First Offshore Wind-Powered Underwater Data Center Deployed Off China's Coast

China’s HiCloud Technology has brought the world’s first offshore wind-powered underwater data center to full commercial operation off Shanghai’s Lingang coast as of May 2026.¹

The $226M facility houses approximately 2,000 servers in pressure-resistant sealed capsule modules submerged 35 meters below the surface, with a total capacity of 24 MW and a Power Usage Effectiveness (PUE) of approximately 1.15.² For context, traditional land-based data centers average 1.5 or higher.

Cooling is entirely passive and requires zero additional power. Hot air from servers converts refrigerant to gas, which rises by buoyancy to an upper cooling layer where seawater absorbs the heat, then gravity returns the liquid back to the server room. No chillers. No HVAC. No freshwater.³ Claimed efficiency gains: 22.8% less electricity consumed, 90%+ reduction in land use, and zero freshwater use.⁴

Power comes from a nearby offshore wind farm supplying approximately 97% of the facility’s energy. The data center sits physically between the wind farm’s first and second phases.⁵ China Telecom has deployed GPU clusters inside the modules for AI workloads, big data annotation, and 5G services.⁶

This isn’t China’s first rodeo. HiCloud’s predecessor facility off Hainan Island has been commercially operational since 2023 – with a 1,300-ton sealed pressure vessel on the seabed serving clients including China Telecom, Tencent, and SenseTime.⁷ The Shanghai facility is the scaled-up, wind-integrated successor. HiCloud’s long-term ambition: 500 MW of underwater capacity.⁸

For comparison, Microsoft’s Project Natick proved the concept worked but Microsoft confirmed in June 2024 it would not build more underwater data centers globally.⁹ China took the baton and ran.

The big questions going forward: long-term reliability in a marine environment, maintenance costs and module accessibility, whether the efficiency claims hold at scale, and whether the West reconsiders underwater deployment as AI energy demands keep climbing.