China Opens 24MW Undersea AI Data Center Powered by Offshore Wind
Abhishek Gautam11 min readQuick summary
Shanghai Lingang undersea data center switched on May 2026: 24MW planned, offshore wind direct feed, seawater cooling, 192 racks. World-first model for AI power and latency.
Read next
- Dell AI Server Revenue +757% to $16.1B, $51B Backlog, FY27 Target $60B
- Nvidia Installs 16-GPU Data Centers on Homes: XFRA Explained
China switched on the world's first undersea data center (UDC) powered directly by offshore wind in late May 2026 — ~10 km off Shanghai's Lingang coast, currently running at ~2.3 MW with 24 MW planned, 192 server racks across four submerged levels, and seawater cooling that cuts land use ~90% and electricity use ~22.8% versus conventional sites.
Built by a China Communications Construction (CCCC) subsidiary, the project is Beijing's bet that AI factories need integrated power + cooling + compute — not just more land-based megawatts.
What Is the Shanghai Lingang Undersea Data Center?
Undersea data centers place sealed modules on the seabed to use ocean water for heat rejection and reduce freshwater draw — Microsoft pioneered early concepts; China now claims the first operational wind-linked production system.
| Spec | Reported figure |
|---|---|
| Location | ~10 km offshore, Shanghai Lingang |
| Current load | ~2.3 MW |
| Planned capacity | 24 MW (~20,000 households equivalent) |
| Racks | 192 across 4 levels |
| Power source | Offshore wind via subsea photoelectric composite cables (direct, bypassing normal grid routing) |
| Wind share | ~95% of electricity (state media estimates) |
| Cooling | Circulating copper-pipe heat exchange + seawater |
| Land savings | >90% vs above-ground DC (developers) |
| Switched on | Late May 2026 |
Shanghai hosts large-model labs, autonomous driving, biotech, and fintech — workloads where latency and power cost both matter.
Why This Matters for AI Infrastructure Globally
Power is the binding constraint on AI scale. China is testing three simultaneous fixes:
- Undersea — cooling + space efficiency
- Offshore wind — renewable direct attach (no long grid losses)
- Integrated engineering — power, cooling, compute as one system
Compare:
- EU CADA — triple datacenter capacity on land (EU Tech Sovereignty)
- NEOM/DataVolt — desert seawater cooling pivot (The Line cancelled)
- Residential XFRA — edge Blackwell nodes (Nvidia Span XFRA)
- Orbital DC hype — China's UDC is a real switched-on alternative to space marketing
Developer read-through: If subsea modules prove O&M economics, expect coastal APAC latency-sensitive inference to shift offshore before inland hyperscale expands.
Cross-read China cheap energy AI datacenters and SMIC/Huawei stack.
Risks and Open Questions
- Subsea maintenance — failed pumps, cable cuts, typhoon exposure
- Sovereignty — data residency rules for offshore Chinese modules serving global clients
- Export controls — advanced GPUs inside modules still subject to US licensing if sourced from Nvidia/AMD
For cable chokepoints in the same region, see undersea cables Middle East risk.
Key Takeaways
- Late May 2026: Shanghai Lingang undersea AI data center online — world-first offshore-wind-direct model
- 2.3 MW now, 24 MW planned, 192 racks, ~22.8% lower power use vs conventional cooling
- ~95% offshore wind power; >90% land-use reduction vs land DCs
- Template for AI-era integrated infra — power, cooling, compute engineered together
- For developers: watch coastal latency SLAs and renewable-attached inference pricing in APAC
Sources
FAQ
Frequently Asked Questions
Where is China's undersea AI data center located?
The Shanghai Lingang undersea data center operates about 10 kilometers offshore from Shanghai's Lingang area in eastern China. It entered operation in late May 2026.
How much power does the Shanghai undersea data center use?
Reports state the facility is currently operating at about 2.3 megawatts with a planned capacity of 24 megawatts, enough to power roughly 20,000 households at full build-out.
How is the undersea data center powered?
Electricity comes primarily from offshore wind farms transmitted directly to submerged modules through subsea photoelectric composite cables, bypassing traditional grid routing. State media estimates offshore wind supplies about 95% of power.
Why build AI data centers under the sea?
Developers cite seawater cooling that reduces electricity consumption by about 22.8%, cuts freshwater use, saves more than 90% of land compared with above-ground centers, and integrates renewable offshore wind for AI workloads near coastal hubs like Shanghai.
Free Weekly Briefing
The AI & Dev Briefing
One honest email a week — what actually matters in AI and software engineering. No noise, no sponsored content. Read by developers across 30+ countries.
No spam. Unsubscribe anytime.
More on AI Infrastructure
All posts →Dell AI Server Revenue +757% to $16.1B, $51B Backlog, FY27 Target $60B
Dell Q1 FY2027 earnings: AI server revenue hit $16.1B (+757% YoY), orders booked $24.4B, backlog reached a record $51.3B. FY27 AI revenue target raised to $60B. Stock surged 32%.
Nvidia Installs 16-GPU Data Centers on Homes: XFRA Explained
Span and Nvidia are deploying XFRA nodes — liquid-cooled boxes with 16 RTX Pro 6000 Blackwell GPUs — on residential homes, tapping unused grid capacity. 100 pilot units launch Q3 2026 on PulteGroup builds.
Nvidia Picks Unitree H2 Plus for GR00T Robot, $620M China IPO Cleared
June 1, 2026: Nvidia sells Isaac GR00T research humanoid on Unitree H2 Plus with Jetson Thor Blackwell. Unitree STAR IPO approved — 4.2B yuan raise, ~$6.2B valuation, 73-day review.
Nvidia Vera CPU: 1.8x Faster Agent Chip, Not Built for Humans
Nvidia Vera CPU launched May 31, 2026 at GTC Taipei: 88 Olympus Arm cores, 1.8x faster than x86 on agent workloads. Anthropic, OpenAI, NYSE, OCI, CoreWeave adopt. Fall 2026 availability.
Free Tool
Will AI replace your job?
4 questions. Get a personalised developer risk score based on your stack, role, and what you actually build day to day.
Check Your AI Risk Score →Written by
Software Engineer based in Delhi, India. Writes about AI models, semiconductor supply chains, and tech geopolitics — covering the intersection of infrastructure and global events. 795+ posts cited by ChatGPT, Perplexity, and Gemini. Read in 164 countries.