A new push aims to move AI computing offshore, but power, latency, and economics may limit floating data centres.
AI is rapidly turning electricity into a critical bottleneck. and a bold proposal is now trying to solve the problem by moving the machines that need that power into the ocean.. The data centres powering the AI boom already consume more electricity than some small countries. and the International Energy Agency has projected that their demand could rise to 945 terawatt-hours a year by 2030—surpassing Japan’s total electricity use.
With compute expanding faster than grid capacity in many regions. some companies are exploring space as a potential “always-on” energy source.. But one start-up believes a more immediate answer may be found closer to home. even if that means leaving land entirely.. Panthalassa is building autonomous floating data centres intended to place computing power in international waters. using solar energy indirectly through its design premise and. more directly. through generating electricity offshore without a connection to local grids.
The Oregon-based company announced $140 million in funding last week for the effort.. It says its floating platforms could bypass overwhelmed electrical grids while delivering carbon-free computing in international waters.. Yet beyond the obvious engineering hurdles of building and operating critical infrastructure in the open sea. the biggest question is whether offshore systems would ease the core constraints that limit data centres—or whether they simply shift the pressure onto other. potentially costlier problems.
A key concern is that the ocean is not a gentle operating environment.. Jonathan Koomey. a former researcher at Lawrence Berkeley National Laboratory in California who studies data centre energy use. cautioned that while wave power is a proven concept. the combination of salt and constant motion is particularly punishing for machinery.. In other words. even if offshore power generation works in principle. it has to survive years of corrosion. vibration. and wear.
Panthalassa’s prototype design is visually striking: the platforms are shaped like a golf ball perched on a tee. with a height of 85 metres—roughly the stature of Big Ben.. The structure is made of plate steel, hauled into the water by boat and then self-propelled to its designated location.. Once there. the company says the system can generate its own electricity. run AI workloads. and communicate back to users without relying on a grid connection. emissions-heavy fuel. or engines dedicated to power.
The “tee” portion of the system contains a long tube open at the bottom.. As waves lift and drop the platform. sea water is pushed through the tube and upward into the hollow “ball. ” which is mostly filled with air to help it float.. Moving water drives turbines. producing electricity that powers onboard graphics processing units and other computing hardware. along with satellite communications equipment.. The overall pitch is to keep computing running continuously while cutting dependence on strained terrestrial infrastructure.
Traditional data centres also face another energy-heavy constraint: cooling.. Water use is often substantial, and cooling requirements can drive significant operational costs.. Panthalassa’s approach differs by placing servers inside sealed modules beneath the water surface.. The company’s design proposes that the container walls will act as heat exchangers. allowing heat to dissipate into surrounding cold seawater as the ocean carries it away through currents and mixing.
Still, the environmental side remains less certain.. While ocean mixing may disperse waste heat, the potential effects on nearby marine ecosystems were described as unclear.. That uncertainty matters because deploying computing at sea at scale would not just add to energy demand—it could also change local temperature patterns. with possible downstream impacts.
Another critical issue is reliability.. Running data centres remotely raises the risk that the systems most likely to fail are also the hardest to repair.. Jacqueline Davis at the Uptime Institute. which assesses data centre performance. says that data centre outages commonly trace back to power and networking as the top root causes.. Both can be especially difficult to manage in a remote environment with limited or no staff on site.
Automation is often used in data centres for monitoring and analytics. but physical intervention is still frequently required. Davis noted. particularly during abnormal incidents such as situations where cooling compressors need manual restarts.. That means that even if a floating platform can generate power and route traffic. the question becomes whether it can handle unusual failures without costly delays or frequent maintenance trips.
Latency is likely to be another determining factor for what kinds of AI workloads offshore systems can realistically support.. Panthalassa plans to send the processed data back to users on land through Starlink satellites.. Compared with fibre-optic cables. satellite links come with higher latency and limited bandwidth. which could shape the operational model for the floating nodes.
The design may fit best for “job-based” AI tasks that can run for hours or even days. then return results—examples mentioned include training advanced models or running scientific simulations.. For consumer-facing applications such as chatbots and search assistants. however. users expect fast response times and continuous interaction with the network. and high latency could become a serious limitation.
Davis also pointed to where the immediate pressure is landing in today’s AI economy: the largest AI training data centres.. Power constraints, she said, affect those facilities most acutely.. Panthalassa’s offshore concept would become more viable only if the total power needs of running trained AI systems eventually grow enough to rival the training phase. she added.. Until that shift happens, floating data centres may struggle to compete with land-based systems for the most power-hungry workloads.
The idea of moving data centres away from land is not unique to Panthalassa, even if its specific design is.. Aikido Technologies is developing floating data centres integrated into offshore wind platforms, and Mitsui O.S.K.. has been studying ship-based computing systems powered by marine energy sources.. Earlier experiments also tested whether water-adjacent computing could improve cooling and efficiency, including Microsoft’s underwater Project Natick.
Despite these efforts, offshore computing remains largely experimental.. The sector still has to demonstrate that ocean-based systems can compete economically with conventional data centres tied to established power grids and fibre networks.. Koomey highlighted a structural advantage that land-based operators have: economies of scale.. Large facilities spread fixed costs over more compute. which makes building big installations on land less risky than attempting comparably large compute platforms at sea.
For now. the promise of floating data centres rests on a delicate balance: they must prove that wave-driven power generation and autonomous operations can be reliable over time. that cooling and heat dispersal do not create unacceptable environmental trade-offs. and that satellite communications can meet performance needs for targeted workloads.. If those conditions hold. offshore infrastructure could become more than a science experiment—it could become a new way to feed the AI boom.. But if any of these constraints prove too expensive or too difficult. the ocean may simply replace one set of data-centre bottlenecks with another.
floating data centres AI energy demand wave power computing data centre cooling satellite latency offshore computing