Overview
In a groundbreaking move, China has commenced full operations of a 24-megawatt underwater artificial intelligence data center off the coast of Shanghai. This facility, housing 2,000 servers, harnesses offshore wind energy and passive seawater cooling to dramatically reduce energy consumption and carbon footprint. This guide provides a comprehensive technical walkthrough for engineers and project managers interested in replicating such a system, covering design principles, deployment steps, and common pitfalls.
This offshore subsea AI data center represents a fusion of renewable energy and advanced cooling technology. By placing servers underwater, it leverages the ocean's natural cooling capacity, eliminating the need for energy-intensive air conditioning. Additionally, it is directly powered by an offshore wind farm, ensuring a continuous supply of clean electricity. The facility is designed for high-performance AI workloads, offering low latency and high reliability.
Prerequisites
Before embarking on a similar project, ensure you have the following resources, knowledge, and approvals:
- Site Assessment Report: Detailed oceanographic data including water depth, current speed, temperature profiles, and seabed geology.
- Regulatory Permits: Environmental impact assessment clearance, maritime construction permits, and communication spectrum licenses.
- Partnerships: Collaborations with offshore wind farm operators, subsea cable providers, and marine engineering firms.
- Technical Expertise: In subsea enclosure design, high-voltage DC power distribution, passive cooling thermodynamics, and AI server management.
- Hardware Components: Pressure-resistant server pods, corrosion-resistant materials, submarine power cables, and remote monitoring sensors.
Step-by-Step Implementation
1. Site Selection and Survey
The first step is choosing an optimal underwater location. The Shanghai facility is situated approximately 30 km offshore in waters around 25–30 meters deep. Key factors include:
- Proximity to an existing offshore wind farm to minimize transmission losses.
- Stable seabed with low sedimentation rate to prevent pod burial.
- Water temperature consistently below 20°C for efficient passive cooling.
- Minimal shipping traffic to reduce risk of anchor damage.
Conduct bathymetric surveys and deploy current meters for at least one full season. Use side-scan sonar to map the seafloor and identify any obstacles.
2. Subsea Enclosure Design
The server pods must withstand hydrostatic pressure at depth. For a 30 m depth, external pressure is about 3 atmospheres (300 kPa). The Shanghai facility uses steel cylindrical vessels with hermetic seals. Design considerations:
- Material: Stainless steel (316L) or titanium for corrosion resistance.
- Internal layout: Rack-mounted servers with heat exchangers integrated into pod walls.
- Sealing: Use double O-ring seals and nitrogen purge to prevent moisture ingress.
- Weight: Ballast to ensure negative buoyancy; include lifting points for deployment.
Each pod in the 24 MW facility likely houses around 500 servers (4 pods total). Perform finite element analysis (FEA) for pressure cycles and fatigue.
3. Power Infrastructure – Integrating Offshore Wind
A 24 MW load requires a dedicated wind farm capacity of at least 30 MW (accounting for intermittency). The power system includes:
- Submarine AC/DC cables from wind turbines to an offshore substation.
- Step-down transformers to medium voltage (e.g., 33 kV).
- Subsea power distribution units (PDU) that convert to low-voltage DC for servers.
- Battery or supercapacitor storage to smooth power fluctuations.
Install redundant cable routes to ensure 99.999% uptime. Use dynamic cable management to accommodate seabed movement.
4. Cooling System – Passive Seawater Circulation
The core innovation is passive cooling. Instead of chillers, the facility uses the surrounding seawater as a heat sink. Implementation steps:
- Heat Exchangers: Install plate heat exchangers on the exterior of each pod, made from a copper-nickel alloy for thermal conductivity and anti-fouling.
- Internal Circulation: Use a closed-loop glycol system inside the pod. Server heat is transferred to the glycol, which then passes through the heat exchanger to the ocean.
- Natural Convection: Design the pod shape and placement to encourage water flow. The Shanghai site uses a slight current to enhance heat dissipation.
- Monitoring: Deploy temperature sensors on both sides of heat exchangers and inside the pod. Automatically adjust server load if internal temperature exceeds 35°C.
This approach eliminates compressor-based cooling, saving up to 30% of total energy.
5. Server Deployment and Sealing
Servers must be adapted for subsea operation. Use enterprise-grade hardware with wide temperature tolerance (0–40°C) and conformal coating on circuit boards to prevent condensation. Deployment sequence:
- Pre-configure servers in land-based assembly facility. Run burn-in tests for 48 hours.
- Load servers into rack trays, then slide trays into pod. Secure with vibration dampeners.
- Connect internal data cables (fiber optic) and power cables to pod's bulkhead connectors.
- Purge pod with dry nitrogen to remove humidity, then seal final hatch.
- Attach external fiber and power wet-mate connectors.
6. Transport and Subsea Installation
Use a dynamic positioning (DP) vessel with a heavy-lift crane. For pods weighing up to 100 tons, a deck barge may be needed. Steps:
- Tow pod on a specialized submersible barge to the site.
- Lower pod using a four-point mooring system to control descent rate.
- Position pod on pre-installed concrete or steel foundation pads to prevent scour.
- Diver or ROV connects power and data cables using wet-mate connectors.
- Perform leak test by pressurizing pod interior to 0.5 bar above ambient and monitoring pressure drop.
7. Commissioning and Remote Operation
Once installed, commission the system:
- Energize power gradually, checking each server rack for proper voltage.
- Run network connectivity tests via the submarine fiber cable.
- Load AI training workloads to verify thermal performance under full load.
- Calibrate remote monitoring dashboards for temperature, power, leakage, and vibration.
- Establish a routine inspection schedule using ROVs every 6 months to check for biofouling and corrosion.
Common Mistakes
- Underestimating Biofouling: Marine organisms can clog heat exchangers. Plan for periodic cleaning using ROV-mounted brushes or ultrasonic antifouling systems.
- Inadequate Cable Protection: Submarine cables are vulnerable to fishing trawls and anchors. Bury cables at least 1 meter under the seabed or use concrete mattresses.
- Ignoring Corrosion Galvanic: Dissimilar metals in seawater cause galvanic corrosion. Use zinc anodes or impressed current protection.
- Poor Thermal Management Under Variable Current: If ocean currents weaken, passive cooling may become insufficient. Include a backup pump system to circulate water actively if needed.
- Neglecting Cybersecurity: Subsea data centers are hard to physically access, but remote attacks are possible. Implement air-gapped networks and encrypted communications.
Summary
China's 24 MW underwater AI data center demonstrates a viable path to energy-efficient, sustainable computing. By combining offshore wind power with passive seawater cooling, this facility achieves remarkable energy savings. Key takeaways for replication: careful site selection, robust subsea enclosure design, seamless power integration, and reliable thermal management. Avoid common pitfalls like biofouling and cable damage. With proper planning, subsea data centers can become a cornerstone of green AI infrastructure.