Enabling Concurrent Multi-Tenancy in NVIDIA Cloud Partners and AI Factories
New capability unifies EVPN/VXLAN control planes across Ethernet switches and NVIDIA BlueField DPUs. GPU cloud operators can now give each tenant anything from a cluster of bare-metal servers down to a single GPU, while fully enforcing isolation in hardware.
Key Highlights:
- Unified switch and DPU orchestration. Netris NAAM (Network Automation, Abstraction, and Multi-tenancy) platform now orchestrates NVIDIA Spectrum-X Ethernet switches and NVIDIA BlueField DPUs as a single EVPN fabric, enabling faster tenant onboarding, faster time to revenue, and reduced operational complexity.
- Concurrent multi-tenancy at single-GPU precision. Operators can allocate as little as one GPU per tenant with hardware-enforced isolation, increasing GPU utilization. This means more tenants per cluster and improved revenue per GPU.
- Supercharged inference performance. Netris orchestrates NVIDIA BlueField DPUs to offload network processing to dedicated hardware, freeing up host computing resources to run AI and other modern workloads.
- Hardware-based network isolation from server to fabric edge. Many compute orchestrators offer in-server DPU programming for tenant isolation, leaving the operator to manage the physical switching fabric manually or develop in-house automation. Netris enables consistent tenant isolation across the entire AI network infrastructure, including the network fabric edge, switch ports facing appliances, and non-DPU-enabled servers.
Netris, the leading provider of network automation for AI infrastructure, announced the latest expansion of its Network Automation, Abstraction, and Multi-Tenancy (NAAM) platform to orchestrate NVIDIA BlueField DPUs as part of a unified Ethernet fabric. With Netris v4.7.0, NVIDIA Cloud Partners (NCPs) and enterprise AI Factory operators can implement hardware-level tenant network isolation at any granularity — from full rack-scale GPU clusters down to individual GPUs within a single server.
Also Read: CIO Influence Interview with Gihan Munasinghe, CTO of One Identity
“By combining NVIDIA BlueField and DOCA with the Netris platform, organizations can accelerate time to production, streamline network operations, and build a foundation for autonomous, agent-driven AI infrastructure.” – Yael Shenhav, VP, NVIDIA
AI cloud operators must continuously reallocate GPU resources across tenants to maximize utilization and revenue. Without concurrent multi-tenancy, the smallest allocatable unit is an entire bare-metal server — often four or eight GPUs — limiting provisioning flexibility and leaving capacity stranded when tenants require fewer resources. Enabling multiple tenants to share a GPU server requires enforcing network isolation inside the server itself.
Implementing this isolation purely in software consumes valuable CPU resources and directly impacts inference throughput. NVIDIA BlueField DPUs address this challenge by offloading and accelerating networking, security, storage, and isolation enforcement to dedicated hardware. Installed directly in the server, BlueField connects into the Ethernet fabric via physical uplinks and exposes multiple Virtual Functions (VFs) to the host, each appearing as an independent network interface.
Netris brings these elements together by orchestrating NVIDIA Spectrum-X Ethernet switches and NVIDIA BlueField DPUs as a single, unified fabric. Drawing on decades of production networking experience, Netris automatically generates and maintains consistent EVPN and VXLAN configuration across all fabric elements. Physical switch ports and DPU VFs can be assigned to the same tenant VPC, enabling bare-metal servers, non-DPU appliances, edge gateways, and virtualized workloads to coexist within a single tenant environment — with hardware-enforced isolation across the entire datapath.
Traditional compute orchestrators operate above the network layer and typically do not extend tenant isolation across physical switches, non-DPU servers, appliances, or data center edge connectivity. By operating at the network layer, Netris complements these platforms — providing the unified control plane they depend on to deliver end-to-end, hardware-enforced tenant isolation across the entire AI data center.
Netris configures BlueField DPUs in zero-trust mode through NVIDIA DOCA, restricting administrative access from the host system and allowing operators to retain full control over network isolation and policy enforcement. By integrating with NVIDIA DOCA HBN, Netris extends this control to host networking, delivering automated, high-performance connectivity at scale for modern AI infrastructure.
“GPU cloud operators need to maximize utilization of their hardware at every moment,” said Alex Saroyan, CEO of Netris. “That means dynamically repurposing GPU resources across training and inference workloads, with tenant isolation enforced at the hardware level. By unifying NVIDIA Spectrum-X Ethernet and NVIDIA BlueField control planes into a single EVPN fabric, Netris enables truly composable AI cloud infrastructure — supporting both large bare-metal clusters and fine-grained, multi-tenant deployments without compromise.”
“As cloud AI factories scale, operators need automated networking that can enforce strict multi-tenant isolation without adding operational complexity,” said Yael Shenhav, VP, Networking Products at NVIDIA. “By combining NVIDIA BlueField and DOCA with the Netris platform, organizations can accelerate time to production, streamline network operations, and build a foundation for autonomous, agent-driven AI infrastructure.”
Catch more CIO Insights: CIO as Orchestrator of Cross-Functional Digital Strategy
[To share your insights with us, please write to psen@itechseries.com ]
