This paper explores the power provisioning strategies for large-scale computing systems, which can be as large as a warehouse. The authors present a detailed analysis of the power usage characteristics of large collections of servers (up to 15,000) over a six-month period, focusing on different application classes. They find that there is a significant gap between the actual and theoretical peak power usage at the cluster level (thousands of servers), ranging from 7% to 16%, and up to 40% in whole datacenters. This headroom allows for the deployment of additional compute equipment within the same power budget with minimal risk of exceeding it. The paper also evaluates the potential of power management schemes, such as power capping and CPU voltage/frequency scaling, to reduce peak power and energy usage. The results show that while these schemes can achieve significant savings, the benefits are more pronounced at the cluster level compared to the rack level. The authors emphasize the importance of designing systems to be power-efficient across the activity range, not just at peak performance levels. The study provides insights into maximizing the utilization of power budgets in datacenters and highlights the economic and strategic implications of power provisioning decisions.This paper explores the power provisioning strategies for large-scale computing systems, which can be as large as a warehouse. The authors present a detailed analysis of the power usage characteristics of large collections of servers (up to 15,000) over a six-month period, focusing on different application classes. They find that there is a significant gap between the actual and theoretical peak power usage at the cluster level (thousands of servers), ranging from 7% to 16%, and up to 40% in whole datacenters. This headroom allows for the deployment of additional compute equipment within the same power budget with minimal risk of exceeding it. The paper also evaluates the potential of power management schemes, such as power capping and CPU voltage/frequency scaling, to reduce peak power and energy usage. The results show that while these schemes can achieve significant savings, the benefits are more pronounced at the cluster level compared to the rack level. The authors emphasize the importance of designing systems to be power-efficient across the activity range, not just at peak performance levels. The study provides insights into maximizing the utilization of power budgets in datacenters and highlights the economic and strategic implications of power provisioning decisions.