Cloud computing has become the fundamental platform for service offerings. Such services frequently face peaks in their variable workload. Thus, the cloudification of critical applications with strict service level agreements (e.g., performability) needs a properly engineered capacity to withstand peak loads. A core problem is the prediction of the value of peaks, especially in bursty workloads. They originate in the cumulative effect of hard-to-predict rare and extreme events. Luckily, system monitoring collects enough vital information for a prediction by statistical methods. Extreme value analysis focuses on the prediction of future peaks. This paper investigates the use of extreme value theory for capacity planning in cloud platforms and services and assesses the technical metrology aspects as well.