NVIDIA Launches Minerva to Deliver End-to-End Observability for AI Factories
NVIDIA Minerva unifies GPU, network, storage, scheduler, and workload telemetry to detect issues early and improve AI factory uptime.
Key Points
- NVIDIA launched Minerva, an AIOps platform for AI factories that unifies telemetry from GPUs, networking, storage, schedulers, and AI workloads into one operational view.
- Minerva builds a dynamic knowledge graph of infrastructure components and jobs, using machine learning to detect anomalies, predict failures, and correlate related events across large clusters.
- The platform includes a native AI assistant and supports external agentic AI tools, enabling natural-language investigation, suggested fixes, and optional automation while keeping human approval in the loop.
- Minerva targets AI factory yield and GPU uptime by helping teams prevent interruptions, resolve incidents faster, and reduce alert noise from fragmented monitoring systems.
Meta description: NVIDIA Minerva unifies telemetry from GPUs, networking, storage, schedulers and AI workloads to help AI factories find problems earlier and stay productive.
Tags: NVIDIA Minerva, AIOps, AI factories, GPU monitoring, agentic AI
Featured image: An abstract illustration of glowing server racks and GPU nodes connected by soft network lines, forming a friendly map of a computing cluster without logos or text.
NVIDIA has launched Minerva, an AIOps platform designed for teams that operate AI factories, the large computing environments that train and serve AI models. For these teams, the value proposition is practical: bring infrastructure and workload signals into one operational view, then use purpose-built AI models to turn that data into insights operators can act on before small issues become major disruptions.
Why AI factories need a single operational view
AI factories are among the most complex operations in the world. They rely on vast numbers of GPUs working across high-speed networks to train models and deliver AI applications. At that scale, one underperforming part can slow an entire job. The earliest warning signs can be subtle: a GPU running hotter than comparable GPUs, or an NVMe drive losing performance under load.
Traditional monitoring tools often split this picture into separate views. Networking teams, compute teams, site reliability engineers and machine learning teams may each see only a slice of the operation. Minerva is designed to connect signals across NVIDIA GPUs, networking, storage, schedulers and workloads, so operators can see the infrastructure and the jobs running on it together.
How Minerva maps the AI factory
The platform builds a dynamic knowledge graph that logically maps the AI factory’s components — GPUs, SuperNICs, switches and cables — along with the workloads they support. That context helps operators trace a slow training job to the component affecting it, even across a facility with hundreds of thousands of GPUs.
Minerva analyzes billions of data points per hour using supervised and unsupervised machine learning. It learns normal behavior, detects anomalies it has not seen before, predicts emerging failures and correlates related events based on the factory’s topology. Instead of presenting operators with every symptom, it can group large bursts of related events into a focused incident with a probable underlying cause.
What operators can do with Minerva
Minerva includes a native AI assistant and connects to external agentic AI tools. This creates a natural-language interface grounded in live telemetry, allowing operators to investigate cluster health and receive suggested fixes through conversation.
Teams can keep a human approval step in the loop and implement changes manually, or connect Minerva to automation through infrastructure APIs and external systems. The result is a workflow that can move from reacting to failures toward preventing them, while reducing the number of alerts operators have to sort through.
The metrics Minerva targets
Minerva focuses on AI factory yield, a measure of the useful work infrastructure produces through GPU hours, goodput, cluster uptime and token production. By correlating workload and infrastructure issues, the platform aims to help operators maximize GPU uptime.
NVIDIA describes the benefits in three areas:
- Longer time between interruptions: Minerva is built to identify trouble before it escalates, so jobs can run longer without incident.
- Faster time to repair: When an incident occurs, Minerva helps narrow down the problem, leading to a significantly quicker resolution.
- Greater insight: Modern AI infrastructure generates a massive volume of data; Minerva filters that data to surface the small set of alerts that deserve attention.
IREN is among the NVIDIA partners using Minerva to optimize its AI factories. IREN chief technology officer Denis Skrinnikoff said the company’s focus as it scales AI factories is maximizing GPU productivity. He said Minerva gives IREN’s operations team one view across the network, the GPUs and the jobs running on them, allowing the team to prioritize proactive maintenance instead of troubleshooting. That visibility, he added, strengthens the operating model for the infrastructure IREN is building next.
Who is deploying Minerva
Joining IREN, Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy Minerva. Dell Technologies, Lenovo and Supermicro will also offer Minerva as part of their AI factory operations portfolios.
The next phase will show how quickly these deployments turn Minerva’s unified telemetry, incident grouping and AI-assisted investigation into measurable changes in AI factory operations: fewer interrupted jobs, faster repairs and less time spent chasing alert noise across separate monitoring tools.
Comments (0)