China switches on underwater data center off Shanghai powered by offshore wind

Shanghai brings computing to the seabed

China has switched on the first phase of an underwater data center off Shanghai, pairing electricity from offshore wind turbines with cooling from the sea. Built by HiCloud near the Lingang Special Area, the installation begins with a demonstration of 2.3 megawatts and is planned to expand to 24 megawatts.

The project represents an investment of about 1.6 billion yuan, roughly $226 million. More than 95 percent of its electricity comes from a nearby offshore wind farm, while seawater removes heat from sealed cabins containing computing equipment. The initial array can accommodate 198 server racks, with additional underwater modules planned as the project grows.

HiCloud is targeting power usage effectiveness, or PUE, of 1.15 or lower. This measure compares all the electricity a data center uses with the electricity delivered to its computing equipment. A lower figure means less power is spent on cooling and other support systems.

Designers estimate that the installation will use about 22.8 percent less electricity than an equivalent facility on land. They also expect a reduction of more than 90 percent in land use and no freshwater consumption for cooling, advantages in a densely developed coastal city where space and water are constrained.

As of October 29, 2025, the Shanghai project combines an operating demonstration with much larger expansion plans. HiCloud and its partners have announced an ambition for 500 megawatts of underwater capacity powered by offshore wind. Moving from the first installation to that scale will put the design, maintenance arrangements and economics to a much broader test.

How the underwater installation works

The data center sits offshore near the Lingang Special Area of the China (Shanghai) Pilot Free Trade Zone. Fiber connections carry information between the submerged modules and users on land, while power cables connect the installation to the offshore wind farm.

The modules are anchored to the seabed to limit movement during storms. Servers remain inside sealed cabins, and seawater passes across heat exchangers attached to those cabins. A heat exchanger transfers warmth from one part of a system to another, allowing the surrounding water to absorb heat without exposing the electronics to the sea.

Seawater can absorb substantial amounts of heat, and temperatures near the seabed are relatively stable. Those qualities make the ocean useful for carrying away the heat produced by servers. The warmed water then disperses into the surrounding water column.

Conventional facilities depend on equipment such as large fans, chillers and cooling towers to keep servers within their operating temperatures. The Lingang design avoids much of that machinery by making the sea the main destination for waste heat.

The sealed environment also keeps out dust and limits changes in humidity, temperature and vibration caused by people or nearby equipment. Earlier underwater trials by other companies reported longer hardware life and fewer failures under such stable conditions. For the Shanghai installation, reliable performance will depend on both the protected interior and the equipment that connects it to shore.

What the efficiency targets mean

Traditional data centers can spend 40 to 50 percent of their electricity budget on chillers and cooling towers. The underwater design aims to bring cooling energy below 10 percent of total consumption, contributing to the estimated 22.8 percent reduction in total power demand compared with an equivalent facility on land.

PUE offers another way to describe efficiency. Operators calculate it by dividing total facility electricity use by the electricity used by computing equipment. A theoretical PUE of 1.0 would mean every watt goes to computing, with none needed for cooling or other overhead.

A target of 1.15 means that, for every unit of electricity delivered to computing equipment, the entire facility would use 1.15 units. The additional 0.15 covers cooling and other support systems. Cooling energy and total overhead are related, but they measure different parts of the electricity budget.

China's national policy requires new data centers to operate below a PUE of 1.25 by the end of 2025. The Shanghai installation's target of no more than 1.15 is tighter than that requirement and comparable with the best facilities on land using advanced cooling and energy management. Many older sites still operate above 1.4.

Electricity efficiency and electricity supply address different costs. Ocean cooling reduces the energy needed to run the facility, while the wind farm supplies more than 95 percent of its power. The demonstration will provide operating data on how those two elements work together.

Why Shanghai wants more computing capacity

Artificial intelligence, video services, electronic commerce and connected industry are increasing demand for computing. Modern AI systems require large groups of graphics and tensor processors for training and operation. These processors perform intensive calculations, consume large amounts of electricity and release considerable heat.

Shanghai's plans call for a major increase in smart computing capacity and industry output. Its targets include 200 exaFLOPS of computing capacity and an intelligent cloud industry worth hundreds of billions of yuan by 2027. FLOPS measures how many mathematical operations a computing system can perform per second; the prefix exa describes a scale of a billion billion.

Local authorities also face tighter limits on land, water use and emissions. Placing computing equipment offshore addresses those constraints by reducing the land footprint and removing the need for freshwater cooling. Developers cite more than 3,000 hours of annual utilization for the East China Sea's offshore wind resources, alongside the seabed's stable conditions for cooling.

Locating computing capacity near coastal users can reduce latency, the delay between sending information and receiving a response. Using nearby renewable electricity also reduces losses associated with transmitting power over distance. The aim is to place computing and energy supply close to where they are needed.

Shanghai's offshore approach complements China's Eastern Data, Western Computing program, which expands capacity and balances where computing takes place across the country. Some workloads can go to large campuses in western China, while capacity near the coast serves areas where substantial amounts of data are generated. The underwater project adds another location option within that wider strategy.

From Hainan trials to a larger Shanghai cluster

HiCloud's Shanghai installation builds on several years of trials in China. Off Hainan Island, the company has already deployed a commercial underwater facility about 35 meters, or 114 feet, below the surface.

The Hainan installation uses pods designed to withstand pressure, each holding 400 to 500 servers. The pods connect to one another and to the mainland through a subsea cable. Their modular design provides a stepping stone toward a larger cluster rather than requiring all computing equipment to be housed in a single structure.

At Shanghai, the immediate expansion plan is to move from the 2.3 megawatt demonstration to 24 megawatts. That would be more than ten times the initial capacity. The larger ambition of 500 megawatts would be more than twenty times the planned 24 megawatt installation, requiring a much bigger system for supplying power, connecting equipment and handling maintenance.

HiCloud and partners including Shenergy Group, the Shanghai branch of China Telecom, INESA and CCCC Third Harbor Engineering have signed an agreement to pursue that 500 megawatt cluster. The partnership brings together computing, telecommunications, energy and marine construction interests.

The sequence shows how the project has developed and where its stated goals lead:

• Several years of Chinese trials preceded the Shanghai installation, including HiCloud's commercial deployment off Hainan.
• By October 29, 2025, the first Shanghai phase had been activated as a demonstration of 2.3 megawatts.
• China's policy deadline for new data centers to operate below a PUE of 1.25 is the end of 2025.
• Shanghai's computing and intelligent cloud industry targets extend to 2027, alongside plans for 24 megawatts at the underwater site and an ambition for a 500 megawatt cluster.

No completion dates were given for the expansion to 24 megawatts or the proposed 500 megawatt cluster. Those plans require additional modules, more turbines and stronger connections to the electricity grid.

Cooling savings bring marine responsibilities

The ocean receives the heat removed from the servers. That creates a thermal plume, an area of warmer water around the installation. Warmer water holds less dissolved oxygen and can stress marine life, especially during summer and marine heat waves.

Developers say testing during earlier pilot deployments found a temperature rise of less than one degree Celsius in nearby water. Continuous monitoring becomes more important as capacity increases, because a larger installation must dispose of more heat.

Potential safeguards include diffusers that spread the discharge, flow controls that reduce operation during sensitive periods, and siting away from vulnerable habitats. These measures would help manage how much heat reaches particular areas and when it arrives.

Marine effects extend beyond temperature. Engineers must consider underwater noise, electromagnetic fields around power and data cables, and disturbance from construction or maintenance to benthic ecosystems, the communities living on or near the seabed.

Marine growth on the outside of the pods also needs attention. Strategies to prevent organisms from attaching should avoid toxic coatings, while corrosion protection must be designed to avoid contaminating the water. Equipment that performs well inside a sealed cabin still needs an exterior capable of enduring the sea.

Regulators will likely require detailed environmental impact assessments, surveys of conditions before development and continued public reporting on water temperature, noise and nearby marine life. Careful oversight could help balance electricity and freshwater savings with protection of the local environment. The land and water benefits need to be considered alongside the effects of releasing heat offshore.

Keeping equipment secure and serviceable

Underwater servers face physical hazards that differ from those in a building. Anchors, fishing gear used for trawling, storms and deliberate interference can threaten cables or pods. Damage to a connection can affect service even when the servers themselves remain protected.

Possible safeguards include exclusion zones, burying cables in seabed trenches, protective cable armor, sonar surveillance and alerts for unusual activity near the installation. Researchers have also warned that interference using sound could affect sensitive equipment in rare circumstances, making acoustic monitoring and robust shielding relevant to the design.

The sealed interior offers a different reliability advantage. Keeping equipment free of dust and at steady temperature and humidity can reduce corrosion and mechanical stress. Previous underwater trials abroad recorded much lower hardware failure rates than typical facilities on land.

Stable operating conditions also help power distribution equipment, storage systems and connections between servers stay within tight tolerances. Those connections are especially important for AI training clusters, where many processors need to exchange information quickly while working on the same task.

Maintenance cannot follow the familiar routine of an engineer entering a server hall and replacing a component. Underwater equipment needs a modular service plan. Operators may replace an entire pod or bring a sealed module to the surface during a planned service window and slide out the equipment requiring attention.

Marine growth, corrosion and the condition of seals will help determine inspection schedules. Quick retrieval and redeployment, clear interfaces between components and standardized parts will be central to making the installation practical. The demonstration will test service procedures as well as computing performance.

Can lower operating costs justify construction at sea?

The investment of about $226 million covers more than server equipment. The build includes subsea cabins, marine construction, specialized power and cooling systems, and integration with the offshore wind farm. Those requirements make the initial capital commitment substantial.

Electricity is the largest expense in a modern data center. Operating costs could be attractive if the site maintains a PUE near 1.15 and obtains power directly from local turbines. Less overhead means more of the electricity purchased is used for computing rather than keeping the equipment cool.

Shanghai's shortage of available land and freshwater also affects the calculation. Avoiding a large campus on land can save spending on real estate, cooling towers and water rights. The reduction in land use of more than 90 percent gives the project a potential advantage beyond its electricity bill.

Offshore construction introduces other costs. Poor weather can delay work and make schedules uncertain. Insurance, vessels and specialized crews add expense, while servicing submerged equipment requires marine operations that a conventional facility would not need.

Experience and standardized designs can reduce some of those premiums. China's established offshore wind supply chain may help with prices and schedules. Reliable operation and manageable maintenance at Lingang would improve the case for adding more underwater capacity.

Large telecommunications and energy partners can also help fill racks with paying customers and coordinate the development with electricity grid planning. The financial case depends on construction, occupancy, electricity use and service costs working together, rather than cooling savings alone.

What earlier underwater projects teach

Microsoft began testing underwater data centers about a decade ago through Project Natick. Its first trial used a sealed steel capsule off the Pacific coast. A later deployment placed a unit measuring 12 meters, containing 864 servers, off the Orkney Islands.

Microsoft reported very low hardware failure rates. The stable underwater environment and an interior filled with nitrogen were considered likely contributors. The company ended the research program in 2024 and said it would not commercialize the concept.

Shanghai is pursuing a different route. Its installation is connected to an offshore wind farm, targets tighter PUE performance and forms part of city and national plans to expand computing capacity. The proposed progression from a demonstration to a much larger cluster puts commercial operation at the center of the effort.

Other companies and countries are also examining underwater or ocean cooling designs. Startups have proposed small subsea modules near shore, while research groups in South Korea, Japan and Singapore have explored marine cooling and possible locations.

Facilities on land continue to improve through liquid cooling using warm water, reuse of waste heat in district energy networks and chip designs that reduce electricity demand. The underwater model will need to show that its benefits exceed those of the best alternatives on land when construction, energy, maintenance and marine risks are counted together.

The demonstration's next tests

Shanghai's first phase will provide operating data on PUE, hardware reliability and service procedures at sea. Independent checks of the electricity balance and environmental performance will matter for public confidence as the project expands.

Monitoring temperature around heat discharge points, recording underwater noise and reporting observations of marine life can show how the installation affects its surroundings. Regulators could make those disclosures conditions of permits and renewals.

The move from 2.3 to 24 megawatts is the first major test of whether installation and maintenance can grow smoothly. Progress toward the proposed 500 megawatt cluster would require standard pod designs, clear maintenance procedures and supply chains capable of delivering modules quickly, alongside the additional turbines and grid connections.

Reliable results in Shanghai could encourage other coastal regions with strong offshore wind resources and limited land to consider similar facilities. For now, the operating demonstration is the point where ambitious estimates meet the daily demands of keeping servers powered, cooled, connected and accessible for repair.

Key Points

• HiCloud has activated a demonstration of 2.3 megawatts off Shanghai, with a planned expansion to 24 megawatts.
• The project represents about 1.6 billion yuan, roughly $226 million, in investment; its initial array can hold 198 server racks.
• More than 95 percent of electricity comes from offshore wind, and seawater cools sealed server cabins.
• Operators target PUE of 1.15 or lower, compared with China's requirement of below 1.25 for new data centers by the end of 2025.
• Designers estimate 22.8 percent lower electricity use, cooling energy below 10 percent, more than 90 percent less land use and no freshwater cooling.
• HiCloud's earlier Hainan deployment sits about 35 meters underwater, with 400 to 500 servers per pod.
• HiCloud and its partners have announced an ambition for a cluster of 500 megawatts powered by offshore wind.
• Expansion will test maintenance, costs, security and the effects of heat, noise and cables on marine life.