Short answer: VMware Private AI Foundation with NVIDIA is a solution on top of VMware Cloud Foundation (VCF) for running AI workloads on NVIDIA GPUs. On VCF 9.x it adds deep learning VMs, GPU-enabled VKS clusters, pgvector databases in Data Services Manager and Private AI Services (model runtime, data indexing, Agent Builder). You need a VCF subscription, the Private AI Foundation license, and an NVIDIA AI Enterprise license when you use vGPU.
What VMware Private AI Foundation with NVIDIA is
The product name is unchanged since the 2024 launch: VMware Private AI Foundation with NVIDIA. It is not a separate platform. According to the Broadcom guide, it builds on VCF and adds GPU- and AI-centric functionality to VCF Operations, VCF Automation and VMware Data Services Manager. The current documentation covers releases 9.0 and 9.1, aligned with VCF 9.0 and 9.1.
Components
| Layer | Component | Purpose |
|---|---|---|
| VCF | Workload domain with GPU-enabled ESX hosts | vGPU (time slicing or MIG) or GPU passthrough |
| VCF | Supervisor and VKS clusters with GPU workers | Kubernetes-based AI applications |
| VCF | Deep learning VM image | Preconfigured VM for data scientists and model validation |
| VCF | VCF Automation catalog items | Self-service provisioning of deep learning VMs and GPU VKS clusters |
| VCF | Data Services Manager | PostgreSQL with the pgvector extension |
| VCF | Private AI Services | Model runtime and API gateway, data indexing and retrieval, Agent Builder |
| NVIDIA | vGPU host driver, GPU Operator | GPU access for VMs and Kubernetes |
| NVIDIA | NVIDIA AI Enterprise, NGC catalog, NIM | Licensed drivers, AI containers and inference microservices |
Licensing: what needs which license
The 9.0 deployment requirements list three licenses: the VCF subscription, the VMware Private AI Foundation with NVIDIA add-on, and NVIDIA AI Enterprise for vGPU. With the VCF license alone you can deploy AI workloads with or without a Supervisor and use GPU metrics in vCenter and VCF Operations. The add-on unlocks:
- Private AI setup in VCF Automation (catalog items for deep learning VMs and VKS clusters)
- PostgreSQL with pgvector with enterprise support in Data Services Manager
- The VMware-delivered deep learning VM image
- The guided deployment workflow in the vSphere Client
- VM classes with GPU reservation
- Private AI Services
The NVIDIA side is licensed separately. The 9.1 documentation states that vGPU deployments need an NVIDIA AI Enterprise license, while GPU passthrough does not. At VMware Explore 2025 Broadcom announced that Private AI Services would become a standard part of a VCF 9.0 subscription, so check your current entitlement in the Broadcom Support Portal before buying the add-on.
Private AI Services versions
Private AI Services runs on a GPU-accelerated Supervisor. From the release notes:
| Version | VCF | VKS Kubernetes | NVIDIA GPU Operator |
|---|---|---|---|
| 2.0.x | 9.0 | 1.32 | 24.9.0 |
| 2.1 | 9.0.2, 9.1 | 1.33 | 25.10.1 |
The model runtime supports vLLM, Infinity and llama.cpp. Version 2.1 adds an MCP tool gallery for agents and observability for models and AI applications. Both versions use PostgreSQL 16.8 with pgvector 0.8.0.
Requirements before you start
- At least three GPU-enabled ESX hosts in the first cluster of the workload domain.
- GPUs listed in the Broadcom Compatibility Guide for GPU and accelerators for AI/ML.
- SR-IOV enabled in the BIOS and on the graphics devices.
- For vGPU: the NVIDIA vGPU host driver in the vSphere Lifecycle Manager image. Note that NVIDIA NIM does not support MIG sharing mode.
- VCF single sign-on configured for the instance with the GPU hosts.
Deployment workflow
- Assign the Private AI Foundation license to the management domain; this activates the guided setup in the vSphere Client.
- Deploy a workload domain with GPU-accelerated hosts and configure vGPU or passthrough.
- Create VM classes for AI workloads and, for disconnected sites, content libraries and a Harbor registry.
- Configure VCF Automation for deep learning VMs and GPU VKS clusters.
- Install Private AI Services on the Supervisor and activate generative AI support in VCF Automation namespaces.
If you are still sizing the platform, see my vSphere 9.0 product line comparison and notes on deploying a minimal VCF 9.0 lab.
Further reading
- VMware VCF Blog: VMware Private AI Foundation with NVIDIA Achieves General Availability (original post)
- Broadcom TechDocs: VMware Private AI Foundation with NVIDIA 9.1
- Broadcom TechDocs: VMware Private AI Services Release Notes
