Nvidia launched Minerva on Thursday, a new AI operations platform designed to monitor and predict failures in large-scale AI data centers. Announced at GTC Berlin, the software aggregates telemetry from GPUs, networks, storage systems, and workloads into a unified view, aiming to reduce the time operators spend diagnosing issues in clusters containing hundreds of thousands of accelerators.

Solving the Siloed Monitoring Problem

In massive GPU clusters, a single faulty componentβ€”such as an overheating GPU or a lagging storage driveβ€”can stall entire training jobs. Traditionally, network, compute, and ML teams monitor separate dashboards, making it difficult to trace a slow job back to its root cause across distributed systems. Minerva uses machine learning to establish baseline performance norms and predict anomalies, grouping related alerts into single incidents with identified probable causes rather than flooding operators with disparate warnings. The platform includes a built-in AI assistant that allows staff to query cluster health using natural language and receive suggested remediation steps. These fixes can be applied manually or integrated directly into existing automation pipelines. Nvidia positions Minerva as part of a broader strategy to offer software and infrastructure management tools alongside its hardware, reinforcing its ecosystem lock-in for major AI builders.

Early Adopters and Industry Partners

IREN, a former Bitcoin miner now investing in AI infrastructure, is an early user. "NVIDIA Minerva gives our operations team one view across the network, the GPUs and the jobs running on them," said Denis Skrinnikoff, CTO of IREN. Nvidia has a deal to invest up to $2.1 billion in IREN. Other planned adopters include Bitdeer AI, IBM, Jio, Lambda, and Tesla. Hardware partners like Dell, Lenovo, and Supermicro will also offer Minerva to their customers.

Key Takeaways

  • Minerva correlates data from GPUs, networks, storage, and workloads to predict failures in large AI clusters.
  • The software groups alerts into single incidents and includes an AI assistant for natural language queries.
  • Early adopters include IREN, IBM, Lambda, and Tesla, with Dell, Lenovo, and Supermicro offering it as well.
  • Nvidia did not disclose pricing for Minerva, but the move reinforces its strategy to sell software alongside chips.

The Bottom Line

This launch signals Nvidia's intent to own the operational layer of AI infrastructure, not just the silicon. By providing the tools to keep massive clusters running smoothly, Nvidia deepens the dependency of hyperscalers and AI-native firms on its ecosystem, making it harder for competitors to sell bare hardware without a comparable management stack.