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

NVIDIA has launched Minerva, an AIOps platform designed to help teams identify and resolve issues early on, ensuring the smooth operation of their AI factories. This comprehensive platform brings together data from various sources, including infrastructure and AI workloads, and applies purpose-built AI models to provide timely insights.

AI factories are incredibly complex operations that involve vast numbers of GPUs working in tandem across high-speed networks to train models and serve AI applications. At this scale, even a single degraded component can significantly impact the entire operation, making it challenging for teams to identify potential issues before they become major problems.

The warning signs may be subtle, such as a GPU running hotter than its peers or an NVMe drive losing performance under load. Traditional monitoring tools often divide the data center into separate views, with different teams seeing only part of the operation. Minerva connects these signals across NVIDIA GPUs, networking, storage, schedulers, and workloads to provide a unified view.

Minerva focuses on AI factory yield, which is measured by factors such as useful work infrastructure produces through GPU hours, goodput, cluster uptime, and token production. The platform analyzes billions of data points per hour using both supervised and unsupervised machine learning techniques. It learns normal behavior, detects previously unseen anomalies, predicts emerging failures, and correlates related events based on the factory’s topology.

One of Minerva’s key features is its ability to build a dynamic knowledge graph that logically maps the infrastructure – including GPUs, SuperNICs, switches, and cables – and the jobs running on it. This context helps operators trace slow training jobs to the component affecting them, even in factories with hundreds of thousands of GPUs. Instead of sending every symptom to an operator, Minerva groups related events into a focused incident with a likely root cause.

Minerva comes equipped with a native AI assistant and can connect to external agentic AI tools, providing a natural-language interface grounded in live telemetry. Operators can investigate cluster health and receive recommended remediation through conversation. Teams can keep a person in the loop for manual implementation of fixes or connect Minerva to automation through infrastructure application programming interfaces and external systems.

Minerva’s ability to correlate workload and infrastructure issues along with its agentic AI capabilities simplifies AI factory operations by reducing the number of alerts operators need to address. This allows AIOps teams to move from reactive to proactive maintenance, maximizing GPU uptime in the process. Minerva delivers value for key metrics that AI factories focus on: longer time between interruptions, faster time to repair, and greater insight into their infrastructure.

Several NVIDIA partners are already using Minerva to optimize their AI factories, including IREN. According to Denis Skrinnikoff, chief technology officer of IREN, ‘the key to providing the best service for our customers is maximizing the productivity of our GPUs.’ With Minerva, IREN’s operations team can focus on proactive maintenance rather than troubleshooting, strengthening their operating model for future infrastructure development.

Other companies planning to deploy Minerva include Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI, and Tesla. Additionally, Dell Technologies, Lenovo, and Supermicro will offer Minerva as part of their AI factory operations portfolios.