Underwater data centres are moving from speculative research into active investment and early-stage regulatory approval, yet the technology’s biggest obstacles may prove to be political rather than engineering problems.
Nir Kshetri, a management professor at the University of North Carolina Greensboro who studies the social and environmental effects of emerging technology, describes underwater data centres as ‘a promising new approach for supporting the growth of AI,’ citing improvements in energy use, cooling efficiency, fresh-water consumption, and land use compared with conventional onshore buildings.
Microsoft’s Project Natick: What the Seafloor Proved
The most detailed evidence to date comes from Microsoft’s Project Natick. In June 2018, the company deployed a sealed pressure vessel containing 864 servers on the seafloor off Scotland’s Orkney Islands, at a depth of 117 feet, with the Orkney grid supplying 100% renewable energy to the unit. (One aggregator source puts the server count at 855, with six units failing over the trial period; Microsoft’s own publications do not specify the precise figure.)
After two years on the seabed, the servers failed at roughly one-eighth the rate of comparable land-based equipment. Microsoft Research’s Project Natick page attributes the improved reliability to reduced exposure to oxygen, humidity, and temperature fluctuations, as well as less mechanical disturbance from maintenance staff. Naval Group, the French defence contractor, led the design and deployment of the pressure vessel. While submerged, the unit also contributed processing capacity to COVID-19 research through Folding at Home and World Community Grid.
Microsoft concluded that underwater data centres ‘are reliable, [are] practical and use energy sustainably.’ Even so, it ended the experiment in 2024 and has not launched a successor project. Kshetri notes that ‘the company didn’t say why,’ though analysts point to regulatory hurdles around environmental permitting and the practical difficulty of upgrading hardware sealed inside a subsea vessel.
Underwater Data Centres Draw Serious Capital
Private investors are not waiting for Microsoft to re-enter the field. Panthalassa, a Portland, Oregon-based public benefit corporation founded in 2016, announced a $140 million Series B round on 5 May 2026, led by Peter Thiel, with participation from John Doerr, Marc Benioff’s TIME Ventures, Max Levchin’s SciFi Ventures, Hanwha Asset Management’s venture fund, Fortescue Ventures, and Super Micro Computer, among others. According to CBInsights, Panthalassa has raised a total of $193 million to date, including a subsequent Series C round.
The company’s Ocean-3 nodes perform AI inference computing at sea and transmit inference tokens back to land by satellite, rather than routing electricity through undersea cables to terrestrial grids. The model sidesteps one of the central logistical criticisms of offshore data infrastructure: the dependence on physical cable connections to shore.
A wind-powered underwater data centre launched in China uses at least 30% less electricity than traditional data centres, according to Kshetri, with offshore wind turbines reducing fossil fuel dependence. Floating installations under development in Japan and South Korea point in a similar direction.
Regulatory Resistance on the Maine Coast
The DeepGreen Western Passage project illustrates how quickly political opposition can materialise. The venture, structured as a Delaware special purpose vehicle incorporated in January 2026, applied to the Federal Energy Regulatory Commission (FERC) for a preliminary permit in February of that year. The proposal covers approximately 400 acres in Western Passage off Eastport, Maine, incorporating 170 underwater tidal turbines powering 34 high-density subsea server pods across roughly 27 acres of active infrastructure, with seawater circulation for cooling.
FERC initially notified DeepGreen of application deficiencies in July, requesting additional information. DeepGreen subsequently downsized the project’s proposed scale and energy output, creating phased stages at the recommendation of Emily Carter, chief of FERC’s New England branch in the Division of Hydropower Licensing. FERC then accepted the preliminary permit, described by the PenBay Pilot as covering a 48-month window for engineering and environmental studies. It is not a construction permit.
Local resistance has been swift. Residents and Passamaquoddy officials submitted petitions against the project almost immediately. The Bangor Daily News reported that Eastport’s Port Authority executive director, Chris Gardner, stated the authority has ‘zero authority over any of the water permitting for the site due to its nature of state and federal permitting.’ Councillor Karen Raye proposed the 180-day moratorium on data centre projects after one of two community ordinances designed to block the development was found by the city’s attorney to be unlawful and unenforceable. The city council passed the moratorium unanimously.
The broader legislative picture in Maine is equally cautious. Maine lawmakers approved a bill blocking data centre construction in the state until November 2027, alongside a council tasked with recommending guardrails to ensure data centres do not drive up energy costs for residents.
Kshetri acknowledges that moving data centres offshore creates ‘new concerns about harm to the marine environment,’ and whether the ocean can sustain AI’s computing demands without becoming its next environmental casualty remains unresolved. With the Maine moratorium in effect and a statewide legislative freeze running until November 2027, the question for developers is less about engineering viability than about whether public consent can be won before the regulatory window closes.
