European Investor
22 Oct 2026, 15:51
NVIDIA Unveils Minerva AIOps Platform to Improve AI Factory Uptime and Yield
NVIDIA introduced Minerva, a new AIOps platform designed to give operators end-to-end observability across large-scale AI factories and help maximize GPU productivity.
Minerva combines telemetry from GPUs, networking, storage, schedulers and AI workloads into a unified operational view. NVIDIA said the platform builds a dynamic knowledge graph that maps infrastructure components such as GPUs, SuperNICs, switches and cables together with the workloads running across them.
The platform uses supervised and unsupervised machine learning to analyze billions of data points per hour, detect anomalies, predict emerging failures and correlate related events. Rather than surfacing every individual alert, Minerva is designed to group related symptoms into a smaller number of actionable incidents with likely root causes.
NVIDIA is positioning the platform around “AI factory yield,” which includes metrics such as GPU hours, goodput, cluster uptime and token production. The company said Minerva can help extend time between interruptions, reduce repair times and filter large volumes of telemetry into more meaningful operational signals.
Minerva also includes a native AI assistant and can connect with external agentic AI tools. This allows operators to investigate cluster health through natural-language queries and receive recommended remediation steps, while still keeping humans in the loop or connecting actions to automated infrastructure systems.
IREN is already using Minerva, while Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy the platform. Dell Technologies, Lenovo and Supermicro are also expected to offer Minerva as part of their AI factory operations portfolios.
The launch expands NVIDIA’s push beyond chips and systems into the software layer used to operate increasingly complex AI infrastructure, giving the company another way to deepen its role across the AI data-center stack.
NVIDIA introduced Minerva, a new AIOps platform designed to give operators end-to-end observability across large-scale AI factories and help maximize GPU productivity.
Minerva combines telemetry from GPUs, networking, storage, schedulers and AI workloads into a unified operational view. NVIDIA said the platform builds a dynamic knowledge graph that maps infrastructure components such as GPUs, SuperNICs, switches and cables together with the workloads running across them.
The platform uses supervised and unsupervised machine learning to analyze billions of data points per hour, detect anomalies, predict emerging failures and correlate related events. Rather than surfacing every individual alert, Minerva is designed to group related symptoms into a smaller number of actionable incidents with likely root causes.
NVIDIA is positioning the platform around “AI factory yield,” which includes metrics such as GPU hours, goodput, cluster uptime and token production. The company said Minerva can help extend time between interruptions, reduce repair times and filter large volumes of telemetry into more meaningful operational signals.
Minerva also includes a native AI assistant and can connect with external agentic AI tools. This allows operators to investigate cluster health through natural-language queries and receive recommended remediation steps, while still keeping humans in the loop or connecting actions to automated infrastructure systems.
IREN is already using Minerva, while Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy the platform. Dell Technologies, Lenovo and Supermicro are also expected to offer Minerva as part of their AI factory operations portfolios.
The launch expands NVIDIA’s push beyond chips and systems into the software layer used to operate increasingly complex AI infrastructure, giving the company another way to deepen its role across the AI data-center stack.
