NVIDIA has launched Minerva, an AI-powered IT operations platform designed to give operators end-to-end visibility across AI factories. The platform combines infrastructure and AI workload telemetry with machine learning to detect anomalies, anticipate failures, correlate incidents, and help teams identify likely root causes before infrastructure problems disrupt production.
Announced on October 22, 2026, Minerva targets a growing challenge for organizations running large GPU clusters: keeping increasingly complex AI infrastructure productive while reducing the time and effort required to diagnose operational problems. NVIDIA says the platform is designed to improve uptime and maximize AI factory yield.
What Is the NVIDIA Minerva AIOps Platform?
Minerva is an AIOps platform, short for artificial intelligence for IT operations. AIOps uses machine learning and automated analysis to interpret operational data, identify abnormal behavior, and help IT teams investigate and resolve problems.
Unlike monitoring systems that focus on individual infrastructure layers, Minerva connects data from computing hardware, networking, storage, schedulers, and AI workloads. Telemetry—the measurements and event records generated by infrastructure and software—provides the operational signals the platform analyzes.
NVIDIA says Minerva can process billions of data points per hour using supervised and unsupervised machine learning. These techniques help the platform recognize known problem patterns, learn normal operating behavior, flag previously unseen anomalies, and identify relationships between events.
The goal is to give operators a unified view of the AI factory rather than requiring them to investigate separate monitoring dashboards for every component.
How Minerva Connects Infrastructure and AI Workloads
Large AI factories bring together thousands of interconnected components. GPUs perform computation, high-speed networks move data between systems, storage supplies datasets and model checkpoints, and schedulers allocate resources to training and inference jobs.
A fault in one component can affect much more than the device itself. For example, a degraded network link could slow communication between GPUs, extending a distributed training job. A storage bottleneck could delay data delivery, leaving expensive accelerators underutilized.
Minerva aims to connect these symptoms with their underlying causes through a dynamic knowledge graph.
A knowledge graph is a structured map of relationships between entities. In Minerva, this graph connects infrastructure components—including GPUs, SuperNICs, switches, cables, and storage—with schedulers and the workloads running on the cluster.
By combining those relationships with live telemetry, the platform can help operators trace a performance problem from an affected workload to a potentially responsible infrastructure component. NVIDIA says this approach is designed to work at AI factory scale, including environments with hundreds of thousands of GPUs.
The system also correlates related alerts. Instead of presenting operators with a flood of individual notifications, it can group connected events into a more focused incident and identify a likely root cause. This is intended to reduce alert fatigue and accelerate troubleshooting.
NVIDIA Minerva AIOps Targets AI Factory Yield
NVIDIA frames Minerva around AI factory yield: how effectively an infrastructure deployment turns its available computing resources into useful AI output.
That concept extends beyond whether a GPU is powered on or a server is reachable. Operators also need to understand whether their infrastructure is completing valuable work efficiently.
NVIDIA identifies several relevant measures:
- GPU hours: The computing capacity available and used over time.
- Goodput: The portion of computing effort that contributes to useful progress, rather than being lost to interruptions, inefficiencies, or repeated work.
- Cluster uptime: The extent to which the infrastructure remains operational and available.
- Token production: The volume of tokens generated by AI models, particularly in inference workloads.
These measures help explain why component-level monitoring is not always sufficient. A cluster can appear operational while delivering less useful work because of degraded networking, inefficient resource allocation, or repeated job interruptions.
By connecting hardware health with workload performance, Minerva is designed to help teams identify these issues and protect productive GPU capacity.
An AI Assistant and Support for External Agents
Minerva also includes a native AI assistant that lets operators investigate infrastructure conditions through natural-language interactions.
Rather than manually correlating logs, metrics, and alerts, an operator can use the assistant to investigate cluster health, examine incidents, and receive recommended remediation steps grounded in live telemetry.
NVIDIA also says Minerva supports integration with external agentic AI tools. Agentic AI refers to systems that can pursue tasks through multiple steps, use tools, and take actions within defined workflows.
This integration creates a path from diagnosis toward operational automation. Teams can retain human oversight and implement suggested fixes manually, or connect the platform to infrastructure APIs and external systems to automate selected responses.
The distinction matters: identifying a likely failure and recommending a fix are not the same as automatically applying that fix. NVIDIA’s announcement describes the platform’s operational capabilities, but it does not establish independently verified reductions in downtime or repair times across customer deployments.
Moving From Reactive Troubleshooting to Proactive Maintenance
Traditional infrastructure operations often begin after an alert, failed job, or performance drop becomes visible. At large AI factories, investigating the cause can be difficult because the initial symptom may appear far from the component responsible.
Minerva is designed to shift this process toward proactive maintenance by identifying emerging anomalies, predicting potential failures, and correlating signals across infrastructure layers.
For operators, the potential benefits include fewer unnecessary alerts, faster incident investigation, longer uninterrupted workloads, and better use of deployed GPUs. These are intended outcomes rather than independently established performance results.
The broader operational challenge is becoming increasingly important as organizations scale AI training and inference. More hardware and more interconnected workloads create additional dependencies, making cross-layer visibility essential to understanding where performance is being lost.
Which Companies Plan to Deploy NVIDIA Minerva?
NVIDIA identifies IREN as an existing Minerva user. IREN CTO Denis Skrinnikoff said the platform gives its operations team visibility across networking, GPUs, and running jobs, supporting a greater focus on proactive maintenance.
NVIDIA also names the following organizations as planning to deploy Minerva:
- Aolani
- Bitdeer AI
- IBM
- Jio
- Lambda
- Quake AI
- Tesla
Separately, Dell Technologies, Lenovo, and Supermicro will offer Minerva as part of their AI factory operations portfolios, according to NVIDIA.
These announcements indicate interest across AI infrastructure operators and hardware suppliers. However, planned deployments should not be interpreted as completed implementations or proof of measured operational improvements.
What NVIDIA Minerva Means for AI Infrastructure Operators
Minerva reflects a shift in AI infrastructure management: monitoring whether hardware is running is no longer enough when the business objective is continuous, efficient AI production.
By combining infrastructure telemetry, workload context, machine learning, and AI-assisted troubleshooting, NVIDIA is positioning Minerva as an operational layer for complex AI factories.
Its practical value will depend on how effectively organizations can integrate their existing systems, act on its diagnoses, and measure improvements in uptime, goodput, and token production. Independent customer results and further deployment details will be important for assessing those outcomes.
For organizations operating large GPU clusters, the central proposition is straightforward: identify infrastructure problems earlier, understand their effects on workloads, and spend less time troubleshooting after production has already been disrupted.
