NVIDIA launches Minerva AIOps platform for AI factory observability

NVIDIA launches Minerva, an AIOps platform that combines telemetry from GPUs, networking, storage, schedulers and workloads to find problems earlier in AI factories, according to NVIDIA's blog.

Minerva builds a knowledge graph of the infrastructure and jobs, groups related alerts into focused incidents with likely root causes, and includes a native AI assistant that connects to external agentic AI tools.

NVIDIA says IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy Minerva. Dell Technologies, Lenovo and Supermicro will offer it as part of their AI factory operations portfolios.

A later NVIDIA technical blog post describes Minerva as a way to turn telemetry from GPUs, Spectrum-X Ethernet fabrics, NVLink scale-up fabrics and schedulers into GPU yield gains. The post says these layers run as one production system even when different teams own them. It gives no performance results or availability. A separate NVIDIA technical post says NVIDIA built an AI Factory Operations Agent to run its own internal DGX SuperPOD, where engineers previously checked compute, network, hardware and scheduling dashboards by hand to find causes of training slowdowns. The agent lets a team of AI agents correlate those signals through one interface.