While a standard rack uses 7-10 kW, an AI-capable rack can demand 30 kW to over 100 kW, with an average of 60 kW+ in dedicated AI facilities. This article provides a condensed analysis of these costs, key efficiency metrics, and optimization strategies. Just like virtual CPUs (vCPUs) relate to physical CPUs in cloud computing, kW/rack defines power use per server rack. This impacts colocation pricing, energy use. Use this TradeOff Tool to estimate the power required by a data center with traditional, or AI/HPC servers. Configure different server, storage, and design attributes to explore different scenarios. White paper 3 presents methods for calculating power and cooling requirements and provides. This growth is heavily influenced by the proliferation of AI, Machine Learning (ML), and High-Performance Computing (HPC) workloads, which drastically increase power consumption per rack.
[PDF Version]