NVIDIA Launches Minerva AIOps Platform for End‑to‑End AI Factory Observability

NVIDIA has introduced Minerva, an AIOps platform designed to give operators a single, dynamic view of an AI factory that can span hundreds of thousands of GPUs. By ingesting telemetry from GPUs, SuperNICs, switches, storage, schedulers and the workloads themselves, Minerva builds a live knowledge graph that maps every component and job in real time.

How Minerva Works

The platform applies both supervised and unsupervised machine‑learning models to billions of data points per hour. It learns normal behavior, flags previously unseen anomalies, predicts emerging failures, and correlates related events using the factory’s topology. Instead of flooding operators with raw alerts, Minerva groups an event storm into a focused incident with a likely root cause.

A native AI assistant provides a natural‑language interface grounded in live telemetry. Operators can query cluster health, receive remediation recommendations, and either act manually or hand off to automation via infrastructure APIs and external agentic tools.

Key Benefits for AI Factories

• Longer mean time between interruptions – early detection keeps training and inference jobs running without pause.
• Faster mean time to repair – precise root‑cause identification cuts troubleshooting time dramatically.
• Reduced alert fatigue – only the handful of actionable alerts surface, letting teams shift from reactive to proactive maintenance.

Why it matters for GPU / AI infrastructure: As GPU clusters grow, the cost of a single degraded component multiplies across the entire workload. Minerva’s end‑to‑end observability directly protects GPU uptime, improves goodput, and maximizes token production — metrics that define the economic viability of any large‑scale AI deployment, including those hosted on platforms like AiGpu.