Optical networks, given their excellent characteristics in terms of high bandwidth and low latency, play a crucial role in enabling today's applications that heavily rely on network connectivity. This is exemplified by the evolution towards grid and cloud computing, as observed during the last decennium. Such cloud computing based applications call for a tighter integration between the control and management of data centers, and the (optical) wide area network connectivity they require. The advent of the resulting "optical clouds" also demands an appropriate network (and data center) planning methodology, which exhibits significant differences compared to, e.g., the classical routing and wavelength assignment (RWA) methods. In this paper, we explain fundamental concepts that cloud computing is based on (i.e., anycast routing and virtualisation), and in particular outline what challenges they bring about in network dimensioning. We survey possible solutions, and specifically argue how large scale optimization (column generation) can be adopted to dimension resilient optical networks for cloud computing. Sample case studies illustrate potential capacity savings enabled by exploiting the particular cloud concepts.