What happened
Dialogue Earth reported that Chinese company HiCloud is operating underwater data centers powered partly by offshore wind, using seawater for cooling. The report also describes environmental concerns involving heated discharge, anti-fouling coatings, monitoring, maintenance and possible effects on marine ecosystems.
Dialogue Earth, in the report republished by Reccessary, said a HiCloud facility near Shanghai’s Lingang area places the core data center about 10 meters underwater and connects it to nearby offshore wind turbines. The report said the company claims seawater cooling saves 18% to 28% of power use for the same server load, while the Lingang project is described as costing CNY 1.6 billion and using no freshwater for cooling. HiCloud also claims the facility receives 95% green electricity. These company and expert figures were not independently confirmed in the supplied source.
The report said HiCloud’s earlier Hainan project placed servers in three sealed seabed capsules in 2023. It described company plans for up to 100 capsules, but the source does not independently establish the full expansion, its location, its financing or its commercial availability. The report also linked the project to China Telecom and Alibaba Cloud through cited reporting, without establishing the terms of those relationships.
Dialogue Earth reported that the Lingang system discharges seawater after it has absorbed heat. Its assessment document describes an output of 150 liters per second at roughly 5 degrees Celsius above intake , while HiCloud’s website says a trial found a temperature increase of less than 1 degree Celsius within two meters. The report said the timing and methodology of that trial were unclear. Experts quoted by Dialogue Earth warned that organisms near intake and discharge points could be affected, and raised concerns about anti-fouling coatings and insufficient study of long-term operation and decommissioning.
The report said HiCloud had not responded to a request for operational data at publication. It also reported that replacing AI chips underwater could be expensive and that typhoons or cable damage could create longer outages than comparable land facilities. No customer access terms, service availability, pricing or workload details are documented in the supplied source.
Source details: reccessary.com ↗
Why it matters
The report presents underwater AI infrastructure as a potentially consequential response to the electricity and freshwater demands of expanding AI computing, while emphasizing that its environmental benefits remain unproven. If the reported cooling savings hold at scale, the approach could reduce some land-based data-center resource demands. But local ecological effects, lifecycle costs, resilience and transparency remain meaningful unknowns, so the technology should not yet be treated as demonstrably green.
Dialogue Earth cited 2025 research from the Beijing Institute of Technology projecting that data centers could account for 4.8% of China’s electricity use by 2030. In that context, reducing cooling demand and freshwater consumption could have practical value for operators building AI capacity. However, the report’s central energy and environmental figures are attributed to HiCloud, cited experts or project documents rather than independently validated measurements.
The reported design shifts part of the environmental burden from land and freshwater use to marine heat discharge, protective coatings, seabed construction and specialized maintenance. That does not establish that underwater facilities have a better overall environmental footprint. It means the relevant comparison requires lifecycle data covering construction, electricity supply, cooling, hardware replacement, emergency repairs and decommissioning.
The technology could matter to infrastructure planners if it reliably uses otherwise-curtailed offshore wind power and supports in areas where land, grid capacity or freshwater are constrained. The source does not show that these benefits have been demonstrated across multiple commercial facilities, nor that the economics work for customers at scale.
Interactive Mechanism: How It Actually Works
Explore the underlying technology behind this development interactively.
A route planner searches possible journeys using explicit rules. What does this illustrate about AI?
What to watch next
Watch for independently collected environmental data, operational energy and water figures, evidence about uptime and maintenance costs, and regulatory scrutiny of expansion. The source does not establish public availability, customer access, pricing or the actual AI workloads running in the facilities.
Independent monitoring should clarify the size and duration of thermal plumes, effects on local species, sediment conditions and possible pollution from anti-fouling or anti-corrosion materials. The source reports expert calls for more operational and decommissioning data, while the project assessment focuses on limited effects.
Future disclosures should separate direct electricity use, cooling savings and the claimed share of green electricity. They should also document hardware replacement intervals, repair times, cable resilience, storm performance and total operating costs. The report says these issues could make underwater facilities more expensive and slower to restore than land-based centers.
Expansion plans in Hainan and interest from other countries will indicate whether the model is moving beyond pilots. Public evidence is still missing on who can use the facilities, whether they serve external customers, what workloads they run, and what prices or access conditions apply.
The main unresolved question is whether underwater AI data centers deliver a lower total environmental impact rather than simply relocating some impacts offshore. That conclusion requires transparent, independently assessed results from operating sites.