| Title |
The Impact of Efficient MEP Equipment Design in AI Data Centers on Energy Efficiency, Reliability, Total Cost of Ownership, and Sustainability |
| Authors |
Joo-Hoon Park ; Joong-Ho Chang ; ; ; ; ; |
| DOI |
https://doi.org/10.5207/JIEIE.2026.40.5.397 |
| Keywords |
AI data center; MEP design; Efficiency; Reliability; Sustainability; TCO |
| Abstract |
The rapid growth of generative AI and large language models has increased power consumption and thermal loads in AI data centers, making efficient MEP (Mechanical, Electrical, and Plumbing) design increasingly important. This study analyzes how MEP design affects energy efficiency, reliability, total cost of ownership (TCO), and sustainability. Air, liquid, and hybrid cooling were evaluated with N, N+1, and 2N UPS redundancy architectures using PUE, WUE, CUE, availability, and expected downtime. For a hypothetical 10 MW facility with 100 racks, an average density of 100 kW per rack, and 8,760 h/year of operation, annual energy consumption, 10- and 15-year TCO, and UPS availability were quantitatively evaluated. Liquid cooling showed the lowest assumed PUE, while hybrid cooling offered a balanced trade-off between energy performance and infrastructure cost. The N+1 configuration substantially improved availability over N while requiring less redundant infrastructure than 2N. Based on these results, a four-layer integrated MEP design model is proposed: (1) rack-density-based zoned cooling, (2) matched UPS redundancy, (3) dedicated plumbing and Coolant Distribution Unit (CDU) capacity, and (4) an intelligent predictive-control layer for future operational optimization. The model integrates energy efficiency, reliability, economic feasibility, and sustainability, providing a quantitative basis for MEP configuration decisions in high-density AI data centers. |