NVIDIA launches Minerva AIOps platform for AI factory observability

NVIDIA unveiled Minerva, an AIOps platform that monitors AI factory infrastructure by consolidating telemetry from GPUs, networking, storage and workloads. It builds a dynamic knowledge graph and collapses event storms into a single incident with a likely root cause. In internal tests, Minerva cut NVMe power-fault repair time from days to six hours.

• Covers GPUs, ConnectX SuperNICs, BlueField DPUs, NVLink, Spectrum-X and schedulers
• Analyzes billions of data points per hour and merged 81 events into one alert
• Ships with native assistant Sherlock and an MCP server for natural-language triage
• IREN, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy it