NVIDIA DSX Air details include Aviz ONES for AI factory networking

NVIDIA used GTC 2026 to position DSX Air and AI factory reference designs around full-stack, multi-tenant systems, and Aviz ONES was presented as the operational layer for validating and deploying AI factory networking across lifecycle stages.

Research Overview

The brief ties NVIDIA’s DSX Air and Vera Rubin DSX AI Factory reference design to an “AI Factory” approach that spans compute, networking, storage, orchestration, security, and operations.

Within that framing, it describes Aviz as providing an operations-focused layer that helps teams validate designs, deploy them, and run networking at scale in a multi-tenant environment.

Key Findings

The integration is presented as shifting integration and troubleshooting earlier by running design and end-to-end simulation before deployment. The workflow described is define, simulate in DSX Air, then move into production using repeatable configuration and validated workflows.

The content states that AI factory failures are more commonly attributed to late integration issues across network traffic paths, orchestration layers, and tenant operations than to incorrect GPU selection.

Technical Breakdown

Aviz ONES in NVIDIA DSX Air is described as enabling design, simulate, and deploy multi-tenant AI factory networking aligned to NVIDIA reference architectures. The brief also says ONES is positioned to validate coverage across front-end and back-end fabrics and the surrounding components used in AI factories.

Operationally, the content divides lifecycle into Day 0 for design and validation, Day 1 for tenant onboarding and fabric provisioning, and Day 2 for telemetry, health, troubleshooting, upgrades, and controlled change management. It further describes observability that tracks 250+ network, server, and GPU metrics, including RoCE telemetry such as PFC, ECN, and queue counters, and states ONES is described as agentless with containerized microservices and YAML templates plus validate/apply/verify workflows.

Operational Impact

The brief frames DSX Air as a way to reduce time to first token by moving integration and troubleshooting into simulation, with timelines described as weeks or months shrinking to days or hours. It also describes Day 0/1/2 operations as “cloud-like,” with integrated NVIDIA AIR simulations.

NVIDIA’s networking and lifecycle framing is supported by a quote from Amit Katz, VP of Networking at NVIDIA: “Aviz ONES adds the operational layer that helps customers turn AI factory networking into a repeatable, multi-tenant platform.”

Conclusion

The update centers on NVIDIA’s DSX Air and AI factory reference designs and places Aviz ONES as the operational layer used to validate, simulate, and deploy AI factory networking across Day 0/1/2 workflows. Blog Signals brief is a fact-based summary of the vendor blog.