Nvidia debuts Minerva AIOps for AI factories

Nvidia (NVDA) has launched Minerva, an AIOps platform built to provide end-to-end observability across what the company calls AI factories. The platform brings together telemetry from infrastructure components and AI workloads, then applies purpose-built AI models to turn this data into actionable insight aimed at increasing uptime and maximizing AI factory yield.

Minerva sits within Nvidia’s broader Vera Rubin platform and is designed to replace fragmented monitoring tools that give networking, compute, site reliability and machine learning teams only partial views of cluster health. By connecting signals from GPUs, networking, storage, schedulers and workloads, it offers operators a unified perspective on complex, large-scale GPU environments.

The platform ingests telemetry from Nvidia GPUs, Spectrum-X Ethernet, NVLink switches, SuperNICs, data processing units, schedulers such as Slurm and Nvidia Run:ai, and the AI workloads themselves. These signals span networking, compute and storage layers, allowing Minerva to correlate workload behavior with underlying infrastructure conditions across entire clusters.

Central to Minerva is a dynamic, topology-aware knowledge graph that maps how GPUs, SuperNICs, switches, cables and jobs relate to one another, both logically and physically. This enables operators to trace issues such as slow training jobs back to specific components, even in AI factories with hundreds of thousands of GPUs.

Minerva processes billions of telemetry data points per hour using supervised and unsupervised machine learning. It learns normal operating patterns, detects previously unseen anomalies, predicts emerging failures and links related events based on the AI factory’s layout. Instead of forwarding every symptom, it groups cascading alerts into focused incidents with a likely root cause.

Minerva includes a native AI assistant and supports connections to external agentic AI tools, creating a natural-language interface grounded in live telemetry. Operators can ask plain-language questions about cluster health, investigate incidents and receive recommended remediation steps without writing queries against raw logs.

Teams can choose to implement fixes manually or connect Minerva to infrastructure application programming interfaces and external automation systems. This flexibility lets organizations keep a human in the loop while still moving from reactive troubleshooting toward more proactive maintenance and operations.

Nvidia frames Minerva’s impact around extending the time between workload interruptions, shortening time to repair and reducing data noise for operators. By spotting trouble before it happens and isolating faults quickly when they occur, the platform is intended to help AI factories maximize GPU uptime and overall productive work.

The platform explicitly targets improvements in AI factory yield, which Nvidia measures through GPU hours, goodput, cluster uptime and token production. By aggregating related problems into single actionable items and filtering out low-value alerts, Minerva aims to give AIOps teams greater insight into these key performance metrics.

IREN is already using Minerva as an early client. Beyond IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy the platform in their AI factory environments, highlighting interest from both hyperscale and enterprise users.

On the systems side, Dell Technologies, Lenovo and Supermicro will offer Minerva as part of their AI factory operations portfolios. By incorporating Minerva into their infrastructure offerings, these vendors extend its availability across a broad range of enterprise AI factory deployments.