ULSAN — Researchers have developed an integrated energy system that captures heat discarded by artificial intelligence data centers, uses it to trap carbon dioxide and converts it back into electricity. The system can cut operating costs by about 48% and life-cycle greenhouse gas emissions by up to 72% compared with conventional setups.
The Ulsan National Institute of Science and Technology (UNIST) said on the 15th that a team led by Professor Lim Han-kwon of its Graduate School of Carbon Neutrality had designed the system and verified its performance. It combines solid oxide fuel cells (SOFC), waste-heat power generation based on an organic Rankine cycle (ORC) and carbon capture technology.
AI data centers run around the clock to handle large-scale computation and consume enormous amounts of electricity. Cooling alone accounts for about 39% of their total power use, making heat removal and power efficiency central challenges.
The team distributed natural gas-fueled SOFC units near the data center to supply power directly. High-temperature exhaust heat from the fuel cells, which generate electricity at high efficiency, is fed into a process that heats and separates a carbon dioxide capture liquid for reuse, curbing carbon emissions.
Heat from the servers is recovered as electricity. Waste heat from the coolant that has cooled the servers evaporates a special refrigerant, and the resulting vapor drives a turbine using ORC technology. The electricity produced is then fed back into the data center's cooling systems.
The researchers compared 24 scenarios for a data center with 20,000 graphics processing units, reflecting energy prices and power grid structures in South Korea, the United States and the European Union.
The analysis found that combining SOFC, carbon capture and ORC cooling cut total costs by 48% and greenhouse gas emissions by up to 72% compared with a conventional grid-and-air-cooling combination. In all three regions, adopting ORC cooling saved up to $19 million a year in cooling power costs. Greenhouse gas emissions fell by about 80,000 to 150,000 tons a year, and factoring in emissions trading schemes added $5 million to $10 million a year in avoided carbon costs.
The optimal power source, however, varied with local energy conditions. In the United States, where natural gas is cheap, direct generation from SOFC held an economic edge, while in South Korea, where gas prices are higher, drawing on the existing power grid was more cost-effective.
"The energy system best suited to a data center depends on regional energy prices and the power mix," Lim said. "This study will provide a practical benchmark for designing energy systems and setting ESG strategies tailored to the scale and location of a data center."
The research was supported by the Ministry of Science and ICT and the National Research Foundation of Korea, and the findings were published in the online edition of the Chemical Engineering Journal, an international chemical engineering journal.
