Ocean Data Centers Emerge as AI Infrastructure Alternative
Companies from China to Singapore are testing underwater and floating facilities to reduce energy use and land demands, but marine environmental risks remain unresolved.

The explosive growth of artificial intelligence is pushing technology companies and governments to rethink where they house the massive computing infrastructure AI requires. A wave of experimental projects across Asia, Europe, and North America is now testing whether the ocean can serve as viable real estate for data centers—potentially easing the strain on electricity grids, freshwater supplies, and available land.
China launched what appears to be the world's first wind-powered underwater data center in Shanghai in June 2025, with full commercial operations beginning in May 2026. The $226 million facility uses seawater as a coolant and consumes at least 30% less electricity than conventional land-based centers, according to details first reported by The Conversation.
Japan, Singapore, South Korea, and the United States are pursuing similar initiatives. Japan opened a containerized data center on a floating platform near Yokohama in 2025, powered entirely by solar panels mounted on the same structure. Singapore's infrastructure firm Keppel is constructing a four-story floating data center scheduled to open in 2028. In Maine, DeepGreen Western Passage has proposed a submersible facility in the Bay of Fundy that would draw power from tidal turbines.
Why it matters
Data centers already account for roughly 1-2% of global electricity consumption, and AI workloads are accelerating that demand. Traditional facilities also consume enormous volumes of fresh water for cooling—a resource under increasing stress in many regions. Ocean-based alternatives could reduce both pressures while keeping infrastructure physically close to coastal population centers where more than half the world's people live. But the approach introduces new environmental and operational trade-offs that remain largely untested at scale.
The Microsoft precedent
Microsoft's 2018 Project Natick deployment near Scotland's Orkney Islands provided early validation. The company submerged 864 servers in a sealed container on the seafloor, connected to shore by undersea cable. After two years, Microsoft reported server failure rates one-eighth those of comparable land facilities—likely due to reduced exposure to oxygen, humidity, temperature swings, and physical disturbances from maintenance activity.
Despite those results, Microsoft ended the project in 2024 without building additional underwater centers. While the company did not publicly explain the decision, analysts have pointed to regulatory complexity, environmental permitting requirements, and the difficulty of upgrading or repairing submerged equipment as probable factors.
Environmental concerns linger
The primary ecological risk involves thermal pollution. Both China's Hainan underwater facility and Portugal's SIN01 land-based center—which uses Atlantic seawater for cooling—discharge water approximately 1 degree Celsius warmer than intake temperatures. While that increase may seem modest, marine species depend on stable temperatures for breeding, feeding, and migration. Multiple facilities operating in proximity could create localized thermal zones that disrupt ecosystems already under pressure.
UNESCO estimates that roughly 60% of marine ecosystems are degraded or used unsustainably. Ocean temperatures are rising due to climate change, threatening coral reefs, mangroves, and seagrass beds. Adding waste heat from data centers compounds existing stress on marine food webs.
Maintenance presents another challenge. Underwater facilities cannot be serviced in place—entire sealed modules may need retrieval to the surface even for single component failures, limiting operational flexibility.
Whether ocean-based data centers can scale sustainably depends on resolving these technical, economic, and environmental questions. The details in this report were first published by The Conversation.
This is an original analysis by the Omega editorial team. Source reporting: AI Watch.
Want systems like this working for your business?
Book a Call
