NVIDIA launches Minerva to trace AI data-center slowdowns to their source

NVIDIA launched Minerva on October 22, 2026, to help operators connect a slow AI job with the infrastructure problem behind it. The platform combines equipment and workload monitoring, applying AI to find faults earlier. IREN is already using it; NVIDIA says several other organizations, including IBM and Tesla, plan to follow.

The problem is not limited to equipment that stops working altogether. NVIDIA describes subtler warning signs: a GPU running hotter than nearby processors, or a storage drive losing performance under load. In large clusters, one degraded component can slow an entire job, making the connection between hardware health and useful computing output important.

From separate dashboards to connected faults

NVIDIA’s contrast is with monitoring tools that give networking, computing, reliability and machine-learning teams separate views. Minerva collects telemetry—operating measurements—from GPUs, networks, storage, job schedulers and AI workloads. Its purpose is to connect those signals rather than leave each team investigating only its own part of the system.

The platform builds a dynamic knowledge graph: a map linking equipment such as GPUs, network adapters, switches and cables with the jobs running on it. NVIDIA says that context lets operators trace a slow training run to an affected component, even in a facility containing hundreds of thousands of GPUs.

NVIDIA says Minerva analyzes billions of data points per hour using machine learning. It learns normal behavior, detects unfamiliar anomalies and predicts emerging failures. The infrastructure map also helps it connect related events. Instead of forwarding every symptom separately, the platform can group a flood of alerts into one incident with a likely root cause.

Finding a fault is separate from fixing it

Minerva includes an AI assistant that lets operators ask about cluster health in ordinary language and receive recommended remedies. NVIDIA says the assistant is grounded in live telemetry. The platform can also connect to external AI tools that take actions, extending its role beyond a conversational interface.

The stated objective is AI-factory yield: the useful work a facility produces, tracked through measures including GPU hours, uptime and token production. NVIDIA pitches earlier warnings and faster repairs as ways to improve that output. Its launch announcement does not provide a measured uptime gain or repair-time reduction.

IREN is using it; other deployments are planned

IREN supplies the announcement’s operating example. Its chief technology officer, Denis Skrinnikoff, says Minerva gives the operations team one view across networks, GPUs and running jobs. He says that visibility helps the team focus more on proactive maintenance instead of troubleshooting as the company scales its infrastructure.

NVIDIA’s deployment and distribution list distinguishes organizations planning to use Minerva from suppliers that will offer it to customers:

• Planned deployments: Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla intend to deploy Minerva, according to NVIDIA.
• Supplier portfolios: Dell Technologies, Lenovo and Supermicro will offer Minerva as part of their AI-factory operations portfolios, NVIDIA says.