Gaussian Process (GP) regression is a popular tool in various fields, especially in domains that require the confidence of resulting predictions, for example, system identification. Running exact GPs on larger datasets is often impractical due to its cubic runtime performance and quadratic memory use. Existing software in this area is often focused on providing approximate solutions for these big data problems. The GPRat project is an existing C++ implementation for exact GP regression that provides shared-memory parallelism using the popular HPX framework. However, for larger datasets, the memory available on typical desktop or server systems is not enough to fulfill the algorithm’s memory requirements. In this work, we address these limitations by developing distributed approaches for all algorithms in GPRat. This includes hyperparameter optimization and the computation of predictions, either with uncertainty or with a full posterior covariance matrix. We also research and implement different strategies for distributing data and computation tasks over multiple nodes. As part of this work, we also investigate the performance of these implementations in a multi-node cluster and a single-node context with multiple NUMA domains. Our results show that speedups can be achieved in the single-node context but that the high costs of data transmission between nodes significantly reduce the speedup provided by additional computing resources. The distributed approach successfully handles datasets that would be intractable for single nodes, but more work is necessary to reduce communication costs.