NVIDIA announced Minerva on October 22, 2026, during GTC Berlin. The new AIOps platform is designed specifically for the large-scale GPU clusters the company calls AI factories. It consolidates telemetry from infrastructure and running workloads, then applies purpose-built AI models to surface problems earlier and maximize useful output measured in GPU hours, goodput, cluster uptime and tokens produced.
The launch follows closely on NVIDIA’s DSX POD announcements for factory-built, containerized AI infrastructure. While those units aim to cut deployment time at prepared sites to as little as 14 days, Minerva addresses the harder, ongoing problem of keeping those dense environments productive once they are online.
What Minerva does
Traditional monitoring tools often split data-center views across networking, compute, site reliability and machine-learning teams. Minerva connects signals across NVIDIA GPUs, SuperNICs, switches, cables, storage, schedulers and the jobs running on them. From those signals it builds a dynamic knowledge graph that maps the physical and logical topology to active workloads.
Operators can use the graph to trace a slow training run or stalled inference job back to a single degraded component, even in clusters containing hundreds of thousands of GPUs. The platform analyzes billions of data points per hour with supervised and unsupervised models to predict hardware failures, detect anomalies and group related alerts into focused incidents with likely root causes.
A native AI assistant and agentic integrations let teams investigate and remediate through natural-language queries grounded in live telemetry. Minerva can run standalone or integrate with NVIDIA DSX OS.
Early results and adopters
In NVIDIA’s internal testing, Minerva reduced mean time to repair for an NVMe power fault from multiple days to six hours. In another case it collapsed 81 noisy events into a single alert, preserving 54 hours of cluster uptime.
IREN is already using the platform. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan deployments. Dell Technologies, Lenovo and Supermicro will offer Minerva as part of their AI factory operations portfolios.
Why it matters
AI factories are capital-intensive. Each megawatt of capacity represents tens of millions of dollars, and every hour of GPU downtime or reduced goodput directly cuts token production and return on investment. As deployments scale from tens of thousands to hundreds of thousands of GPUs, fragmented observability becomes a bottleneck.
By treating infrastructure and workload telemetry as one system and applying AI to turn raw signals into actionable, root-cause-aware insight, Minerva aims to shift operations from reactive firefighting to proactive maintenance. For operators, the practical value is longer intervals between interruptions, faster repairs and higher effective utilization of the most expensive resource in the stack: the GPUs themselves.
Sources
[1] NVIDIA Blog: NVIDIA Unveils Minerva to Bring End-to-End Observability for AI Factories (October 22, 2026)
[2] WebProNews: NVIDIA Built Minerva to Find the One Bad Component Slowing an AI Factory
[3] Bitdeer AI: Bitdeer AI Brings NVIDIA Minerva's AI-Native AIOps
[4] Supermicro: Supermicro Integrates NVIDIA Minerva into SuperCloud Suite
Top comments (0)