How AVIZ ONES Enables Secure Multi-Tenant North-South Networking in NVIDIA Spectrum-X AI Fabrics
February 11, 2026
ONES & SONiC
Summary
AI infrastructure has evolved from dedicated GPU clusters into shared platforms supporting multiple users, applications, and business units. As AI environments scale, secure and consistent multi-tenancy across north-south networks becomes essential. AVIZ ONES extends tenant-aware networking beyond GPU fabrics into front-end and storage connectivity, transforming NVIDIA Spectrum-X into a cloud-ready, multi-tenant AI fabric.
Introduction
AI infrastructure is no longer single-tenant. Modern AI platforms are shared environments where GPUs, storage, and external networks must serve multiple teams simultaneously. As utilization increases, maintaining performance, security, and operational simplicity becomes critical.
NVIDIA Spectrum-X delivers high-performance Ethernet optimized for GPU east-west traffic. However, real-world AI environments require reliable north-south connectivity to storage systems, management platforms, and external enterprise networks. AVIZ ONES addresses this requirement by enabling secure, scalable multi-tenancy across north-south and storage networks within Spectrum-X fabrics.
Importance of North-South Connectivity in Multi-Tenant AI Fabrics
North-south traffic connects GPU workloads to essential external systems, making it a foundational component of AI platforms.
Key connectivity requirements include
- Access to external data sources
- Integration with enterprise and cloud networks
- Communication with management and orchestration systems
Without built-in multi-tenancy, organizations face limited isolation, manual workflows, scaling constraints, and higher operational risk. As AI platforms increasingly resemble cloud platforms, tenant-aware north-south networking becomes a necessity.
Figure 1: North-South Fabric Creation & Deployment
Spectrum-X as a Hybrid AI Fabric with Multiple Network Roles
Spectrum-X is designed as a hybrid AI fabric supporting multiple network functions within a single architecture.
The fabric combines
- A high-performance GPU east-west network
- A scalable front-end north-south network
- A dedicated storage network
AVIZ ONES acts as the orchestration layer that designs, provisions, and manages these networks as one logical fabric. This approach shifts infrastructure behavior from static networking to a shared AI platform model.
Tenant-Aware Networking Across North-South and Storage Fabrics
AVIZ ONES extends multi-tenancy beyond GPUs into front-end and storage networks.
Each tenant is treated as a first-class entity with
- End-to-end traffic isolation
- Shared infrastructure without tenant visibility overlap
- Consistent policy enforcement across GPU, front-end, and storage networks
From GPU servers to fabric egress points, isolation is preserved while maintaining shared infrastructure efficiency.
Figure 2: Tenant Management
End-to-End Tenant Isolation from Server to Border
A core principle of Spectrum-X multi-tenant design is consistency across the entire traffic path.
Tenant traffic behavior includes
- Tenant association at the GPU server
- Isolation throughout fabric traversal
- Controlled, tenant-specific egress paths
This applies uniformly to
- North-south access to external systems
- Storage access from GPU workloads
- Management and control traffic
Each tenant experiences the fabric as a dedicated network, while the underlying infrastructure remains shared.
Figure 3: Tenant GPU Allocation
Platform-Level Benefits of Multi-Tenant AI Networking
The value of this approach lies in the platform capabilities it enables rather than individual features.
Organizations can
- Offer AI infrastructure as a shared internal platform
- Support multiple teams without redesigning networks
- Integrate AI fabrics into cloud and PaaS ecosystems
- Increase utilization while reducing operational risk
Networking becomes an enabler that operates transparently in the background.
Conclusion
By extending multi-tenancy beyond GPUs into north-south and storage networks, AVIZ ONES transforms NVIDIA Spectrum-X into a cloud-ready AI fabric. This architecture supports shared AI platforms, managed services, and next-generation AI workloads. The result is a secure, scalable foundation for modern multi-tenant AI data centers.
Frequently Asked Questions
What problem does AVIZ ONES solve in NVIDIA Spectrum-X AI fabrics?
Spectrum-X is built for GPU east-west traffic but real AI environments also need north-south connectivity to storage systems, management platforms, and external enterprise networks. Without multi-tenancy built into those paths, organisations face limited isolation, manual workflows, and higher operational risk. AVIZ ONES adds tenant-aware orchestration across north-south and storage networks so the whole fabric works as a shared platform.
Why does north-south connectivity matter so much in a multi-tenant AI environment?
GPU workloads do not just talk to each other. They need to reach external data sources, enterprise and cloud networks, and management and orchestration systems. Without tenant-aware north-south networking, multiple teams sharing the same AI platform either bleed into each other's traffic or require manual network redesigns every time a new team is added. As AI platforms grow to resemble cloud platforms, that isolation has to be built in from the start.
How does AVIZ ONES extend multi-tenancy beyond GPUs into storage and front-end networks?
ONES treats each tenant as a first-class entity across all three network functions inside Spectrum-X: the GPU east-west network, the front-end north-south network, and the dedicated storage network. End-to-end traffic isolation is enforced from the GPU server all the way to the fabric egress point. Each tenant sees the fabric as their own dedicated network even though the underlying infrastructure is shared.
What does AVIZ ONES actually do as the orchestration layer for Spectrum-X?
ONES designs, provisions, and manages the GPU east-west, front-end north-south, and storage networks as one logical fabric. Instead of treating those three networks as separate infrastructure with separate management, ONES brings them under a single orchestration layer. That shift moves the whole platform from static networking to a shared AI infrastructure model that can serve multiple teams without manual reconfiguration.
What does end-to-end tenant isolation actually look like from server to egress?
Tenant association starts at the GPU server. Isolation is maintained all the way through fabric traversal and each tenant gets its own controlled egress path. North-south access to external systems, storage access from GPU workloads, and management traffic all follow the same isolation rules. One tenant never gets visibility into another tenant's traffic at any point in the path.
What business capabilities does multi-tenant AI networking unlock for organisations?
Organisations can offer AI infrastructure as a shared internal platform, support multiple teams without redesigning networks each time, integrate AI fabrics into cloud and PaaS ecosystems, and increase utilisation while reducing operational risk. Networking stops being something teams have to work around and starts running transparently in the background while the platform scales.
Why is AI infrastructure moving from dedicated single-tenant clusters to shared multi-tenant platforms?
Dedicated clusters made sense when only one team needed GPU resources. Now GPUs, storage, and external networks have to serve multiple teams simultaneously and utilisation has to stay high to justify the cost. Single-tenant models waste capacity. Shared multi-tenant platforms let organisations get more work out of the same hardware while keeping each team's traffic, data, and policies properly separated.
How does AVIZ ONES turn NVIDIA Spectrum-X into a cloud-ready AI fabric?
Cloud platforms are multi-tenant by design. Every customer gets isolation, consistent policy enforcement, and self-service access without seeing anyone else's infrastructure. AVIZ ONES brings those same properties to Spectrum-X by extending tenant-aware networking across GPU, storage, and north-south networks together. The result is an AI fabric that supports managed services and shared platform models the same way a cloud provider would.
See how Aviz ONES transforms NVIDIA Spectrum-X into a secure, cloud-ready multi-tenant AI fabric with consistent north-south networking and unified orchestration.
