Multi-Tenant AI Clouds, Proven Before the Hardware Lands: Netris and Rafay on NVIDIA DSX Air

Introduction

Netris and Rafay have published a joint demo, Multi-Tenancy for NCPs and Neoclouds, on the NVIDIA DSX Air Demo Marketplace. It is available today. Operators can now see the full multi-tenant AI cloud stack working together in simulation: Netris orchestrating and isolating the network fabric, and Rafay turning that infrastructure into governed, self-service services that tenants consume.

The timing matters. AI cloud operators are signing contracts for capacity that is still on a loading dock, and every week a GPU cluster sits idle after delivery is lost revenue. Multi-tenancy is where most of that bring-up time goes, because it has to hold at every layer at once: the network, the compute, and the experience each tenant sees.

NVIDIA DSX Air already lets teams simulate AI infrastructure at full fidelity before hardware arrives. Netris and Rafay now extend that simulation through the two layers that decide whether an operator can sell the capacity: hard network isolation per tenant, and the platform tenants actually use.

Two layers of multi-tenancy, validated as one system

A GPU cloud tenant needs two kinds of isolation. Its traffic must be separated across every fabric, and its users need a portal, quotas, and services that belong only to them. Netris and Rafay each own one of these layers, and the joint demo shows them working as one.

Netris: the network layer. Netris NAAM (Network Automation, Abstraction, and Multi-Tenancy) orchestrates the AI networking stack.

  • Hard multi-tenancy with hardware-level isolation across east-west, north-south, and out-of-band management fabrics
  • Instant network isolation as each tenant is provisioned
  • Automated fabric addressing and leaf-spine configuration, including NVIDIA Spectrum-X Ethernet
  • The same controller moves from DSX Air into the live environment at deployment

Rafay: the platform and consumption layer. The Rafay Platform turns that infrastructure into governed services tenants consume.

  • Self-service provisioning of bare metal, VMs, Kubernetes, and AI services through a branded portal
  • Tenant onboarding, quotas, and governance from a single system of record
  • Automated tenant networks through the documented Netris integration, including storage peering
  • Metering that supports chargeback and billing

What the demo shows

The demo runs on a simulated NVIDIA HGX scalable unit inside DSX Air, with GPU nodes attached to redundant east-west leaf switches and fabric addressing managed by Netris. Rafay sits above the fabric as the system of record. When an operator onboards a tenant in Rafay, Netris builds that tenant's isolated networks, and the tenant can immediately consume compute through self-service.

Use cases: from evaluation to Day 2

Validate the business before the racks arrive. A neocloud has signed for a new GPU cluster and has two anchor tenants waiting. Instead of learning on live hardware, the team stands up a digital twin in DSX Air with Netris and Rafay already running. They onboard both tenants, confirm network isolation holds, and walk each customer through the portal they will use. When the hardware powers on, the configuration is already proven, and the first tenant goes live sooner.

Rehearse changes without touching revenue. An NVIDIA Cloud Partner wants to add a third tenant and launch a new service tier on existing capacity. Testing that on production hardware would take GPUs away from paying workloads. The team rehearses the change in the twin first, verifies the new tenant's isolation and quotas, then applies it to the live environment with confidence.

The Netris and Rafay demo is live today on the NVIDIA DSX Air Demo Marketplace. The integration behind it is part of the Rafay Platform today, with Netris SoftGate supported as of Rafay GPU PaaS v3.1-41.

Why it matters for CEOs and CTOs

  • Faster time to revenue. Tenants are onboarded in simulation, so capacity can be sold as soon as the hardware is live.
  • Lower deployment risk. Isolation, cabling, and configuration issues surface in the twin, not in front of customers.
  • Protected GPU utilization. Day-2 changes are rehearsed off production, keeping paid workloads running.
  • A validated, two-vendor stack. Network and platform layers are proven together before you commit, which shortens evaluation and procurement.

In their words

"Time is the enemy right now. The value here for customers is that they can evaluate an end-to-end solution in a very, very short window of time. With a working environment at hand that can be demonstrated to downstream users, everyone has the opportunity to make faster progress."

Haseeb Budhani, CEO, Rafay

"GPU cloud operators need to bring capacity online quickly and adapt as customer demand changes. Netris NAAM, combined with Rafay’s compute orchestration, delivers a cloud-like operating experience across compute and networking. This demo gives operators hands-on experience of dynamically allocating GPU capacity with hardware-enforced tenant isolation with no manual switch configuration"

Alex Saroyan, CEO, Netris

See it for yourself

The fastest way to understand multi-tenancy across both layers is to run it. Launch the Netris and Rafay demo on the NVIDIA DSX Air Demo Marketplace, onboard a tenant, and watch the network and the platform respond together. Then bring your own topology and test the services you plan to sell before your hardware ships.

LAUNCH THE DEMO ON DSX AIR · Read the Rafay + Netris integration docs

‍

Share this post

Want a deeper dive in the Rafay Platform?

Book time with an expert.

Book a demo
Tags:

You might be also be interested in...

Self-Service Bare Metal Is Where a GPU Cloud Starts, Not Where It Ends

Read Now

Product

Rafay is bringing NVIDIA Kubernetes Engine (NVKE) into its Kubernetes-as-a-Service (K8s) offering

A working tenant cluster about 21 minutes after logging in to a bare VM. Each new cluster after that in about three minutes. The whole lifecycle in five nvkectl verbs.

Read Now

Product

NVIDIA AI Cloud-Ready at Rack Scale: Rafay on GB200 NVL72 with NVIDIA Infra Controller

The rack is becoming the unit of AI compute. Rafay is extending its NVIDIA AI Cloud-Ready validation from single systems to full GB200 NVL72 racks with NVIDIA Infra Controller.

Read Now