NVIDIA Minerva Brings AI Powered Observability to AI Factories
NVIDIA launches Minerva, an AIOps platform using live telemetry and AI to predict failures, cut alert noise and raise GPU uptime across AI factories at scale.
Summary
NVIDIA launched Minerva on October 22, 2026, an AIOps platform that combines telemetry from AI workloads, GPUs, networking, storage and schedulers to improve AI factory yield, measured through GPU hours, goodput, cluster uptime and token production. Its dynamic knowledge graph maps jobs to GPUs, SuperNICs, switches and cables across factories containing hundreds of thousands of GPUs. Supervised and unsupervised machine learning analyzes billions of data points per hour to learn normal behavior, detect unseen anomalies, predict failures, correlate events by topology and reduce alert storms to incidents with likely root causes. This can expose subtle problems such as an unusually hot GPU or an NVMe drive slowing under load before one degraded component delays a job.
Minerva includes an AI assistant grounded in live telemetry and supports external agentic AI tools, letting operators investigate cluster health, receive remediation recommendations and either apply fixes manually or automate them through infrastructure APIs and external systems. NVIDIA says this should extend time between interruptions, accelerate repairs, reduce alert noise and maximize GPU uptime, although it provided no measured improvement. IREN is using Minerva, and CTO Denis Skrinnikoff said its unified view supports proactive maintenance. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan deployments. Dell Technologies, Lenovo and Supermicro will offer Minerva in their AI factory operations portfolios.
Positives
- Billions of data points per hour can be analyzed to detect unseen anomalies, predict failures and identify likely root causes.
- Hundreds of thousands of GPUs can be mapped with workloads, SuperNICs, switches and cables in one dynamic knowledge graph.
- GPU hours, goodput, cluster uptime and token production give operators defined measures for improving AI factory yield.
- IREN is already using Minerva, while Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan deployments.
- Dell Technologies, Lenovo and Supermicro will add Minerva to their AI factory operations portfolios.
Risks & concerns
- One degraded component can slow an entire AI job across factories containing vast numbers of GPUs.
- Hotter GPUs and NVMe drives losing performance under load can produce subtle warning signs that operators may miss.
- Traditional monitoring separates networking, compute, site reliability and machine learning teams into incomplete operational views.
- NVIDIA provided no quantified evidence for Minerva's promised gains in interruption frequency, repair speed or GPU uptime.
- Pricing and deployment dates for the companies planning to adopt Minerva remain unspecified.