The Pulse
NVIDIA Launches Minerva to Trace AI-Factory Faults
NVIDIA launched Minerva, an AIOps platform that combines infrastructure and workload telemetry to help operators identify problems in AI factories. IREN is already using Minerva, while seven other organizations plan to deploy it.
AI.info Team · Written with AI · fact-checked ·
Minerva links the signals separate teams see
NVIDIA launched Minerva on October 22, pitching the AIOps platform as a way to find problems in AI factories before they disrupt work. The operational challenge is that infrastructure and workloads generate connected warning signs, while teams responsible for networking, compute, site reliability and machine learning often monitor separate systems. A GPU running hotter than its peers or an NVMe drive slowing under load can affect a job without making the cause obvious from any one view.
Minerva pulls telemetry from NVIDIA GPUs, networking, storage, schedulers and AI workloads into a shared picture. NVIDIA says the platform builds a knowledge graph mapping components—including GPUs, SuperNICs, switches and cables—to the jobs running on them. Operators can use that context to trace a slow training run to a component, including in facilities with hundreds of thousands of GPUs.
Billions of data points feed its incident view
NVIDIA says Minerva analyzes billions of data points per hour with supervised and unsupervised machine learning. The system learns typical behavior, detects anomalies, predicts possible failures and connects related events using the factory’s topology. Rather than leave operators to sort through a surge of individual alerts, it can group related symptoms into an incident with a likely root cause.
The platform also includes a conversational AI assistant grounded in live telemetry and can connect to external agentic AI tools. Operators can ask about cluster health and receive recommended fixes. Teams can apply those changes themselves or connect Minerva to automation through infrastructure APIs and external systems, according to NVIDIA.
IREN backs the single-view pitch
IREN is among the organizations already using Minerva. “NVIDIA Minerva gives our operations team one view across the network, the GPUs and the jobs running on them, so we focus more on proactive maintenance rather than troubleshooting,” said Denis Skrinnikoff, IREN’s chief technology officer. His description reflects the product’s central claim: connect system-level signals so teams can spend less time identifying where a problem began.
Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla also plan to deploy Minerva. Dell Technologies, Lenovo and Supermicro plan to offer the platform as part of their AI-factory operations portfolios. NVIDIA’s announcement does not give deployment dates for those organizations.
NVIDIA gives no measured uptime gains
The company says Minerva is designed to lengthen the time between interruptions, speed repairs and filter out alerts that matter less. But its launch announcement does not provide benchmark results, customer performance figures or quantified improvements in uptime or repair time. The distinction matters: the announcement describes intended benefits and partner plans, not measured results across deployed systems.
For now, the concrete proposition is an operations tool that correlates telemetry across hardware and jobs, with human review or automated remediation as options. Its value will depend on whether that unified view helps operators identify faults and restore work faster in live AI factories.