingang project uses what developers call a "direct offshore wind connection." Electricity from nearby offshore wind farms is sent through subsea cables directly to the submerged data modules, bypassing the usual grid-routing path and bringing power generation much closer to computing demand.

The cooling system is equally innovative. Instead of relying on energy-intensive air conditioning or freshwater cooling loops, the facility uses seawater as a natural cooling source through a circulating copper-pipe heat exchange design. According to project estimates, this setup reduces electricity consumption by 22.8%, eliminates freshwater use entirely, and cuts land use by more than 90% compared with traditional facilities.

To understand the potential scale of this approach, consider China's broader data center footprint. Li estimated that China's data centers consume about 250 billion kilowatt-hours of electricity each year, with roughly 80 billion kilowatt-hours going to environmental cooling. If similarly scaled data centers were placed underwater, cooling demand could fall to around 30 billion kilowatt-hours. That would save about 50 billion kilowatt-hours of electricity every year, equivalent to avoiding the burning of about 16.5 million tons of standard coal annually.

What Are the Real-World Challenges?

The ocean is not an empty parking lot. Saltwater corrosion, subsea cable maintenance, repair access, storm exposure, and continuous heat release into local marine environments all present significant operational challenges. A data hall on land can be entered by technicians with a badge and a toolbox. Underwater, even routine repairs become a marine engineering job.

Similar underwater data center proposals have already drawn regulatory scrutiny elsewhere. In San Francisco Bay, a proposed underwater server project faced questions over permits and possible ecological effects, including concerns about heat and disturbance in sensitive waters.

China is not starting from zero. Microsoft's Project Natick tested the feasibility of subsea data centers years earlier, including a two-year deployment off Scotland's Orkney Islands. Microsoft described the project as an effort to understand the benefits and difficulties of deploying subsea data centers powered by offshore renewable energy. The key difference is scale and intent: Natick was a research program, while the Shanghai project is being presented as commercial infrastructure tied to AI workloads, offshore wind, and regional computing demand.

How Are Tech Giants Addressing the Power Crisis on Land?

While China experiments with underwater solutions, Silicon Valley is turning to nuclear power as a more immediate answer to AI's energy demands. Microsoft and OpenAI have struck deals to open power centers in Ohio, with Microsoft also securing a footprint in Illinois. Most notably, Microsoft inked a deal with Constellation Energy, the nation's largest nuclear power provider, committing more than $1.5 billion to restart Pennsylvania's Three Mile Island nuclear plant, supported by a $1 billion federal loan from the U.S. Department of Energy.

Meta signed a 20-year power purchase agreement with Constellation Energy, while Google signed a corporate agreement with Kairos Power to purchase energy from a fleet of advanced small modular reactors (SMRs), adding up to 500 megawatts of clean power to the grid. Amazon expanded its partnership agreement with Talen through 2042, part of its larger $500 million push into nuclear technologies, partnering with Dominion Energy and Energy Northwest to build SMRs near its data center hubs in Virginia and Washington.

The U.S. government is backing nuclear energy development across small modular reactors, which offer compact, factory-built reactors that are faster and cheaper to deploy than traditional plants. The U.S. Department of Energy selected the Tennessee Valley Authority and Holtec as the first recipients of $400 million each in funding to advance small modular reactor deployments.

Steps to Understanding AI's Energy Solutions

• Offshore Wind and Seawater Cooling: Undersea data centers use direct connections to offshore wind farms and natural seawater circulation to reduce electricity consumption by roughly 23% and eliminate freshwater use entirely, representing a model for coastal regions with heavy AI demand.
• Nuclear Power Partnerships: Major tech companies including Microsoft, Meta, Google, and Amazon are signing long-term agreements with nuclear providers to secure steady, carbon-free electricity, with commitments ranging from $1.5 billion to $500 million across multiple projects.
• Small Modular Reactors: The U.S. government is funding advanced SMR deployments through the Department of Energy, offering a faster and cheaper alternative to traditional nuclear plants, though experts estimate at least another decade before they can support large-scale data center consumption.

However, not every nuclear project is moving smoothly. Wonder Valley, an off-grid 7.5-gigawatt, 58-building data center campus in northwestern Alberta, is facing multiple challenges, including no finalized land purchase and no new provincial permits. Its sister site, Wonder Valley Utah, is also receiving community pushback.

New nuclear plants must secure licensing and permits, scale supply chains, and prove reliable operations before they become a viable solution. These requirements are unlikely to be met in the short term. Georgia Power took 15 years to open an expansion of its Plant Votgle nuclear facility, with a price tag of $36.8 billion, more than double the original $14 billion estimate.

"Meeting the AI infrastructure power demand will require data center operators to shift development approaches, co-developing their projects with the new on-grid wind, solar, and natural gas generation and battery storage required to power the facility," said Jay Dietrich, research director of sustainability and energy at Uptime Institute.

Jay Dietrich, Research Director of Sustainability and Energy at Uptime Institute

Dietrich cited natural gas generation as a short-term option that is widely deployable and available now. For geothermal energy and SMRs, it will be at least another decade before they are ready to support data center energy consumption.

Venture capital firms including DCVC (Data Collective), Founders Fund, and Andreessen Horowitz have invested millions into nuclear fission and fusion startups. DCVC is heavily invested in Oklo, a startup with OpenAI's CEO Sam Altman as chairman, which is developing fast neutron reactors fueled by HALEU (high-assay, low-enriched uranium) that can pull power from recycled nuclear waste. The U.S. Department of Energy estimates that the country has accumulated over 90,000 metric tons of spent nuclear fuel from commercial nuclear power plants as of early 2026.

For now, the Shanghai project represents a live test of whether AI infrastructure can move into new environments without simply moving its problems with it. Lin-gang officials have already signaled larger ambitions, with a strategic cooperation agreement signed in 2025 for a 500-megawatt offshore underwater data center project, suggesting developers see this as a platform, not a one-off experiment. If the model works, the next data center boom may look less like a warehouse park and more like a piece of marine energy infrastructure.