Data Center Total Costs Halved by Recycling Waste Heat

[Financial News] An analysis has found that supplying electricity to an artificial intelligence (AI) data center with fuel cells, while recycling heat from the fuel cells and servers for carbon dioxide capture and power generation, respectively, could reduce total costs by approximately 48% and greenhouse gas emissions by up to 72% compared with conventional methods.
On the 15th, a research team led by Lim, Hankwon of the Graduate School of Carbon Neutrality at Ulsan National Institute of Science and Technology (UNIST) said it had reached this conclusion after designing an integrated energy system that uses solid oxide fuel cells as a data center power source and recycles heat generated during power generation and cooling, then analyzing the system’s economic feasibility and life-cycle greenhouse gas emissions.
According to the research team, data centers must receive a stable supply of enormous amounts of electricity because they operate numerous servers around the clock. Installing solid oxide fuel cells (SOFCs), which generate electricity from natural gas, near a data center and supplying power directly would allow operators to add modules as needed to meet rising electricity demand.
In the integrated energy system proposed by the research team, heat released by the fuel cells is used for carbon dioxide capture. Solid oxide fuel cells have the highest power-generation efficiency among fuel cells, but they emit carbon dioxide and high-temperature heat. The system captures the carbon dioxide with a special liquid that absorbs it, then uses heat from the fuel cells to heat the liquid, separate the carbon dioxide, and reuse the liquid.
Heat generated by the servers is also used to produce electricity. In this Organic Rankine Cycle (ORC) process, heat from the coolant used to cool the servers evaporates a special refrigerant, and the resulting vapor drives a turbine to generate electricity.
The research team assumed a data center with 20,000 GPUs and analyzed eight combinations based on power sources, cooling methods, and whether carbon capture was used, taking into account energy prices and power-generation structures in South Korea, the United States, and the European Union (EU).
The analysis showed that an integrated system combining solid oxide fuel cells, carbon capture, and Organic Rankine Cycle cooling generally ranked higher than other combinations when costs and greenhouse gas emissions were assessed together. Organic Rankine Cycle cooling demonstrated strong economic and environmental performance in South Korea, the United States, and the European Union. It could reduce electricity costs for annual cooling by up to $19 million. Because less electricity would need to be generated, greenhouse gas emissions were also estimated to fall by approximately 80,000 to 150,000 tons annually. When systems that impose a cost on carbon emissions, such as the EU Emissions Trading System (EU ETS), are taken into account, the reduction in emissions could further cut annual carbon costs by approximately $5 million to $10 million.
However, the relative advantages of power-supply methods varied by region. Solid oxide fuel cells were economical in the United States, where natural gas is inexpensive, while in South Korea, where natural gas is costly, using the existing power grid was more advantageous when minimizing costs was the priority.
Lim, Hankwon said, "The energy system best suited to a data center varies depending on regional energy prices and carbon emissions associated with power generation." He added, "This study will help design energy systems and develop ESG strategies that reduce costs and carbon emissions according to the size of a data center and local conditions."
The findings were published online on August 14 in Chemical Engineering Journal (CEJ), a leading international journal in the field of chemical engineering.
[email protected] Yeon Ji-an Reporter