NVIDIA Minerva Tackles AI Factory Downtime With End-to-End Observability and Predictive Analytics

NVIDIA unveiled Minerva at GTC Berlin, an AIOps platform designed to maximize uptime and productivity in AI factories by delivering end-to-end observability across infrastructure and AI workloads1

1

2

2

. The platform addresses a critical challenge in large-scale AI infrastructure: detecting and resolving issues before a single degraded component slows down entire training jobs across hundreds of thousands of GPUs1

1

.

Connecting Infrastructure and Workload Telemetry

Minerva breaks down the traditional siloed approach to data center monitoring by unifying telemetry from NVIDIA GPUs, networking equipment, storage systems, schedulers, and AI workloads into a single view1

1

. The platform builds a dynamic knowledge graph that logically maps both the physical infrastructure—including GPUs, SuperNICs, switches, and cables—and the jobs running on it1

1

. This integrated approach allows operators to trace a slow training job back to the specific component causing the problem, even in AI factories with hundreds of thousands of GPUs1

1

2

2

.

Machine Learning-Powered Anomaly Detection and Root Cause Analysis

The platform analyzes billions of data points per hour using supervised and unsupervised machine learning models1

1

. Minerva learns normal behavior patterns, detects previously unseen anomalies, and predicts emerging failures by correlating related events based on the factory's topology1

1

. Instead of overwhelming operators with every symptom, it groups event storms into focused incidents with likely root cause analysis1

1

2

2

. Warning signs like a GPU running hotter than its peers or an NVMe drive losing performance under load—signals that might be easy to miss—become actionable insights1

1

.

AI Assistant Enables Natural Language Operations

Minerva includes a native AI assistant and connects to external agentic AI tools, adding a natural-language interface grounded in live telemetry1

1

. Operators can investigate cluster health and receive recommended remediation through conversation1

1

2

2

. Teams maintain control by implementing fixes manually or connecting Minerva to automation through infrastructure application programming interfaces and external systems1

1

.

Measuring Impact on AI Factory Yield

The platform focuses on AI factory yield—a measure of useful work infrastructure produces through GPU hours, goodput, cluster uptime, and token production1

1

. Minerva delivers measurable improvements across key metrics: longer time between interruptions by spotting trouble before it happens, faster time to repair by pinpointing problems quickly, and greater insight by filtering the flood of data down to alerts that actually matter1

1

.

Early Adoption Across Industry Leaders

IREN, where NVIDIA can invest up to $2.1B under a May deal, is among the early adopters using Minerva to optimize its AI factories1

1

2

2

. Denis Skrinnikoff, chief technology officer of IREN, stated that Minerva gives operations teams one view across the network, the GPUs, and the jobs running on them, enabling more proactive maintenance rather than troubleshooting1

1

2

2

. Tesla, IBM, Lambda, Bitdeer AI, Jio, Quake AI, and Aolani plan to deploy the platform1

1

2

2

. Dell Technologies, Lenovo, and Supermicro will offer Minerva as part of their AI factory operations portfolios1

1

2

2

.

Implications for AI Operations Software

The launch signals NVIDIA's expansion beyond selling chips to providing comprehensive AI operations software for companies building AI data centers2

2

. By enabling proactive maintenance instead of reactive troubleshooting, Minerva addresses the operational complexity that comes with scaling AI infrastructure. The platform's ability to maximize GPU uptime directly impacts the economics of running AI factories, where every hour of downtime represents significant lost compute capacity. As AI factories continue to grow in scale and complexity, tools that can predict failures in AI data centers and deliver incident resolution through intelligent automation will become increasingly critical for operators managing these massive systems.