China's 24 MW underwater AI data center off Shanghai's coast is now fully operational, housing 2,000 servers. The facility uses offshore wind power and passive seawater cooling to reduce energy consumption by an estimated 30% versus land-based alternatives.
Key facts
- 24 MW power capacity for the underwater data center.
- Houses 2,000 servers in a subsea facility.
- Operates at 35 meters depth off Shanghai.
- Claims 30% energy reduction vs land-based data centers.
- Uses dedicated offshore wind power for operations.
China activated the 'world's first' offshore wind-powered underwater data center for full commercial operation, per state media reports. The 24 MW facility sits at 35 meters depth off Shanghai, hosting 2,000 servers cooled passively by seawater and powered by a dedicated offshore wind farm.
The operator claims the subsea deployment cuts energy usage by ~30% compared to conventional land-based data centers, primarily by eliminating mechanical cooling systems. [per Xinhua and Hailan Cloud, May 2026] The project targets low-latency AI inference workloads for coastal regions, though specific latency figures were not provided.
Infrastructure details and unknowns
The facility's total capital expenditure, server configurations, and GPU types remain undisclosed. The 24 MW power draw suggests capacity for roughly 2,000-4,000 high-end AI accelerators (e.g., NVIDIA H100 at 700W each), depending on utilization. [per Uptime Institute 2024 PUE benchmarks] The offshore wind connection implies intermittent power supply, though the operator did not detail battery backup or grid interconnects for continuity.
Strategic context
This deployment fits China's pattern of coupling renewable energy with AI compute — a strategy distinct from the U.S. approach of scaling terrestrial data centers with grid power. The underwater model also avoids land-use conflicts and proximity to population centers, potentially easing permitting. However, maintenance access at 35 meters depth introduces operational complexity not present in land-based facilities.
The unique take
What the AP wire would miss: this is less a breakthrough in underwater computing and more a test case for whether renewable-powered edge AI can compete on reliability with terrestrial hyperscale centers. The 30% energy savings claim is modest — hyperscale operators like Google already achieve <1.10 PUE (power usage effectiveness) on land with advanced cooling. The real question is whether the subsea location's latency advantage for coastal AI workloads offsets the maintenance overhead and power intermittency.
What to watch
Watch for the operator to release latency benchmarks comparing the subsea facility to Shanghai's terrestrial data centers, and for any disclosure of GPU types or utilization rates. Also track whether China announces additional underwater deployments — a second facility would signal scaling intent beyond this pilot.
