Aviz Networks

Aviz Networks Explore our Networking 3.0 stack - universally compatible, SONiC-ready, with 24/7 global support

ONES is the Build layer of the Aviz AI Factory stack - open, SONiC-based, and running the fabric your GPUs sit on.New: O...
09/01/2026

ONES is the Build layer of the Aviz AI Factory stack - open, SONiC-based, and running the fabric your GPUs sit on.

New: ONES Forecast. Same dashboards you already use - now showing where things are headed, not just where they've been.

Predicts before it's an incident:

- CPU, memory & temperature
- Disk wear & endurance
- PSU voltage, current & fan health
- RoCE traffic & congestion
- Optical link power & temperature
- Routing table & ASIC capacity

No new tooling. Just a toggle on the charts you already use.

Full breakdown of ONES Forecast: how it works, and what it means for AI-driven infrastructure - https://na2.hubs.ly/H07yJR_0

Network operations shouldn’t wait for someone to ask the right question.That’s the shift from “Ask Once” to “Always-On” ...
08/29/2026

Network operations shouldn’t wait for someone to ask the right question.

That’s the shift from “Ask Once” to “Always-On” with AI-NOC. Aviz brings one networking layer for AI factories, integrating upward with the platforms and applications you already use while abstracting the infrastructure below. From building and managing the fabric with ONES, to gaining token and traffic visibility with AI-Parser, AI-NOC brings it together for agentic operations.

With background agents continuously monitoring the environment, AI-NOC can surface what matters, identify potential issues, and help operators act before they become incidents.

The goal: move from reactive monitoring to proactive, always-on network operations.

Read the blog to learn more: https://na2.hubs.ly/H07wz-C0

Tenants are provisioned by standalone scripts, and the information lives in a spreadsheet. That's one of the pain points...
08/28/2026

Tenants are provisioned by standalone scripts, and the information lives in a spreadsheet. That's one of the pain points we hear from nearly every neocloud team building an AI factory.

Aviz builds AI Factory as one networking layer: it integrates upward with the orchestration stack already in place (Kubernetes, Slurm, Run:AI) and abstracts away everything below it (GPUs, DPUs, storage, servers). Nothing already deployed has to be removed for it to start working.

On NVIDIA's GB200 and GB300 NVL72 systems, 72 Blackwell GPUs share one NVLink fabric that standard VLANs can't isolate, NVLink is not Ethernet. ONES Fabric Manager gives operators one control plane:
- Manages GPU partitioning within a rack
- Coordinates cross-rack InfiniBand partitioning when a tenant spans more than one rack
- Commits as one atomic operation, with clean rollback if anything fails

Same API call whether a tenant fits in one rack or spans several, on GB200 or GB300.
Read the blog to learn more: https://na2.hubs.ly/H07w1r20

08/27/2026

Networks still run on standalone scripts, spreadsheets, and a reference architecture that keeps getting modified by hand, while GPUs get provisioned like cloud infrastructure.

Aviz treats the network the same way: built, observed, and operated like the rest of the AI factory, not bolted on after the fact. Nothing already deployed has to be ripped out for it to start working.

We're at AI Infra Summit 2026 (Sept 15–17, Santa Clara Convention Center), by appointment only. If your GPU fabric strategy has any of these gaps, book a 1:1 with Aviz. Meetings are limited- https://na2.hubs.ly/H07tLdv0

08/26/2026

The scariest alerts in an AI factory aren't the ones that fail outright. They're the ones that refuse to tell you what's wrong.

Thomas Scheibe (Chief Product Officer, Aviz Networks) breaks down what a control room actually watches for: GPU utilization, token efficiency across the estate, and failure scenarios spanning services, network links, and transceivers. All of it rolls up to one thing, user experience.

But Mohan Atreya (Chief Product Officer, Rafay) flags the harder problem: network flapping. Not a clean pass or fail, just a system stuck in between, and no way to know if it self-corrects or needs intervention.

The gap between a clear failure and a grey one is exactly where AI-NOC and agentic network operations earn their keep, catching what traditional monitoring misses before it becomes downtime.

Watch the full podcast now- https://na2.hubs.ly/H07rSj70

Aviz Open Network Enterprise Suite operationalizes NVIDIA AI Factory fabrics end-to-end, aligned with NVIDIA reference a...
08/25/2026

Aviz Open Network Enterprise Suite operationalizes NVIDIA AI Factory fabrics end-to-end, aligned with NVIDIA reference architecture and validated on Spectrum-X, BlueField, InfiniBand, and NVLink.

One place this shows up: Host-Based Networking. HBN moves VTEP termination, tenant VRFs, and BGP routing off the leaf switch and onto the BlueField-3 DPU in each server, distributing intelligence closer to compute.

ONES orchestrates the full lifecycle:
→ Day 0: DPU and switch configuration, validated before deploy
→ Per-tenant VF allocation, fully automated
→ EVPN sync, pushed atomically across the fabric
→ DPU readiness checks so no deploy goes out with missing hardware
Distributed intelligence. Centralized control. That's how AI Factory fabrics stay simple to operate as they scale.

Read the full breakdown - https://na2.hubs.ly/H07nKVk0

Break it in simulation, not in production. That's the whole idea behind Aviz ONES, now live on the NVIDIA DSX Air Demo M...
08/24/2026

Break it in simulation, not in production. That's the whole idea behind Aviz ONES, now live on the NVIDIA DSX Air Demo Marketplace as Aviz ONES Multi-tenant AI-Fabric Provisioning & Monitoring.

Inside the demo:
1.) Simulated AI fabric deployment, start to finish, no physical hardware required
2.) Day 0 orchestration: fabric design and configuration generated and validated upfront
3.) Tenant provisioning: allocate GPU resources and enforce network isolation per tenant
4.) Day 2 monitoring: continuous visibility into the fabric after deployment

Why it matters:
1.) Infrastructure teams can validate an AI fabric design before it ever touches production
2.) Multi-tenancy becomes a network-level capability, not an afterthought bolted on later
3.) Operations teams get real AI visibility into the fabric instead of static, after-the-fact dashboards

Try the demo: https://na2.hubs.ly/H07p45J0
Read how we built it below- https://na2.hubs.ly/H07p3Q80

Thank you to our engineering team!

08/23/2026

FTAS for AI Factory is Aviz's vendor-neutral lab that continuously validates AI infrastructure, GPU, DPU/NIC, fabric, storage, and orchestration, so customers get a proven, deployable design instead of finding failures late in a PoC or production.

One area that stands out: resiliency.
- FTAS runs resiliency testing through every release
- It deliberately injects real failure conditions: node failovers, link flaps, container restarts, route push and withdrawal
- Each scenario has a defined convergence window, set from Aviz's SONiC experience, customer recommendations, and community guidelines
- If the fabric doesn't converge, traffic doesn't resume, or there's unexpected loss, the test is marked failed

Watch the full video to learn more - https://na2.hubs.ly/H07mC3V0

Your NVIDIA AI Factory has hundreds of moving parts, GPUs, NICs, storage, orchestration. When something changes, do you ...
08/22/2026

Your NVIDIA AI Factory has hundreds of moving parts, GPUs, NICs, storage, orchestration. When something changes, do you know who changed it, and when?

ONES operationalizes AI Factory fabrics through Day 2, where change control lives. Audit Service is what makes that real.

Every action, UI or API, logged automatically, server-side:
→ One chronological Activity Log
→ "View Related" for full event context
→ Instant filters, CSV export, real-time Syslog forwarding

No recollection. No reconstruction. Just accountability at scale.
Read the full breakdown: https://na2.hubs.ly/H07mC6H0

Not every CVE on your list is actually urgent.Most teams find out which one matters only after it's exploited.Patchmaged...
08/20/2026

Not every CVE on your list is actually urgent.

Most teams find out which one matters only after it's exploited.
Patchmageddon Live. September 17, 9 am PST.

Network Copilot's Patch Management Agent discovers every device, flags what's actually exploitable, and fixes it through your existing ITSM process. Audit pack included.

For network and security engineers, NOC/SOC leads, compliance and GRC teams, IT operations leaders, and CISOs.

Bring your own backlog. We'll triage it live, on non-production data. Register your interest below- https://na2.hubs.ly/H07lm0N0

Address

2150 N First Street Suite #442
San Jose, CA
95131

Website

Alerts

Be the first to know and let us send you an email when Aviz Networks posts news and promotions. Your email address will not be used for any other purpose, and you can unsubscribe at any time.

Shortcuts

Share

Category