The emergence and rapid maturation of a new class of weather forecast models built using artificial intelligence (AI) and machine learning (ML) approaches in recent years is generating substantial interest within and beyond the weather enterprise. One factor driving this interest is that these models – which include Google DeepMind’s GraphCast, NVIDIA’s FourCastNet, and Microsoft’s Aurora, among many others – appear to meet or exceed the skill of numerical weather forecast models when using standard metrics such as the 500mb anomaly correlation for medium-range forecast applications. The potential that this new approach to weather modeling could further push the barriers of forecast accuracy garners much public attention, but the opportunities they afford are much greater. For example, a trained “AI-NWP” model can be used to generate a forecast similar to the operational, deterministic GFS in just a few minutes, but without the need for a massive supercomputer. Such computational efficiency opens the door to massive ensembles for uncertainty quantification and probabilistic forecasting, among many other applications.