Cloud vendors like GCP and AWS offer an increasingly wide array of heterogeneous hardware instances with different costs and performance characteristics, which can be purchased with per-second billing. Nonetheless, it’s hard for developers to select the best instance to minimize workload cost given runtime constraints when so many options are available. To address this problem, Viktor Leis and Maxmillian Kuschewski develop a white-box model to estimate the cost of a given analytical query processing workload on a specific hardware configuration. They discuss how this system could be used to build a cost-optimal, cloud-native OLAP system.