China is sinking AI data centers into the ocean
China is taking data center cooling underwater, with construction kicking off this June on a wind-powered undersea facility six miles off Shanghai’s coast. Backed by Hailanyun, the project uses seawater cooling to slash electricity use by 30%, with offshore turbines providing 97% of its power. Why it matters: As AI clusters guzzle massive amounts of freshwater and energy, undersea deployments are graduating from tech demos into commercial infrastructure.

AI is facing a massive climate contradiction: the more computing power we build, the more water and electricity we burn just to stop the servers from melting down.
A typical on-land data center churns through hundreds of thousands of gallons of water every day for evaporative cooling, with cooling systems eating up around 40 percent of its electricity. To make matters worse, many are built in dry regions like Arizona or parts of Spain to avoid humidity issues, pitting AI directly against local water needs.
Now, China is pursuing a radical alternative by dropping servers straight into the ocean. Construction began this June on a $223 million, wind-powered undersea data center roughly six miles off the coast of Shanghai, led by maritime tech firm Hailanyun.
Natural Chilling and Offshore Wind
The mechanics behind undersea data centers are straightforward: sea pipes cycle cold ocean water through radiators on the backs of server racks, pulling heat away without massive energy-hungry chillers.
Hailanyun claims an assessment conducted with the China Academy of Information and Communications Technology shows the site uses at least 30 percent less power than land-based counterparts. On top of that, an adjacent offshore wind farm will supply roughly 97 percent of the facility's power, according to spokesperson Li Langping.
The initial phase houses 198 racks, enough for 396 to 792 AI servers, with operations scheduled to start in September. Li says it will pack enough compute to finish the equivalent of a GPT-3.5 training run within a single day. That is still modest compared to mainland giant campuses that house up to 10,000 racks, but it is an aggressive first step.
Reviving Microsoft's Shelved Playbook
Subsea computing is not a brand-new concept. Microsoft pioneered the idea over a decade ago with Project Natick, sinking a sealed capsule with more than 800 servers off Scotland. When retrieved two years later, the nitrogen-filled pod actually logged lower failure rates than terrestrial servers because it lacked human interference and oxygen-driven corrosion.
Yet Microsoft quietly shelved active underwater deployments, turning Natick into a theoretical research platform. Hailanyun, on the other hand, went from a pilot project in Hainan in late 2022 to a commercial buildout in Shanghai in less than 30 months.
It is not without risks. Marine biologists warn that localized thermal discharge during marine heatwaves could deplete dissolved oxygen and disrupt sea life. Others point to obscure vulnerabilities, like a 2024 study showing how specific underwater acoustic frequencies could sabotage hardware. Hailanyun pushes back, pointing to a 2020 Pearl River test where surrounding water temps rose by less than one degree Celsius.
Other coastal hubs are watching closely: South Korea has planned subsea installations, while Singapore and Japan are testing floating surface facilities.
Why it matters
For the AI ecosystem, cooling and power constraints are rapidly becoming the ultimate ceiling on scale. Coastal megacities like Shanghai have virtually no vacant land and limited grid bandwidth for gigawatt-scale data clusters. Siting servers six miles offshore directly connects high-density AI clusters to plentiful offshore wind, bypassing both real estate bottlenecks and municipal freshwater supplies.
This commercial push also marks an interesting divergence between US and Chinese infrastructure strategies. While US tech giants focus heavily on securing land and nuclear partnerships, China is taking a bold bet on maritime industrialization.
In our view, pairing subsea passive cooling with direct offshore wind generation is an exceptionally clever engineering shortcut. However, maintenance remains the glaring question mark. Swapping out a failed GPU on land takes minutes; repairing flooded or unresponsive nodes deep in the ocean requires naval logistics. Unless Hailanyun proves that the nitrogen-sealed hardware can reliably survive without physical human touch for years, operating margins could drown under maritime maintenance overhead.
Sources (1)
- scientificamerican.com(opens in a new tab)Originalscientificamerican.com