Nutanix announces partnerships with Nvidia, Supermicro and Netris. The company wants to bring AI agents and private inference into production faster and at lower cost.
At Nvidia GTC in Berlin, Nutanix is seeking friends in every layer of the AI stack. With Nvida it is working on support for the new Vera CPU, with server maker Supermicro it is building a ready-made system for inference, and with network specialist Netris it is tackling the separation between customers on shared GPU infrastructure. In each case, Nutanix supplies the software that manages the partners’ hardware.
Thomas Cornely, Executive Vice President Product Management at Nutanix, points to the infrastructure. “Taking AI from the experimentation phase to production remains a challenge for many companies, because infrastructure complexity and hardware procurement often slow progress.”
Nutanix works on Vera support
AI agents run code, call tools and APIs, and retrieve data from databases, ERP and CRM systems. That work runs on the CPU, between the steps in which the GPU generates an answer. According to Oscar Wahlberg, Senior Director Product Management at Nutanix, the CPU becomes a bottleneck as soon as companies scale up to tens of thousands of agent tasks at the same time.
Nutanix is working with Nvidia on support for the Vera CPU. It will come first to Nutanix Kubernetes Platform Bare Metal. The company is still investigating support in virtualised Nutanix environments. IT teams should thus be able to manage Vera systems with the same platform, the same storage and the same hypervisor as the rest of their hybrid multicloud.
Code that an agent writes itself can run on the Nutanix Cloud Platform in Kubernetes on virtual machines, shielded by the AHV hypervisor and Flow network virtualisation. For NKP Bare Metal, Nutanix is planning an integration with Nvidia OpenShell, which lets agents work in an isolated environment. Flow’s microsegmentation can move to Nvidia’s BlueField DPUs, so the Vera CPU can spend its capacity on the agents themselves.
Through the Agent Gateway in Nutanix Enterprise AI, administrators track token costs, limit the number of requests and control access to MCP servers. An integration with NeMo Guardrails should help detect prompt injection.
Ready-made inference with Supermicro
With Supermicro, Nutanix is working on a jointly designed system for private inference, which lets companies run open-weight models on their own infrastructure and keep the cost per token under control. Nvidia is once again close by: every Supermicro server contains eight Blackwell Ultra GPUs.
A separate Nutanix management cluster keeps the GPUs running if management fails. A storage cluster with Nutanix Unified Storage handles data loading, checkpoints and cache warming. The network runs on Nvidia Spectrum-4 switches, ConnectX SuperNICs and BlueField-3 DPUs, with NKP and NAI as software.
Matthew Thauberger, Chief Revenue Officer at Supermicro, speaks of “a pre-validated, rack-scale solution that bypasses traditional lead times”. Supermicro will sell the system from early 2027.
Network separation with Netris
Netris automates networks for AI infrastructure and uses it to separate customers. Nutanix is linking its Nutanix Cloud Platform to Netris’s NAAM platform for network automation and multitenancy.
The combination targets so-called “neoclouds”, providers that rent out GPU infrastructure as a service, and companies that share GPUs internally. As soon as a provider creates a new customer, the platform automatically configures the physical network. The separation between customers runs from the virtual workloads down to the switches. Customers set up VPCs, NAT and L4 load balancers themselves via a web console, a REST API or Terraform.
According to Ketan Shah, Vice President Product Management at Nutanix, the cloud software needed to monetise GPU clusters at scale is very complex to build in-house. With Netris, he wants to offer operators a ready-made route “from bare metal to margins”.
First solutions in early 2027
The Nutanix and Netris solution will arrive as a tech preview in early 2027, around the same time as the system with Supermicro. Nutanix has not yet given a date for Vera support, and the company is not yet announcing prices. Nutanix warns that features and timing of all announcements may still change.