Summary
NVIDIA has launched Minerva, an AIOps platform that connects infrastructure and workload telemetry across AI factories. The company says it uses machine learning to detect anomalies, group related alerts and help operators investigate incidents.
NVIDIA announced Minerva on October 22, 2026, describing it as an AIOps platform for monitoring and operating AI factories. It combines telemetry from infrastructure and AI workloads, then uses machine-learning models to help teams spot problems and investigate their causes.
AI factories bring together large numbers of GPUs, high-speed networks, storage and software that schedules and runs jobs. A fault or slowdown in one component can affect a larger training run or AI service. NVIDIA says Minerva is designed to give operations teams a joined-up view of these systems rather than separate dashboards for compute, networking and machine-learning workloads.
How Minerva connects factory signals
The platform builds a dynamic knowledge graph that maps infrastructure—including GPUs, SuperNICs, switches and cables—to the jobs running on it. This topology provides context for tracing a slow workload to a component that may be affecting it, including in clusters NVIDIA says can contain hundreds of thousands of GPUs.
NVIDIA says Minerva analyses billions of data points per hour using supervised and unsupervised machine learning. The models learn patterns of normal operation, detect anomalies, anticipate emerging failures and correlate events using the factory’s topology. Related alerts can be grouped into a single incident with a likely root cause, helping teams focus on a smaller number of actionable problems.
The platform also includes an AI assistant that operators can query in natural language. NVIDIA says its responses are grounded in live telemetry and can include recommended remediation. Teams can apply fixes themselves or connect Minerva to automation through infrastructure APIs and external systems.
A focus on AI factory yield
NVIDIA frames Minerva around AI factory “yield”: useful work produced by the infrastructure, measured through GPU hours, goodput, cluster uptime and token production. The company says the platform is intended to reduce interruptions, speed repairs and filter out less useful alerts. Those are operational goals; the announcement provides no quantified before-and-after results for uptime, repair times or yield.
IREN is among the organisations NVIDIA says is already using Minerva. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy it. Dell Technologies, Lenovo and Supermicro will offer the platform as part of their AI factory operations portfolios.
By bringing workload and infrastructure signals into one view, Minerva targets a practical challenge of large-scale AI operations: identifying which component is affecting a job before the issue becomes a broader interruption. Its value will depend on how well that shared view and incident guidance work in the varied environments where organisations deploy it.