HiCloud Subsea Data Center Hits Full Operation Off Shanghai
Key insights
- HiCloud's facility uses zero freshwater and zero grid power, drawing 95% of its 24MW capacity from co-located offshore wind.
- Phase 1 operates at 2.3MW with a 10x scaling path to 24MW already engineered into the existing subsea structure.
- Seawater passive cooling cuts electricity consumption 22.8% versus land-based equivalents, according to HiCloud's operational data.
Why this matters
Freshwater access and grid capacity are now documented bottlenecks on datacenter permitting in the US, Europe, and China, so a validated seawater-cooled, wind-powered alternative changes the site-selection calculus for any operator planning capacity in coastal regions. The 22.8% efficiency gain is concrete enough to pressure land-based operators to justify their cooling overhead, particularly as power purchase agreements tighten. China demonstrating a working 24MW subsea blueprint before Western hyperscalers gives Chinese AI infrastructure firms a replicable export architecture for Southeast Asian and Middle Eastern coastal markets where land and freshwater constraints are acute.
Summary
HiCloud's $226M underwater data center off Shanghai's Lingang Special Area is now fully operational, making it the first facility of its kind to combine offshore wind power with subsea deployment at commercial scale.
The installation sits 35 meters below the ocean surface, housing 2,000 GPU servers in a sealed pressure vessel that uses ambient seawater for passive cooling, eliminating freshwater consumption entirely. A co-located offshore wind farm supplies 95% of the facility's power, with the architecture delivering 22.8% lower electricity consumption than comparable land-based setups. Phase 1 runs at 2.3MW demonstration capacity, with a ceiling of 24MW as deployment scales.
Essentially: HiCloud is betting that ocean-floor infrastructure solves the two hardest constraints in AI buildout -- land scarcity and cooling water.
- 2,000 GPU servers support LLM training and big-data annotation workloads at current Phase 1 capacity
- No grid connection and no freshwater use are the headline engineering claims separating this from land-based alternatives
- Full 24MW ceiling represents roughly 10x the current operational capacity, with scaling path already built into the design
If the architecture replicates, coastal nations with offshore wind resources gain a new template for building AI compute without the land, water, or grid constraints that are bottlenecking datacenter expansion inland.
Potential risks and opportunities
Risks
- Subsea hardware failures at 35m depth could strand GPU capacity for weeks if retrieval logistics are not yet mature, creating SLA exposure for LLM training customers relying on the facility.
- Offshore wind intermittency without disclosed battery backup means the 5% non-wind power source and grid-independence claim warrants scrutiny -- any gap could trigger compliance issues under Chinese green datacenter certification standards.
- Competing hyperscalers (Microsoft, Google, Amazon) that have invested heavily in land-based liquid cooling infrastructure face narrative pressure to justify freshwater usage publicly if HiCloud's zero-freshwater claim scales and attracts regulatory attention in drought-affected regions.
Opportunities
- Subsea pressure vessel and marine engineering firms (Saipem, TechnipFMC, CNOOC Engineering) are positioned to win fabrication contracts if HiCloud or copycat operators expand to additional coastal sites in Asia or the Middle East.
- Offshore wind developers (Orsted, Vestas, Ming Yang Smart Energy) gain a new anchor-tenant model where AI datacenters co-locate directly with wind assets, improving project economics and capacity factor utilization.
- Coastal sovereign wealth funds and port authorities in Singapore, UAE, and South Korea could fast-track subsea datacenter concessions using HiCloud's operational blueprint as a proof point, opening a new infrastructure asset class for institutional investors.
What we don't know yet
- Hardware maintenance and server replacement procedures at 35 meters depth have not been disclosed -- turnaround time and cost per intervention are unknown.
- Whether the 22.8% efficiency figure is independently audited or derived from HiCloud's own operational telemetry remains unconfirmed as of May 2026.
- Regulatory status of scaling from 2.3MW Phase 1 to the full 24MW ceiling -- including Chinese maritime and environmental approvals -- has not been reported.
Originally reported by tomshardware.com
Read the original article →Original headline: China's World's First Offshore Wind-Powered Underwater Data Center Enters Full Operation — 24MW, 2,000 GPU Servers, Seawater Cooling Off Shanghai Coast