Skip to main content

Blog post · November 19, 2026 · aviznetworks.com

Aviz Networks details AVIZ AI NOC for NVIDIA Spectrum-X AI factories

Summary

Aviz Networks describes AVIZ AI NOC, a multi-agent platform that runs on AVIZ ONES and handles AI factory network operations through natural-language requests. Tested in a simulated NVIDIA Spectrum-X environment, it matters to enterprise network leaders managing GPU infrastructure.

Architecture and Lab Scope

The lab extends a prior simulation in NVIDIA DSX Air, where AVIZ ONES handled Day 0 design and deployment of a Spectrum-X fabric. The environment includes separate East-West and North-South fabrics, a dedicated out-of-band management network, and centralized ONES management.

AVIZ AI NOC is introduced for Day 1 and Day 2 operations. The article describes it as using an open-weight gpt-oss-120B model fine-tuned for agentic operations, served through an Aviz-provided LLM endpoint.

Operational Boundaries

ONES remains the authoritative source for fabric inventory, telemetry, and tenant management. AI NOC retrieves data through the ONES connector and sends supported write operations to the Fabric Manager REST API. Tenant modifications require explicit operator confirmation, and long-running GPU allocation operations run asynchronously.

Day 1 Tenant and GPU Management

Operators create the Blue and Red tenants, each with a GPU limit of 8, by making natural-language requests. According to the article, AI NOC checks resources and presents planned changes before applying them, and ONES assigns VNI and VLAN information. The Fabric Manager then pushes EVPN/VXLAN configuration to leaf switches.

Allocating 4 GPUs each to Blue and Red follows the same conversational workflow, and the switch reconfiguration can take several minutes. AI NOC reports operations in progress and lets operators check their status later. Supported lifecycle actions also include listing tenants, inspecting allocations, deallocating GPUs, and deleting tenants after resources are released.

Day 2 Monitoring and Audits

The article describes conversational queries for East-West and North-South fabric health, interface utilization, packet drops, CPU and memory use, and critical alerts. AI NOC can also combine these telemetry queries into one overall health view of Fabric00.

For audits, AI NOC retrieves ONES data and helps generate reports against operator-defined thresholds, organized by device, network domain, severity, and measured value. The article lists threshold categories including utilization, errors, availability, alerts, configuration consistency, and tenant allocation.

Correlation and Troubleshooting

AI NOC correlates inventory, topology, running configuration, and telemetry from ONES so engineers can review several sources of evidence. Examples include identifying switches with interface drops, comparing leaf switch running configurations, and correlating critical alerts with affected interfaces.

In an inference degradation scenario described in the article, AI NOC highlights East-West switches with drops, high utilization, or critical alerts. The article states that engineers validate final root-cause conclusions and any production changes against underlying evidence.

Agentic Operations Compared With Traditional Automation

Traditional automation runs predefined workflows triggered by scripts, templates, or API requests. The article says agentic operations interpret operator intent, select tools, coordinate multiple operations, and build responses from the results. In this model, ONES still provides authoritative network state and executes supported changes.

This Blog Summary is a fact-based summary of the vendor blog. For enterprise decision-makers, the article presents AI NOC as a conversational layer over ONES for Day 1 and Day 2 AI factory operations, with engineers retaining validation of root-cause conclusions.

Our summary. The original keeps its own byline.

Read the original on aviznetworks.com

Companies in this article

Topics