NVIDIA Unveils Minerva to Bring End-to-End Observability for AI Factories

NVIDIA announced Minerva on October 22, 2026, describing it as an AIOps platform that helps operators of large GPU installations catch problems sooner and keep their AI factories productive. The company says the software combines telemetry from GPUs, networking, storage, schedulers and workloads, addressing traditional tools that give compute, networking, reliability and machine learning teams separate views.

According to NVIDIA, Minerva builds a dynamic knowledge graph mapping hardware such as GPUs, SuperNICs, switches and cables to the jobs running on them, allowing a slow training run to be traced to a faulty component even across hundreds of thousands of GPUs. It processes billions of data points per hour with supervised and unsupervised machine learning to learn normal behavior, flag new anomalies, predict failures and collapse floods of alerts into a single incident with a likely root cause. A built-in AI assistant and links to external agentic tools let operators query cluster health in natural language, then apply recommended fixes manually or through automation.

The pitch centers on AI factory yield, which NVIDIA measures through GPU hours, goodput, cluster uptime and token production. The company promises longer stretches between interruptions and faster repairs but gives no specific performance figures.

IREN is already using Minerva, and its chief technology officer, Denis Skrinnikoff, said it gives his team a single view that favors proactive maintenance over troubleshooting. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan deployments, while Dell Technologies, Lenovo and Supermicro will offer it within their AI factory operations portfolios.