Rafay wins NVIDIA certification for AI virtual machines

Rafay has achieved NVIDIA-Certified Hypervisors status for its Virtual Machines-as-a-Service offering, covering use with NVIDIA HGX systems and NVL72 rack-scale systems.

The designation applies to Rafay's VM service within its broader AI infrastructure management platform and is aimed at organisations using virtualisation to run AI workloads on shared GPU systems.

NVIDIA's hypervisor certification programme tests performance-related behaviour across compute, memory, data-path efficiency and large language model inference. It is designed to identify virtualisation platforms and infrastructure software suited to AI and accelerated computing workloads.

The certification gives enterprises and cloud providers added assurance when using virtualisation in production AI infrastructure. Rafay's platform is designed to let operators offer access to bare metal, virtual machines, Kubernetes clusters, SLURM environments and AI services through a single control layer.

Multi-tenant demand

The announcement reflects a broader shift in the AI infrastructure market, as operators look for ways to share costly GPU resources across multiple users, teams and applications. That has pushed virtualisation higher on the agenda for enterprises, neocloud operators, telecommunications groups and sovereign AI projects seeking stronger isolation and better utilisation.

Rafay's platform centres on turning GPU infrastructure into a multi-tenant environment with identity controls, role-based access, quotas and policy management. It also offers self-service access, allowing developers and data scientists to provision resources directly rather than rely on manual infrastructure processes.

Customers can use virtual machines as one consumption model alongside bare metal, Kubernetes and SLURM. The platform also supports usage metering and chargeback models for organisations seeking to package AI infrastructure as an internal or external service.

In practical terms, the certification targets systems including NVIDIA HGX H200 and HGX H100, as well as GB200 NVL72 and GB300 NVL72 platforms. It applies to the specific product version and NVIDIA platform tested.

For operators building out AI infrastructure, a key commercial question is how to avoid leaving expensive accelerators idle. Virtualisation has become one way to improve utilisation by dividing infrastructure among multiple workloads while keeping governance and security controls in place.

That is particularly relevant in environments where hundreds or thousands of users may need access to GPU resources for different training, inference or development jobs. Shared infrastructure models can also help providers create standardised service tiers for customers or business units.

Broader NVIDIA ties

The certification also adds to Rafay's existing work with NVIDIA-related programmes and projects. These include GPU Platform-as-a-Service reference architectures, NVIDIA AI Cloud Ready, NVIDIA Cloud Partners, AI factories, NVIDIA DSX OS, NVIDIA Infra Controller, NVIDIA AI Enterprise software and AI services delivered through its Token Factory.

Those ties place Rafay within a growing ecosystem of software and infrastructure suppliers trying to make AI compute easier to deploy and govern. As demand for accelerated computing expands, software layers that control provisioning, tenant separation, access policies and billing are becoming more central to how operators package GPU infrastructure.

Haseeb Budhani, Chief Executive Officer and Co-Founder of Rafay Systems, said virtualisation is becoming a core requirement for operators of shared AI infrastructure.

"Virtualization is a key use case that AI Factory operators expect to leverage to address multi-tenancy requirements," said Haseeb Budhani, Chief Executive Officer and Co-Founder of Rafay Systems.

He said the certification is intended to underline the company's focus in this part of the market.

"Achieving this certification under the NVIDIA-Certified Hypervisors program is yet another proofpoint to the community for Rafay's focus on making it easier for neoclouds and enterprises to govern and operate NVIDIA-powered AI Infrastructure at scale."

Budhani also pointed to changing customer priorities as AI deployments mature beyond initial hardware purchases.

"Organisations are moving rapidly from acquiring GPUs to asking how those resources can securely support hundreds or thousands of users, applications, and AI workloads," said Budhani.

"Virtualization gives operators another important consumption model. Rafay provides the orchestration, governance, and self-service framework around those environments so infrastructure can become a scalable AI platform rather than a collection of isolated resources."