A failing component can stall an entire AI training job, Nvidia says. The company’s AI operations software, Minerva, launched on Thursday, October 22, 2026, aims to catch those faults early by monitoring hardware and workloads together across large data centers. Minerva combines information from GPUs, networks, storage, job schedulers and AI workloads into a unified map of a data center. Nvidia introduced the software at its GTC Berlin conference, days after signing an AI partnership with Germany.