communityfix.org

Lingang, Shanghai, China (HiCloud offshore underwater data centre)

#00220

OngoingCity

Case study of

#00212 Relocate compute into submerged or floating offshore modules cooled by seawater

Implementer

HiCloud Technology, with Shenergy Group, Shanghai Telecom and CCCC Third Harbor Engineering

Timeline

Since Oct 1, 2025

Location

Lingang, Shanghai, China (HiCloud offshore underwater data centre)30.9000, 121.9200

Description

The only commercial-scale underwater data centre in operation as of 2026. Servers sit in sealed watertight units offshore from Lingang, positioned between phases of an existing offshore wind farm; construction completed October 2025 and full commercial operations reported from May 2026.

The decisive design choice is siting inside the wind farm rather than immersion itself. The reported PUE of 1.15 and the claimed 22.8% electricity reduction against a conventional facility derive from removing the chiller plant; the claimed 100% wind supply comes from co-location. A land-based facility inside the same wind farm would capture the larger share of the benefit.

Phase 1 is 2.3 MW and the full build is 24 MW (~2,000 servers). At $226 million for 24 MW this is not obviously cheap relative to terrestrial hyperscale.

An earlier Chinese deployment, Highlander's shallow-water unit off Lingshui, Hainan (completed December 2022), reported an 80% cost reduction against deep-water designs and named China Telecom and SenseTime as customers.

No measured environmental monitoring data (thermal plume extent, benthic effects, biofouling rates) has been published for either site.

Metrics

6
Phase 1 capacity2.3MW
Full build capacity24MW
Reported PUE1.15ratio electricity reduction vs conventional facility (no chiller plant)
Claimed electricity reduction versus conventional22.8%
Freshwater consumption0litres/year (seawater cooling, no evaporative draw)
Claimed land use reduction>90%

Funding

$226,000,000 · Consortium of Chinese state-linked enterprises

Lessons learned

  • Co-location with offshore wind generation, not immersion itself, produces most of the measured efficiency benefit. The transferable lesson is about siting next to the power source, which works equally well on land.
  • At 24 MW and $226 million, this demonstrates technical feasibility rather than economic competitiveness with terrestrial hyperscale; feasibility at this scale should not be extrapolated to gigawatt-class AI demand.
  • Shallow-water designs are reported as substantially cheaper than deep-water ones, which identifies where engineering effort belongs if this approach is pursued elsewhere.
  • Two years into commercial operation there is still no published thermal plume or benthic monitoring data. Environmental consenting regimes in other jurisdictions should require this data as a permit condition, since it does not appear to be produced voluntarily.

Documented Jul 28, 2026

Author AvatarGerard Antoun

communityfix.org