State-of-the-art industrial flow predictions rely heavily on Reynolds-Averaged Navier-Stokes (RANS) solvers, however, there is a need for reliable and time-resolved predictions of unsteady flow phenomena. Unsteady phenomena can be efficiently resolved with implicit Large-Eddy Simulations (iLES). However, the main difficulty in industrial applications arises from complex geometries, which require reasonable meshes near the wall, and at the same time high Reynolds number flows that, in combination, severely effect the CFL limit [1, 2]. Typically this CFL restriction reduces the robustness and the largest stable time step size for typical Semi-Implicit schemes [3] and, hence, the computational feasibility of iLES for the short turnaround time in industrial design processes. In this work, we employ a spectral hp element method [4] and combine it with a modal tensor basis to leverage it’s efficient kernels for unstructured meshes [5]. Further, we circumvent the CFL constraint with an implicit Velocity-Correction scheme [6]. The scheme is particularly amenable to matrix-free operators because it is linear in the time-step and enables fast and memory-lean operator evaluations. While implicit schemes typically lead to significantly higher computational costs, we employ a matrix-free kernel to remove the need of building matrices and thus greatly reduce the computational overhead. We present strongly improved robustness on industrial problems and the efficiency of the matrix-free kernels on large-scale unstructured meshes in comparison to current Semi-Implicit schemes.