Nvidia unveils Minerva for end-to-end observability in AI factories

Nvidia has introduced Minerva, an AIOps platform built to watch over the sprawling GPU clusters the company likes to call AI factories. The platform was announced on October 22, 2026, at Nvidia’s GTC Berlin event. It landed alongside new DSX POD reference designs for AI factories. What Minerva actually does Minerva consolidates telemetry from GPUs, networking, storage, job scheduling and AI workloads into one unified view. Under the hood, the platform applies both supervised and unsupervised machine learning models. According to the research findings on the launch, it can analyze billions of data points per hour. Those models are designed to predict hardware failures, detect anomalies and correlate events across the stack. The platform also includes a topology-aware knowledge graph. Minerva maps how GPUs, SuperNICs and cables physically and logically connect to one another. When alerts start firing, the system groups them together with their likely root cause. Minerva also ships with a native AI assistant that accepts natural language queries. Operators can ask questions about cluster health in plain language rather than writing queries against raw logs. The metrics Nvidia wants op...