Aviz Details How SONiC Can Support AI Networking Fabrics
Aviz’s vendor post argues that SONiC can serve as an AI networking fabric to reduce lock-in, speed switch adoption for new ASICs, and provide a foundation for telemetry-driven automation in multi-vendor, multi-tenant environments.
Research Overview
The post frames AI cluster buildouts as a driver for rethinking network operating system choices, especially when organizations weigh total cost of ownership for AI workloads.
It presents SONiC as Software for Open Networking in the Cloud and links adoption goals to open operational control, hardware flexibility, and faster paths from proof of concept to deployment.
Key Findings
The post lists three themes for why SONiC is favored: open choices and control intended to avoid vendor lock-in, and cost savings across other CapEx and OpEx items.
It also describes deployment and testing expectations, including end-to-end testing of traffic types and scenarios and compatibility across multiple switch vendors and ASIC ecosystems.
Technical Breakdown
The post says SONiC is intended to be portable to new ASICs and that it is often among the first network operating systems ported to new hardware due to broad usage by hyperscalers.
It describes microservices architecture as a way to support customization and states that SONiC can provide normalized data for use with AI to manage an AI fabric.
Multi-tenancy and networking capabilities
The post states that protocols for multi-tenancy are already in place and cites VRF, VXLAN, and EVPN as mechanisms for workload isolation and segmentation.
For performance targeting, it says SONiC supports speeds up to 800GbE and references RoCE support with features such as Priority Flow Control, ECN, and RoCE-specific optimizations.
Operational Impact
For selection, the post advises enterprises to choose proven open-source-based SONiC solutions, run comprehensive end-to-end testing for their traffic and simulated scenarios, and ensure multi-vendor compatibility to support future hardware changes.
It also describes an evaluation approach tied to Aviz Networks’ ONE Center, which it positions as a facility for real-world, multi-vendor proof of concept testing and comparison of SONiC solutions without initial investment.
Leadership Perspective
The post includes a viewpoint from an Aviz leader, describing how customers react when switching vendors and listing goals that align with SONiC adoption: reducing lock-in, maintaining control, and redirecting savings.
It further states that Aviz works with partners including NVIDIA, Cisco, Edgecore, Wistron, and Celestica and ties this ecosystem approach to RoCE-based deployments using SONiC.
Overall, the post argues that SONiC’s open-source design, multi-vendor ecosystem, multi-tenancy support, and described telemetry and normalization capabilities can support AI networking deployments while organizations assess end-to-end testing and cost-of-ownership tradeoffs; this “Blog Signals brief” is a fact-based summary of the vendor blog.