Planned offshore wind farm clusters have a rated capacity of more than 10 GW. The layout optimization and yield estimation of wind farms is often performed with computationally inexpensive, analytical wake models. As recent research results show, the flow physics in large (multi-gigawatt) offshore wind farms are more complex than in small (sub-gigawatt) wind farms. Since analytical wake models are tuned with data of existing, sub-gigawatt wind farms they might not produce accurate results for large wind farm clusters. In this study the results of a large-eddy simulation of a 15 GW wind farm are compared with two analytical wake models to demonstrate potential discrepancies. The TurbOPark model and the Niayifar and Porté-Agel model are chosen because they use a Gaussian wake profile and a turbulence model. The wind farm has a finite size in the crosswise direction, unlike as in many other large-eddy simulation wind farm studies, in which the wind farm is effectively infinitely wide due to the cyclic boundary conditions. The results show that new effects like crosswise divergence and convergence occur in such a finite-size multi-gigawatt wind farm. The comparison with the wake models shows that there are large discrepancies of up to 40% between the predicted wind farm power output of the wake models and the large-eddy simulation. An energy budget analysis is made to explain the discrepancies. It shows that the wake models neglect relevant kinetic energy sources and sinks like the geostrophic forcing, the energy input by pressure gradients and energy dissipation. Taking some of these sources and sinks into account could improve the accuracy of the wake models.