Rafay Platform: Workload Orchestration for AI Factories & Neoclouds Powered by Aviz ONES & NVIDIA Spectrum-X Ethernet Fabrics

TL;DR

With the combination of Spectrum-X, Aviz ONES, and Rafay Platform, AI factories and neoclouds are equipped with an end-to-end fabric to cloud stack. Aviz ONES and Rafay implement multi-tenant, self service, secured, cost attributed, and optimized GPU consumption. Spectrum-X builds the lossless Ethernet fabric. This automates GPUs for faster deployment and better utilization of the overall infrastructure. Compared to the alternative of managing the infrastructure as a fixed cost by platform teams, the automation of this stack enables these teams to monetize the offered service.

Introduction

While enterprises and neocloud providers rush to create AI factories, a variety of issues will slow down that important work. One of those issues is a lack of enterprise-ready, multi-tenant service infrastructure. Infrastructure only with lots of GPUs is not enough. Closing this gap requires a governed, secure, and observable way to take a service from the network fabric all the way to the application. Fabric performance, GPU-aware orchestration, and workload governance are the focus of Rafay, Aviz ONES, and NVIDIA Spectrum-X. With their partnership, they are working to create a unified suite of AI solutions.

Why Rafay?

Traditional data centers are transforming into AI factories. AI factories contain high density GPU fabrics built specifically for AI training and serving at scale. Enterprises and neoclouds are investing heavily in AI factories. There is a gap though between GPU cores and enterprise grade, multi-tenant service delivery. That gap is what Rafay Platform fills as the AI factory and neocloud workload scheduler, governance, and scale platform for multi-tenant service delivery. The Rafay Platform is a workload scheduler and service governance platform that integrates and orchestrates cloud-based AI fabrics and NVIDIA Spectrum-X Ethernet with lossless technology, and Aviz ONES fabric orchestration. This solution provides continuum fabric-to-cloud and improves the utilization and acceleration of AI workloads for platform teams.

What Are Rafay’s Core Capabilities?

Lifecycle Orchestration: Facilitates the deployment, scaling, and upgrades of AI applications that leverage GPU-accelerated compute in private, hybrid, and sovereign computing environments.

Secure Multi-Tenancy: Strong tenant isolation is provided through enterprise-grade RBAC, policy controls, and audit trails across all Business Units.

Enterprises vary widely from small local businesses to multinational corporations. Depending on the size, enterprises manifest different resource requirements.

Policy & Cost Governance: Track your usage across all of your tenants using exposure of metrics for use in billing and chargebacks.

GPU-Aware Allocation: GPU Elasticity with Dynamic Binding Dynamically bind GPUs to workload to ensure deterministic performance, and maximized utilization of each and every GPU.

Workload Orchestration: Plans and dictates both GPU compute and AI workbenches as well as inference on a per-tenant basis via a catalog.

Ecosystem Integration: Works with NVIDIA NIM, Run:AI, Kubeflow, Ray & Jupyter natively across the entire AI pipeline.

The outcome: Which is pretty much what it is. The idea is infrastructure becomes your source of new ideas, the GPU is a utility you rent, not a CapEx spend; we’re talking about delivery speed, delivery efficiency across all layers of the AI infrastructure stack.

How Does It All Fit Together?

• NVIDIA Spectrum-X Ethernet — lossless, congestion-free fabric foundation with SuperNICs.
• Aviz ONES — GPU-aware fabric orchestration, tenant segmentation, and lifecycle automation.
• Rafay Platform — orchestrates networked GPUs into governed compute (Bare Metal, Kubernetes, or VMs as a Service).

What Is the Joint Value?

Capability	Delivered By	Value to Customers
GPU-Aware Orchestration	Aviz ONES + Rafay	Predictable GPU performance from fabric to pod through automatic binding and lifecycle management.
Secure Multi-Tenancy	Aviz ONES + Rafay	Network- and Kubernetes-layer isolation and governance for sovereign & regulated deployments.
Workload Orchestration	Rafay Platform	Automatically schedules, scales, and manages the entire lifecycle of AI/ML workloads across clusters & environments.
Telemetry & Observability	Aviz ONES	From one place, get fabric and workload telemetry. From a single pane of glass, see how link health, ECMP balance and GPU utilization all perform.
Self-Service GPU Consumption	Rafay Platform	On-demand access to GPU clusters and AI workbenches through governed, single-click catalogs.

What Are the Use Cases?

• AI/ML Training & Inference — governed scheduling and lifecycle automation for deterministic GPU clusters.
• Enterprise AI Factories — sovereign-ready, multi-tenant AI infrastructure with cost attribution.
• GPU PaaS — governed orchestration and billing controls that monetize GPU infrastructure for AI factories and neoclouds.

What Does the Reference Architecture Look Like?

Single Architecture covering Fabrics to Cloud: Rafay provides tenancy and governance orchestration and Aviz ONES provided the GPU & network resource distribution over NVIDIA Spectrum-X fabrics resulting into a secure & multi-tenant GPU cloud.

Conclusion

It’s transforming the compute farm into an organized, billable service that grows as needed-powered by an on-prem to cloud Fabric+Cloud Solution leveraging NVIDIA Spectrum-X’s lossless fabric, Aviz ONEs GPU-aware orchestration engine, and Rafay’s multitenant, workload governance platform so platform teams have a high-performance, tenants-isolated, predictable platform that will monetize their communal compute. This simplifies all AI delivery-use cases of training through inferencing.