We prove quantitative convergence estimates for single layer neural networks in the NTK regime to gaussian processes at positive training time