Nutanix announced three collaborations today at NVIDIA GTC Berlin: a jointly engineered private inference system with Supermicro built on servers with eight NVIDIA B300 Blackwell Ultra GPUs per node, Day Zero support for NVIDIA Vera CPUs starting with NKP Bare Metal, and a multitenant stack with Netris that extends tenant isolation into the physical network. The Supermicro system pairs Nutanix Agentic AI (NAI) with dedicated management and storage clusters and NVIDIA Spectrum-X networking, and it is scheduled to ship in early 2027 through Supermicro and its channel partners.
Private Inference With Dedicated Management and Storage
The Supermicro system combines Nutanix Cloud Platform, Nutanix Kubernetes Platform (NKP), and Nutanix Enterprise AI on validated Supermicro GPU servers. Nutanix says the stack covers provisioning from bare metal through model-as-a-service and a secure Agent Gateway, with a built-in AI catalog for common AI applications. The stated goal is to cut custom configuration and integration delays and to control the cost per token of private inference as GPU capacity grows, though neither company has published pricing or a measured cost-per-token figure.
A dedicated, resilient Nutanix management cluster on the NX-8170-G10 provides the single-tenant control plane, an arrangement meant to keep a management-plane failure from disrupting GPU execution. A separate Nutanix Unified Storage (NUS) cluster on the NX-8150-G10 handles the high-bandwidth I/O for data ingestion, checkpointing, and cache warming; Nutanix describes NUS as NVIDIA Enterprise-certified storage that supports context caching and model serving.
엔비디아 스펙트럼-X networking ties the clusters together with Spectrum-4 Ethernet switches (SN5610 and SN5600), ConnectX-7 and ConnectX-8 SuperNICs, and BlueField-3 DPUs for infrastructure offload and in-band management. The companies describe the fabric as deterministic, non-blocking GPU-to-GPU connectivity. “Collaborating with Nutanix, we want to accelerate time-to-online for enterprise AI,” said Matthew Thauberger, chief revenue officer at Supermicro.
NVIDIA Vera Support Starts With NKP Bare Metal
Nutanix plans to work toward Day Zero support for NVIDIA Vera CPUs, beginning with NKP Bare Metal and expanding later to virtualized Nutanix environments, with no date set for either. The CPU side of an agentic application is the work between GPU operations: tool calls, code execution, retrieval, orchestration, and outcome processing. Nutanix says it will pair Vera CPUs with NVIDIA 오픈쉘 and NKP to build safe, private runtime environments for agents, and it has published no performance figures for the planned implementation.
Netris Extends Multitenancy Into the Physical Network
Nutanix and Netris are building a jointly engineered stack that combines Nutanix Cloud Platform with the Netris Network Automation, Abstraction, and Multitenancy (NAAM) platform for neoclouds, NVIDIA Cloud Partners, and enterprises running multitenant GPU infrastructure. Nutanix supplies compute orchestration across virtual machines, containers, CPUs, and GPUs, along with storage, Kubernetes, security, the Nutanix Agent Gateway, and model-as-a-service and AI platform services. Netris supplies physical network automation, abstraction, and multitenancy, translating each tenant’s requirements into network services for bare-metal and virtualized GPU nodes and configuring the physical fabric as the tenant’s resources are provisioned.
The network layer exposes VPCs, elastic IPs, NAT, and L4 load balancers through a web console, a REST API, and Terraform, and covers edge, East-West, North-South, scale-up, and host-based traffic. Tenant isolation runs from virtual workloads through DPUs to the physical switches, with hard isolation enforced in the networking hardware. Netris automates Ethernet, InfiniBand, DPUs, and virtual and edge networking from Day Zero design through live operations, and the company counts more than 70 live deployments. The Nutanix and Netris stack is planned for tech preview in early 2027; no general availability date has been set.
아마존