Beyond PUE: A Boundary-Aware Primary-Energy Framework for Evaluating Self- Powered Computing Infrastructure, with Application to a Hybrid Offshore Floating Data Center

In the world of modern computing, the servers that power our digital lives consume vast amounts of electricity. To measure how efficiently a data center uses that power, engineers rely on a standard score called Power Usage Effectiveness, or PUE. This number compares the total energy a facility consumes against the energy actually used by the computers themselves. The difference accounts for the "overhead" needed to keep the machines running, such as cooling systems and power distribution. A lower score is better; a perfect score of 1.0 would mean every watt of electricity goes directly to the processor, with none lost to cooling or support systems. For decades, this metric has driven improvements, helping facilities reduce waste. However, this score has a blind spot. It only looks at what happens inside the building's walls. It assumes the electricity arrives from an external power grid, where the losses of generating that power are someone else's problem. But as companies begin building data centers in remote locations that generate their own electricity on-site, this narrow view can hide a much larger truth about efficiency.

A new study by independent researcher Leon Sandler investigates this blind spot by examining a specific, proposed design for a floating data center located offshore. This concept, known as the Hybrid Seabed-Energy Floating Data Center, envisions a platform that sits in the ocean, cooled by the deep sea, and powered by a mix of natural gas turbines and wave-energy converters. The original proposal for this project claimed a highly efficient PUE of roughly 1.04, suggesting it was a model of sustainability. Sandler's work tests whether this low score truly reflects an efficient use of fuel when the facility is responsible for generating its own power. By looking beyond the facility's walls to the entire process of turning fuel into computing power, the study reveals that the original claim was misleading.

The researchers built a detailed model of the proposed floating platform to see how the different energy sources would actually perform. The design relied heavily on gas turbines to provide a steady base of power, with a smaller array of floating buoys intended to capture energy from ocean waves. When the team calculated the average power output of these wave buoys using realistic data from existing devices, they found a surprising result. Despite being a central part of the original concept, the wave energy array contributed less than 1.2 percent of the total power needed. In practical terms, the waves provided only a tiny fraction of the electricity, meaning the facility was almost entirely dependent on burning natural gas. The study concludes that the wave component was far too small to be considered a genuine hybrid power source or a reliable backup; it was merely a minor supplement.

The most significant finding emerged when the researchers applied a new way of measuring efficiency that accounts for the entire journey of energy, from the fuel source to the computer chip. They introduced a "primary energy efficiency" metric, which includes the energy lost during the conversion of gas into electricity and the extra energy required to capture carbon emissions from the exhaust. When this broader view was applied, the picture changed dramatically. While the facility's internal PUE remained impressive at around 1.08 to 1.09—comparable to the best land-based centers—the overall efficiency of turning fuel into computing power was only between 25.5 and 38.7 percent. This is significantly lower than a conventional data center connected to a standard power grid, which operates at about 43 percent efficiency, and far lower than a facility powered by renewable energy on the grid, which can reach 93.5 percent.

The study clarifies that a low PUE score does not automatically mean a self-powered facility is sustainable. In this case, the facility's high efficiency inside the building was completely offset by the heavy losses in the power generation process outside it. The gas turbines lost a large portion of their fuel energy as heat during conversion, and the equipment needed to capture carbon emissions consumed even more power. These losses happened before the electricity ever reached the servers, so they were invisible to the standard PUE metric. The research demonstrates that for any facility that generates its own fossil-fuel power, looking only at the building's internal efficiency gives a false sense of environmental performance.

The author is careful to note that this conclusion applies specifically to this type of offshore, gas-powered architecture and does not mean all self-powered computing is a bad idea. The study highlights a genuine advantage to such platforms: they are completely independent of the main power grid, making them immune to blackouts and grid failures that can disrupt land-based centers. However, this reliability comes with a steep energy cost when fossil fuels are used. The research suggests that for these floating data centers to be truly sustainable, the focus must shift from just improving the cooling and internal systems to addressing the massive inefficiencies of on-site power generation. Until the energy source itself becomes cleaner and more efficient, the promise of a "zero grid dependency" data center may come at a higher environmental price than previously thought.