Enterprise

Aviz and Spectro Cloud describe a governed AI Factory stack with GPU-aware networking, Kubernetes fleet management, and Day-0 to Day-2 automation.

ONES 4.2 adds zero-touch Day-0 bootstrapping, expanded controller and SSD telemetry, modular licensing, multi-tenant GPU sharing, and API-driven networking.

The blog argues that AI depends on networking, outlining scale-up, scale-out, open architectures, and vendor-agnostic planning to control AI cost per token.

Network Copilot Dashboard Analytics consolidates users, connectors, queries, file operations, and storage into one view with time-range filters.

Aviz Podcast Episode 21 says AI value forms at the edge and that networking is needed to scale AI infrastructure beyond GPUs.

Aviz will present at ONUG Dallas on open networking and SONiC, using eBay’s 100G-to-400G shift to explain bandwidth scaling and NetOps.

Podcast discussion on building AI-era networking: moving from one fabric to multiple networks for AI factories, with openness and standardization to manage complexity and cost.

The post says network standardization can cut AI-serving costs by improving GPU utilization, job completion time, and cluster predictability, citing tokens-per-watt-per-dollar.

Aviz ONES 4.2 lets multiple tenants share compute on one server while GPUs stay bound to a single tenant via device-level allocation and EVPN VXLAN/VRF.

Aviz Networks with Red Hat outlines ONES orchestration, ASN packet visibility, and Network Copilot for AI factory operations on RHEL and OpenShift.

Aviz addresses enterprise SONiC objections with validation, observability, orchestration, SLAs, and training via ONES and CCS.

Alan Weckel of 650 Group projects AI networking could exceed $200B by decade’s end, driven by Ethernet fabrics, telemetry, automation, and open networking.