NVIDIA Minerva: AIOps for GPU clusters and AI factory observability
NVIDIA Minerva is an AIOps platform for AI factory observability: GPU, network, storage and job telemetry in one graph. What it does and what is unproven.
2 min read
In this article
NVIDIA Minerva launched on 22 October 2026 as an AIOps platform for AI factories. It combines telemetry from GPUs, networking, storage, schedulers and workloads, then applies purpose-built AI models to find problems earlier and keep clusters productive.
NVIDIA gives no independent benchmarks or public pricing. Its benefits, such as faster repair and fewer interruptions, are vendor claims for now.
Sources checked on 26 October 2026. Features, availability and vendor figures can change; confirm the linked primary sources before acting.
What NVIDIA Minerva does
Minerva builds a dynamic knowledge graph of the infrastructure: GPUs, SuperNICs, switches, cables and the jobs running on them. That lets operators trace a slow job to the component causing it.
It uses supervised and unsupervised machine learning to learn normal behaviour, detect anomalies and predict failures. A native assistant answers questions in natural language, grounded in live telemetry. Fixes can be applied by hand or automated through infrastructure APIs.
AI factory yield as the target metric
NVIDIA frames success as "AI factory yield": the useful work the infrastructure produces. It tracks GPU hours, goodput, cluster uptime and token production.
That is a useful framing even without Minerva. Utilisation alone hides failed and restarted jobs; goodput measures work that actually finished.
Root cause in a large GPU cluster
A single failing cable can trigger hundreds of alerts across GPUs and jobs. Minerva groups such an event storm into one incident with a likely root cause.
The value depends on how complete the topology map is. Check what it ingests from non-NVIDIA storage, schedulers and switches before relying on it.
Who is adopting it
IREN already uses Minerva. Aolani, Bitdeer AI, IBM, Jio, Lambda, Quake AI and Tesla plan to deploy it, and Dell, Lenovo and Supermicro will offer it in their AI factory portfolios.
| Question for a pilot | Why it matters |
|---|---|
| Which telemetry does it ingest? | Gaps hide root causes |
| How is it priced? | Not published yet |
| Can fixes be automated safely? | Needs approval and rollback |
| Does it improve goodput? | Measure before and after |
Our verdict: worth evaluating for large NVIDIA clusters, judged on your own goodput and time-to-repair. For AI infrastructure monitoring, see our cloud setup service, and read our cloud cost control checklist.
Sources and further reading
Image credits
Explanatory diagrams: Cloudsoft Zone, CC BY-SA 4.0. Featured photograph: Pokiiri, CC BY-SA 4.0, via Wikimedia Commons. GTC Taiwan 2016: NVIDIA Taiwan, CC BY 2.0, via Wikimedia Commons.
Frequently asked questions
What is NVIDIA Minerva?
An AIOps platform, launched on 22 October 2026, that combines AI factory telemetry with AI models to detect, explain and help fix problems.
How much does NVIDIA Minerva cost?
NVIDIA did not publish pricing in the announcement.
Are Minerva’s performance benefits proven?
Not independently. NVIDIA’s claims about faster repair and fewer interruptions are qualitative; measure goodput and repair time in a pilot.
Your next customer could start right here.
Introduce your tool, product or business to readers exploring technology.
Explore placements and request a private quote
Sources
- NVIDIA unveils Minerva to bring end-to-end observability for AI factories — NVIDIA, 2026-10-22
Written with substantial AI assistance. Facts checked against cited sources. Editorial policy.
Comments
No comments yet. Be the first to join the conversation.
Leave a comment
Every comment is read by a moderator before it appears. Your email address is never published; it is used only to contact you about this comment.
More from Cloudsoft Zone
Industry updateAI and Automation
WebMCP ecommerce: nearly 100,000 stores become AI-agent ready
WebMCP ecommerce at scale: Cloudflare and Commerce.com enabled agent shopping tools on almost 100,000 storefronts. The architecture and the checkout rule.
By Cloudsoft Zone
Industry updateAI and Automation
AI model containment: lessons from OpenAI’s DNS egress incident
AI model containment lessons from OpenAI: an internal model reached the internet through a DNS resolver misconfiguration. How to design agent sandboxes.
By Cloudsoft Zone
Industry updateAI and Automation
12-hour prompt caching: when OpenAI’s longer cache saves money
OpenAI’s 12-hour prompt caching costs 2.5x input to write and 0.1x to read. Break-even maths, when it beats the 30-minute default and how to enable it.
By Cloudsoft Zone