NVIDIA Launches Minerva to Bring End-to-End Observability to AI Factories

NVIDIA has launched Minerva, a new AIOps platform designed to unify infrastructure and workload telemetry for large-scale AI data centers. Released today, the platform helps operators identify problems early and maximize AI factory yield by analyzing billions of data points per hour across compute, networking, storage, and active machine learning jobs.

Modern AI infrastructure often segregates monitoring data across separate teams, making subtle hardware degradations—such as an overheating GPU or an underperforming NVMe drive—difficult to isolate before they disrupt an entire training run. Minerva addresses this by building a dynamic knowledge graph that maps the physical infrastructure, including GPUs, SuperNICs, switches, and cables, directly to the workloads running on top of them.

Unified Telemetry and Root-Cause Analysis

By applying both supervised and unsupervised machine learning, Minerva learns baseline behavior, detects novel anomalies, and predicts emerging component failures. Instead of overwhelming operators with disparate warning signs, the platform correlates related events based on factory topology to distill a massive event storm into a single focused incident with a probable root cause.

The software also integrates a native AI assistant alongside support for external agentic AI tools. This allows operators to query cluster health using a natural-language interface grounded in live telemetry data. Teams can choose to keep a human in the loop for manual remediation or connect the platform to automated systems via infrastructure APIs.

Yield, Availability, and Industry Adoption

Minerva targets key operational metrics known collectively as AI factory yield, which measures useful work produced through GPU hours, cluster uptime, goodput, and token generation. According to NVIDIA, the platform extends the time between interruptions, accelerates repair times by pinpointing exact component faults, and filters out non-critical alerts.

Initial deployments and planned integrations span several enterprise and cloud providers. IREN is currently utilizing the platform, while Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla plan to deploy it. Additionally, hardware vendors Dell Technologies, Lenovo, and Supermicro intend to offer Minerva as part of their respective AI factory operations portfolios.

Source

Official announcement or documentation