NVIDIA Minerva Uses AI to Improve AI Factory Productivity and GPU Uptime

NVIDIA Minerva is now available as an AIOps platform designed to help teams identify problems early and keep AI factories productive. By integrating telemetry from infrastructure and AI workloads, Minerva applies purpose-built AI models to turn operational data into timely, actionable insights.

AI factories are among the most complex operations in the world, with vast numbers of GPUs working across high-speed networks to train models and serve AI applications. At this scale, degradation in a single component can slow an entire job. Warning signs may be easy to overlook, including GPUs running hotter than others or NVMe drives performing poorly under load.

The platform focuses on AI factory yield, a measure of the useful working infrastructure produced through GPU time, goodput, cluster uptime and token generation.

A dynamic view of AI factory operations

Traditional monitoring tools often divide data centers into separate operational views. Networking, computing, site reliability and machine learning teams may each focus on different parts of the environment. Minerva connects these signals across NVIDIA GPUs, networking, storage, schedulers and workloads.

Using these signals, Minerva builds a dynamic knowledge graph that maps infrastructure such as GPUs, SuperNICs, switches and cables to the jobs running on it. This context helps operators trace slow training jobs to the components affecting them, even across AI factories with hundreds of thousands of GPUs.

Minerva can analyze billions of data points per hour using supervised and unsupervised machine learning. It learns normal operating patterns, detects previously invisible anomalies, predicts potential failures and correlates related events based on factory topology. Rather than sending every symptom to operators, the platform groups event storms into focused incidents with a higher likelihood of identifying the root cause.

Minerva includes a native AI assistant and connects to external agentic AI tools. This provides a natural-language interface based on live telemetry, allowing operators to investigate cluster health and receive remediation recommendations through a conversation. Teams can assign fixes to personnel for manual implementation or connect Minerva to automation through infrastructure application programming interfaces and external systems.

How NVIDIA Minerva can improve AI factory productivity

Correlating workloads with infrastructure issues and using agentic AI capabilities can simplify AI factory operations. Instead of dealing with a storm of alerts, Minerva aggregates related issues into a single actionable item. This enables AIOps teams to move from reactive troubleshooting to proactive maintenance while helping maximize GPU uptime.

Minerva provides value across key AI factory performance metrics:

• Longer time between interruptions: Minerva identifies problems before they occur, helping jobs run longer without interruptions.
• Reduced repair time: When an incident occurs, Minerva pinpoints the problem and helps teams resolve it more quickly.
• Greater insight: Modern AI infrastructure generates vast amounts of data. Minerva filters that data to highlight the alerts that matter most.

AI factory companies planning to use Minerva

IREN is one of the NVIDIA partners using Minerva to optimize its AI factories.

“As we scale our AI factory, the key to best serving our customers is maximizing GPU productivity,” said Denis Skrinnikov, IREN Chief Technology Officer. “NVIDIA Minerva gives operations teams a single view across the network, GPUs and the jobs running on them, allowing them to focus on proactive maintenance rather than troubleshooting. This visibility strengthens the operational model for the infrastructure we build next.”

IREN is joined by Aolani, Bitdia AI, IBM, Geo, Lambda, Earthquake AI and Tesla, which are planning to deploy Minerva.

Dell Technologies, Lenovo and Supermicro also plan to offer Minerva as part of their AI factory operations portfolios.

Learn more about NVIDIA Minerva.

See the NVIDIA News terms of service for information about software product news.