Modeling and performance analysis of a novel air-liquid hybrid cooling system for high-power containerized AI data centers

With the rapid development of artificial intelligence (AI) computing and edge computing infrastructure, high-power-density data centers are placing increasingly stringent demands on the heat dissipation capability, rapid deployment, and flexible scalability of cooling systems. In addition, the mixed deployment of AI and general computing, along with the differentiated heat dissipation requirements of various electronic components, makes traditional air cooling methods increasingly inadequate. To address these challenges, this study proposes a novel air-liquid hybrid cooling system for containerized AI data centers, integrating a water-cooled pump-driven heat pipe in-row air conditioner, a staged two-phase cooling distribution unit (CDU), and dry-wet hybrid cooling towers. Heat transfer and power consumption models were developed for the system, and the corresponding models for the aforementioned three subsystems were validated against measured data. Compared with a baseline air-liquid hybrid cooling system, the proposed system achieved higher energy efficiency ratios (EER) across different outdoor temperature conditions. Specifically, the EER ranged from 9.94 to 31.49 at an air-to-liquid cooling load ratio of 3:7 and from 12.87 to 33.79 at 2:8. A year-round assessment was conducted for 21 representative cities across different climate zones in China. The results showed that the annual EER (AEER) of the proposed system ranged from 10.97 to 16.88 at an air-to-liquid cooling load ratio of 3:7 and from 13.61 to 19.79 at 2:8, representing improvements of 13.15%–32.98% and 16.46%–30.38% over the baseline system. The corresponding power usage effectiveness (PUE) ranged from 1.109 to 1.141 and from 1.101 to 1.123, while the water usage effectiveness (WUE) ranged from 0.565 to 0.988 L/kWh and from 0.375 to 0.789 L/kWh. The results can serve as a reference for the design and deployment of hybrid cooling systems for containerized AI data centers in different climate zones.