Presentation Information
[BT-1-03]Building and Operating Data Centers for High-Density AI Infrastructure
〇Naoki Horie1 (1. KDDI CORPORATION)
Keywords:
AI Data Center,Liquid cooling server,GPU server
With the rapid widespread adoption of artificial intelligence and the increasing scale of AI models, AI data centers are facing significant challenges. These include unprecedented levels of concentrated power consumption and high heat generation compared to conventional data centers, alongside the demand for rapid deployment to keep pace with rapidly evolving GPU servers. Specifically, the GPU servers generating high heat exceeding 100 kW per rack has pushed traditional air-cooling methods to their limits, rendering the adoption of liquid cooling technologies inevitable. However, constructing and operating liquid cooling systems necessitate novel technical expertise distinct from conventional data center operations, such as mitigating coolant leakage risks and achieving tightly integrated control between IT equipment and cooling facilities. Consequently, these challenges have constituted a major barrier to the practical implementation of such advanced infrastructure in society.