NVIDIA introduced Minerva, an observability platform for AI infrastructure. It links information about GPUs, networks, storage, schedulers, and jobs in a knowledge graph to identify anomalies and failure causes, with agents assisting investigations and suggesting responses. IREN is using it, and adoption by other companies was also announced. Training or inference jobs can run slowly even when individual devices appear healthy, so the platform focuses on relationships between components. The public description is a vendor feature announcement; reductions in failures across all operating environments have not been independently demonstrated.