Nutanix Kubernetes Platform Achieves CNCF Kubernetes AI Conformance

The Nutanix Kubernetes Platform (NKP) solution is now a CNCF Certified Kubernetes AI Conformant Platform. NKP 2.18 satisfies all twelve requirements of the Certified Kubernetes AI Conformance Program validated on production infrastructure with real GPU workloads.

What is the Kubernetes AI Conformance Program?

At KubeCon + CloudNativeCon North America 2025, the Cloud Native Computing Foundation launched the Certified Kubernetes AI Conformance Program, a community led effort to define and validate standards for running AI workloads reliably and consistently on Kubernetes®. It defines the minimum set of capabilities, APIs, and configurations a Kubernetes platform must offer, like accelerator management, gang scheduling, autoscaling, inference networking, observability, and security. These capabilities are defined on top of standard Kubernetes conformance, to run AI/ML workloads dependably.

The program follows the model that made Kubernetes itself interoperable. The original Certified Kubernetes Conformance Program brought consistency to more than 100 distributions and platforms across every major cloud and on-premises vendor. The AI Conformance Program applies that same proven, community-driven process to AI infrastructure.

And the timing could not have been better. Research done by Linux Foundation shows that organizations are increasingly building custom AI solutions, running AI workloads on Kubernetes, and viewing open source as a critical part of their AI strategy. As AI moves into production at that scale, fragmentation in form of bespoke platforms, inconsistent APIs, and non-portable workloads becomes a business risk. With AI Conformance the community aims for a shared, vendor neutral baseline that keeps AI workloads portable across environments and free of lock-in.

Why conformance matters for enterprise AI

The goal of the Kubernetes AI Conformance program is to simplify AI/ML on Kubernetes, support interoperability and portability for AI workloads, and enable a growing ecosystem of AI tools on a standard foundation.

Setting up a Kubernetes cluster for AI/ML can be a complex undertaking. An AI conformant platform like NKP is designed to handle these underlying complexities for you, supporting your efforts to optimize your environment for scalability , performance, portability, and interoperability - across public cloud, on-premises, and edge.

For a detailed look at all twelve requirements and how NKP satisfies each one, see the NKP AI Conformance documentation.

How NKP meets the requirements

NKP's conformance was validated on NKP 2.18 running Kubernetes v1.35 with NVIDIA GPUs on Nutanix infrastructure. A few highlights of how the platform delivers each capability area:

Accelerator management. NKP ships the NVIDIA GPU Operator as a platform component, automating driver, container toolkit, device plugin, and runtime lifecycle on GPU nodes. The Dynamic Resource Allocation (DRA) APIs that are generally available since Kubernetes v1.34  are served out of the box in NKP, for fine-grained, structured accelerator requests. For utilization, NKP supports GPU sharing through time-slicing, allowing multiple pods to oversubscribe a single physical GPU, and offers vGPU-enabled node pools managed through the same device-plugin path as physical GPUs. 

Scheduling and orchestration. Distributed AI workloads need all-or-nothing placement. NKP delivers gang scheduling through Kueue from the NKP AI Applications catalog. A job that fits its quota is admitted atomically, and a job that doesn't is refused in full, no partial placement, no stranded accelerators. For elasticity, NKP's Cluster API foundation means the upstream Cluster Autoscaler scales GPU node pools independently, adding accelerator nodes when GPU-requesting pods are pending and reclaiming them when idle, with the same mechanism across public cloud and on-prem, including bare metal. The Horizontal Pod Autoscaler works on accelerator pods too, including scaling on custom AI metrics such as live GPU utilization served through NKP's Prometheus Adapter.

Inference networking. Serving models in production means constantly rolling out new model versions without breaking the applications that call them. NKP supports the Kubernetes Gateway API and provides Envoy Gateway in its catalog to make this safe: you can send a small percentage of live traffic to a new model version to test it before a full rollout (canary or blue/green deployments), or route specific requests to a specific model version based on a request header. If the new version underperforms, you can shift traffic back without client changes and minimal disruption.

Observability. The DCGM Exporter surfaces per-GPU utilization, memory, temperature, power, and interconnect metrics on a standard Prometheus endpoint that NKP's integrated monitoring stack (Prometheus, Grafana, Alertmanager, Loki, Thanos) discovers and scrapes automatically. The same ServiceMonitor contract collects metrics from AI workloads themselves -  job queue depth from Kueue, and per-model inference latency histograms from KServe.

Security. GPUs are expensive, shared resources, so it matters that no workload can grab one it wasn't given. NKP is designed so that a container is granted access to a GPU only if its pod explicitly requests one, helping to enforce proper allocation Each GPU is handed to exactly the workloads it was assigned to, helping to prevent unauthorized pods from accessing the device. With this, teams sharing a cluster can achieve stronger isolation between their AI workloads.

One standard. Every environment.

AI Conformance is especially valuable for NKP customers because NKP delivers the same platform stack across public clouds and on-premises infrastructure including bare metal. Certification helps support consistent behavior for core AI capabilities across those environments, making workloads easier to move without re-platforming.

That is the value of open standards: portability without lock-in. Nutanix is proud to support the CNCF’s community driven effort to make AI on Kubernetes interoperable, reproducible, and production-ready, and to deliver a conformant platform from day one.

Learn more: the Certified Kubernetes AI Conformance Program is developed in the open at github.com/cncf/ai-conformance. For NKP AI capabilities, see the NKP documentation and the NKP AI Applications catalog.

_{©2026 Nutanix, Inc. All rights reserved. Nutanix, the Nutanix logo and all Nutanix product and service names mentioned are registered trademarks or trademarks of Nutanix, Inc. in the United States and other countries. Kubernetes is a registered trademark of The Linux Foundation in the United States and other countries. All other brand names mentioned are for identification purposes only and may be the trademarks of their respective holder(s).}