Can Floating Data Centers Meet the Growing Energy Demands of AI?

Data centers powering the AI revolution are consuming more electricity than many small countries. According to an
International Energy Agency report,
energy demand could soar to 945 terawatt hours per year by 2030, surpassing Japan’s entire electricity consumption.
As AI technology continues to evolve, the hunger for power has led companies to consider not just earthly solutions but also outer-space options to harness constant solar energy.
Intriguingly, startup Panthalassa is working on an autonomous floating data center that harnesses computational power in the open sea.

Recently, the Oregon-based company announced $140 million in funding, stating that its innovative platform could bypass overloaded power grids while offering carbon-free computing at sea. However, it remains to be seen if relocating computing power offshore can truly ease the challenges faced by current data centers; it might simply replace one expensive issue with another.

“Wave energy is a well-established technology, but the ocean presents a brutal environment,” warns Jonathan Koomey, a data center energy utilization expert at the Lawrence Berkeley National Laboratory.
“Saltwater and wave action can lead to mechanical malfunctions.”

Shaped like a golf ball atop a tee, Panthalassa’s floating data center stands 85 meters high (almost the height of Big Ben) and is constructed from sheet steel.
These structures are towed into position and can autonomously generate power to run AI workloads without relying on grid electricity, zero emissions, or traditional engines.

The “tee” section of the platform features a long tube with an open base that captures seawater as waves cause it to rise and fall. This movement forces water through the tube into the hollow “ball” section, which floats primarily due to trapped air.
The dynamic movement of water activates turbines that generate electricity, powering the vehicle’s GPUs, computing hardware, and satellite communication systems.

Conventional data centers consume significant quantities of water for cooling AI hardware.
In contrast, Panthalassa’s servers are housed in enclosed modules below the water surface, allowing the container walls to act as heat exchangers that dissipate heat into the cooler seawater.
While ocean mixing helps eradicate waste heat, the implications for local marine ecosystems remain uncertain.

Panthalassa aims to achieve what few data center operators have dared: to maintain crucial computing infrastructure beyond the oversight of human engineers.
“Our data consistently identifies power and network issues as the leading causes of data center outages,” states Jacqueline Davis of the Uptime Institute, a global authority on data center performance.
“These complications can be particularly challenging to manage in remote locales with limited or no staffing.”
Panthalassa did not respond to inquiries from the New Scientist before this article was published.

Automation within data centers mostly focuses on monitoring and analysis, with physical human intervention often necessary during unusual events, such as manual compressor restarts.

This presents a significant challenge for Panthalassa. Latency is another hurdle.
Data processed on the floating platform is sent to users on land via Starlink satellites, which have limited bandwidth and higher latency compared to fiber optic cables.
Consequently, while the node may work effectively for long-duration AI workloads—such as training advanced mathematical models—applications that require swift responses, like chatbots and search assistants, could face challenges.

“Constraints in power are most severely impacting extensive AI training data centers,” noted Davis. She suggests that Panthalassa’s model is more plausible when power demands for advanced AI become significant enough to align with AI training needs.
Until then, floating data centers may struggle to remain competitive with land-based options.

Although Panthalassa’s approach is innovative, the concept of offshore data centers isn’t entirely new.
Aikido Technologies is developing floating data centers integrated with offshore wind platforms, while Mitsui O.S.K. Lines is exploring ship-based computing systems designed to exploit ocean energy sources.
Past initiatives, like Microsoft’s underwater Natick project, have examined whether housing servers in or near water enhances cooling and efficiency.

Nevertheless, offshore computing remains predominantly experimental. Besides engineering challenges, firms must demonstrate that ocean-based systems can economically compete with traditional data centers linked to power grids and fiber networks.
“Data centers can achieve economies of scale by building at larger sizes, which is why they are typically so extensive today,” remarks Koomey.
“Scaling up to accommodate the fixed costs of computing is considerably more challenging and riskier on water.”