Nutanix Powered by AMD: Simplifying AI Infrastructure with Enterprise Kubernetes

If you are moving AI from pilots into production, you are not choosing a GPU in isolation. You are choosing an operating model for the servers, Kubernetes® clusters, networking, applications, data, security, and upgrades that surround it. The wrong integration choices can leave your platform team maintaining a specialized environment for every accelerator or server configuration.

With Nutanix Kubernetes Platform (NKP) Metal, you can bring AMD Instinct MI350X GPUs, AMD networking, and AMD 5th Gen EPYC servers into a common platform experience. The NKP AI Application Catalog and Nutanix-certified hardware portfolio give you a clearer path to deploy AMD-powered Kubernetes infrastructure while using consistent approaches to cluster operations, application delivery, lifecycle management, and support.

That means you can choose infrastructure based on your workload, data, performance, availability, and cost requirements. This is a fast track to production AI compute with AMD and dual-native freedom from Nutanix.

The problem is not simply access to GPUs

You can have access to powerful GPUs and still face a long path to a production service. Your team must coordinate the operating system, device drivers, kernel modules, container runtime, Kubernetes operators, networking, storage, private registries, workload scheduling, upgrades, and support boundaries.

When you assemble those pieces manually, every new accelerator or server can create another integration project. That can lead to:

• Long integration projects before a cluster can be placed into service
• Different installation and upgrade procedures for different accelerator vendors
• Unclear compatibility between GPU drivers, network components, operating systems, and Kubernetes versions
• Operational risk when teams need to change a driver or operator while workloads are running
• More tools to maintain, more runbooks to document, and more places for configuration drift to develop
• Pressure to choose hardware based on what is easiest to integrate rather than what best fits the workload

You should not have to become an expert in every layer of the accelerator stack just to make an AI workload available to developers. The better approach is to standardize the repeatable infrastructure work so your team can focus on capacity, policy, reliability, and the applications your business is trying to run.

Choose infrastructure for the workload

Your AI environment rarely stands alone. You may also need AMD-based CPU platforms for virtualization, storage-dense workloads, edge deployments, VDI, and general enterprise applications that share the same data center and operational teams.

The Nutanix-certified AMD portfolio gives you options for different deployment needs. Each platform is supported with platform-specific Nutanix management capabilities, including Foundation for automated discovery, imaging, and cluster setup; Life Cycle Management for software and firmware upgrades; and Nutanix Cluster Check for proactive health checks and actionable guidance.

This gives you a way to match infrastructure to the workload while keeping deployment, imaging, upgrades, health checks, and support within a more consistent operating model. You can modernize incrementally instead of creating a separate AI island alongside the rest of your environment.

The value becomes clearer when you connect it to the decisions your team is making today:

• Private AI inference and model serving: Run containerized inference services on AMD Instinct MI350X-based bare-metal infrastructure when data sovereignty, predictable latency, or control over the hardware stack matters. NKP integrates the GPU and network operators into the Kubernetes platform; workload-level performance and sizing should still be evaluated for each model and service

• Production AI platforms for regulated or sensitive data: Keep models, data, and inference services in a customer-controlled data center or edge location, while using platform-level lifecycle and health management to reduce the operational burden of running Kubernetes outside a public cloud.

What you can do with the integrated solution

With Nutanix NKP, you can now run AMD Instinct MI350X GPU infrastructure on Cisco UCS C885A M8 servers configured with eight MI350X GPUs per server, using Ubuntu 24.04 LTS. In this configuration, the AMD GPU Operator and AMD Network Operator are available through the NKP Application AI Catalog, with integration for the AMD Pensando Pollara 400 AI Network Interface Card and Multus networking.

For this validated configuration, the value for your platform team is a common operating model around those components. Within the applicable NKP, operator, and hardware support releases, the integration provides a documented, repeatable way to:

• Simplify deployment by installing and managing the AMD GPU and network operators directly from the NKP applications catalog with a click.
• Build and manage GPU drivers using private registries and tools such as DKMS and Kubernetes Module Management (KMM).
• Support GPU workloads alongside the high-performance networking services they depend on.
• Move between supported operator versions without creating a new end-to-end process for every change.
• Remove GPU components cleanly without disrupting shared module-management and registry resources.
• Run GPU and network components together as part of the same Kubernetes cluster.

This addresses the gap between acquiring the validated AI infrastructure and making it usable as a production service. Specifically for the Cisco configuration supported today, NKP brings GPU acceleration, networking, and Kubernetes operations together, giving teams a consistent way to deploy, manage, and maintain AI workloads without having to build their own integration and lifecycle processes. 

Why this matters 

The current cloud-native conversation is moving from AI experimentation to production AI, with platform engineering, operations, security, and infrastructure choice all part of the decision. If you are responsible for delivering AI services, the question is practical: how do you make these workloads dependable without creating a new operating model for every infrastructure choice?

NKP extends this approach with a curated AI application catalog, multi-accelerator support, CNCF AI Conformance, and a dual-native operating model for running Kubernetes workloads on bare metal or with virtual machines.

For you, that means more freedom to evaluate an accelerator based on workload requirements, availability, cost targets, or sourcing strategy - not solely on which ecosystem is easiest to integrate. It also means you can preserve common processes for cluster deployment, application delivery, upgrades, observability, and governance as your infrastructure evolves.

Give your team a simpler path to production AI

Access to accelerators is only the beginning. You need an infrastructure platform that your team can operate, secure, upgrade, and scale without creating a separate set of processes for every hardware choice.

With the NKP/AMD integration, you can bring AMD GPU and networking capabilities into a consistent Kubernetes operating model, choose from a broader AMD server portfolio, and support containers, virtual machines, data, and AI across the environments where your business needs them. That is the practical path from AI infrastructure experimentation to a production platform.

To explore the supported configurations and capabilities, see the NKP AMD GPU Management documentation and NKP AMD Network Operator Support documentation.