Skip to content
Neural Daily

ToolsNews

NVIDIA Minerva brings AIOps to AI factories

At GTC Berlin NVIDIA launched Minerva, an AIOps platform that unifies infrastructure and workload telemetry for AI factories. Purpose-built AI models filter alerts, predict failures and point to the likely root cause.

By: Mikola Máté Patrik · · 2 min read

NVIDIA Minerva dashboards with telemetry charts, a GPU health grid and an AI assistant panel in front of data center racks
Image: NVIDIA · Press kit · Official NVIDIA blog key visual · Source

What happened

At GTC Berlin NVIDIA launched Minerva (NVIDIA blog), an AIOps platform, meaning an operations system that applies machine learning to monitoring data. It targets AI factories: clusters of thousands of GPUs, high-speed networks and storage that train and serve models.

What it does

  • One view: it connects signals from GPUs, networking, storage, schedulers and running jobs, which separate teams usually watch in separate tools today.
  • Knowledge graph: it logically maps the infrastructure (GPUs, SuperNICs, switches, cables) and the jobs on it, so a slow training run can be traced to the component behind it, even across hundreds of thousands of GPUs.
  • ML analysis: it processes billions of data points per hour with supervised and unsupervised learning, learns normal behavior, detects previously unseen anomalies and predicts emerging failures.
  • Alert grouping: based on topology it collapses an event storm into a single incident with a likely root cause.
  • AI assistant: a built-in natural-language interface grounded in live telemetry answers questions about cluster health, recommends fixes and connects to external agentic tools.
  • Automation: fixes can stay human-in-the-loop or be wired to automation through infrastructure APIs.

The platform optimizes for what NVIDIA calls AI factory yield, combining GPU hours, goodput (the work that is actually useful), cluster uptime and token production. NVIDIA gave no concrete improvement figures, only that time between interruptions gets longer and repairs get faster.

Who uses it

IREN already uses Minerva. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy it, while Dell, Lenovo and Supermicro will include it in their AI factory operations portfolios.

Why it matters

At this scale one GPU running hot or one NVMe drive slowing down can hold back an entire training job, and traditional monitoring spreads that across separate dashboards and a flood of alerts. Minerva promises to turn it into a single actionable item with a cause attached. For teams running their own GPU clusters that could mean fewer lost GPU hours, and for cloud tenants more stable capacity.

Availability

NVIDIA did not announce pricing or a general availability date. Details are on the Minerva product page, and Dell, Lenovo and Supermicro will also offer it.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles