NVIDIA Unveils Minerva, an AIOps Observability Platform for Mega-Scale AI Factories

NVIDIA has launched Minerva, a dedicated AIOps platform designed to provide end-to-end observability for massive AI infrastructure. Aimed at modern "AI factories" housing up to hundreds of thousands of GPUs, the system connects physical hardware telemetry directly with running AI workloads to identify bottlenecks and keep clusters operating at peak capacity.

Modern AI training runs and inference workloads rely on vast clusters of interconnected processors. In these environments, subtle failures—such as a single GPU running hotter than normal or an NVMe drive degrading under heavy read-write loads—can quietly drag down performance across an entire job. Minerva targets this vulnerability by shifting operations from reactive troubleshooting to proactive remediation.

Unifying Cluster Topology With AI Workloads

Traditional data center monitoring tools often segment operations into distinct silos, leaving machine learning, compute, networking, and site reliability teams with fragmented visibility. Minerva bridges these boundaries by collecting signals across NVIDIA GPUs, SuperNICs, switches, cables, storage systems, schedulers, and workloads.

Using this telemetry, Minerva builds a dynamic knowledge graph that maps physical hardware alongside the specific tasks running across the cluster. The platform processes billions of data points every hour using supervised and unsupervised machine learning algorithms. By learning baseline behavior, detecting novel anomalies, and correlating related events against the cluster's topology, Minerva can compress noisy "event storms" into a single, focused incident with an identified root cause.

Minerva also features an integrated native AI assistant capable of linking to external agentic AI systems. Operators can query cluster health and inspect infrastructure using natural language, receiving contextual remediation advice grounded in live telemetry. Teams can either review suggestions manually or connect Minerva directly to infrastructure automation via APIs.

Maximizing AI Factory Yield

The platform is designed around improving "AI factory yield"—a measurement encompassing cluster uptime, GPU hours, effective throughput (goodput), and token generation. NVIDIA outlines three key operational advantages:

• Longer intervals between interruptions: Predictive failure detection resolves issues before jobs stall.
• Faster time to repair: Pinpointed root causes drastically cut the time spent diagnosing hardware faults.
• Actionable insights: Machine learning models filter through massive data streams to surface only high-priority alerts.

"As we scale our AI factories, the key to providing the best service for our customers is maximizing the productivity of our GPUs," said Denis Skrinnikoff, chief technology officer at data center operator IREN. "NVIDIA Minerva gives our operations team one view across the network, the GPUs and the jobs running on them, so we focus more on proactive maintenance rather than troubleshooting."

Broad Industry Adoption

A wide range of cloud and enterprise computing operators are preparing to integrate the platform. Alongside IREN, companies including Tesla, IBM, Bitdeer AI, Lambda, Jio, Quake AI, and Aolani plan to deploy Minerva.

Enterprise server manufacturers Dell Technologies, Lenovo, and Supermicro will also integrate Minerva directly into their AI factory infrastructure portfolios.