Rafay’s NVIDIA Certification Unlocks the AI Factory for the Enterprise

SUNNYVALE, Calif. – August 27, 2026 – As enterprises scramble to acquire vast fleets of GPUs, a more complex challenge is emerging from the shadows of the AI gold rush: how to transform these expensive, powerful accelerators from isolated resources into a cohesive, efficient, and monetizable platform. Today, AI infrastructure firm Rafay Systems took a significant step toward solving this puzzle, announcing its Virtual Machines-as-a-Service (VMaaS) offering has achieved NVIDIA-Certified Hypervisors status.

This certification validates that Rafay’s platform can run virtualized AI workloads on NVIDIA’s most advanced hardware—including the HGX and rack-scale NVL72 systems—at near bare-metal performance. While seemingly a technical milestone, the strategic implications are profound. It signals a critical maturation in the AI infrastructure market, moving beyond the raw pursuit of hardware to the sophisticated orchestration required to build true, production-grade “AI Factories.” For business leaders, this development provides a blueprint for maximizing the ROI on their massive AI investments.

The AI Factory Imperative: From Raw GPUs to Managed Platforms

The era of simply buying GPUs and hoping for the best is over. Organizations are now grappling with the operational reality of supporting hundreds or thousands of users, applications, and AI models on a shared infrastructure. This creates immense challenges around security, governance, and resource allocation. The concept of the “AI Factory”—a centralized, streamlined engine for AI development and deployment—has become the strategic goal, but achieving it has been elusive.

Rafay’s certification directly addresses this operational gap. By providing a validated layer of virtualization, the company enables secure multi-tenancy, allowing organizations to safely partition their GPU clusters for different teams, customers, or projects. This is a crucial capability for the emerging class of “Neoclouds,” telecommunications firms, and sovereign AI initiatives that are building their own AI cloud services.

"Organizations are moving rapidly from acquiring GPUs to asking how those resources can securely support hundreds or thousands of users, applications, and AI workloads," said Haseeb Budhani, CEO and co-founder of Rafay Systems, in the announcement. "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."

This platform approach is what sets apart the next generation of AI infrastructure. Instead of just offering raw compute, Rafay allows operators to provide a self-service experience for developers and data scientists. Users can provision GPU-enabled virtual machines, Kubernetes clusters, or other compute environments on demand, all governed by centralized policies, access controls, and usage quotas. This transforms a static hardware investment into a dynamic, responsive internal service or even a public-facing revenue stream.

Beyond Bare Metal: Virtualization's Performance Renaissance

For years, virtualization was viewed with skepticism for high-performance computing. The performance overhead, or “tax,” of running a hypervisor was considered too great for latency-sensitive AI training and inference workloads. This forced many organizations to dedicate entire physical GPUs to single tasks, leading to poor utilization and spiraling costs. NVIDIA’s certification program is systematically dismantling that old assumption.

The NVIDIA-Certified Hypervisors program involves a rigorous validation suite that measures performance across critical AI behaviors, including compute, memory access, data-path efficiency, and LLM inference. To earn the certification, a solution must prove it can deliver performance within a tight margin—often cited as within 5%—of running on bare metal. This is achieved through deep engineering that ensures the hypervisor accurately exposes the underlying hardware topology and implements key performance optimizations.

Rafay’s achievement on cutting-edge systems like the NVIDIA GB200 NVL72 rack-scale platform is particularly noteworthy. It assures customers that they can embrace the flexibility and multi-tenancy of virtualization without making a significant performance trade-off. This validation is a powerful de-risking mechanism for CIOs and infrastructure leaders who need to guarantee performance for their most demanding AI initiatives.

This trend is not happening in a vacuum. Other players, like Mirantis and virtualization giant VMware, are also working closely with NVIDIA to bring performant virtualization to the AI stack. However, Rafay’s strategic focus extends beyond the hypervisor itself. The company’s platform provides a unified control plane for a diverse set of consumption models—from virtual machines and containers to bare metal and serverless inference endpoints—positioning it as a comprehensive operating system for the AI Factory.

Orchestrating the AI Cloud Economy

The ultimate goal for any significant technology investment is not just operational efficiency but value creation. Rafay’s strategy, bolstered by this NVIDIA certification, is squarely aimed at enabling a new AI cloud economy. By integrating orchestration with features like usage metering, chargeback, and tenant management, the Rafay Platform empowers operators to monetize their infrastructure.

This is a game-changer for entities building AI services. A telecommunications provider, for example, can leverage its data centers and fiber networks to offer sovereign AI cloud services, packaging GPU compute and AI models into standardized, metered offerings for enterprise customers. Internally, a large financial services firm can use the same platform to track GPU usage by different business units, ensuring fair cost allocation and maximizing the return on its hardware fleet. Rafay’s concept of a “Token Factory,” which allows operators to publish AI models as token-metered inference services, is the logical conclusion of this strategy.

This deep integration with NVIDIA's ecosystem, from the Inception program for startups to collaborations on GPU-as-a-Service architectures, highlights a symbiotic relationship. NVIDIA builds the world’s most powerful AI engines, and partners like Rafay build the sophisticated dashboards, controls, and business logic needed to drive them effectively.

"Virtualization is a key use case that AI Factory operators expect to leverage to address multi-tenancy requirements," Budhani stated. For business and technology leaders, this certification is more than just a technical seal of approval; it is a clear signal that the tools to build, manage, and monetize scalable AI platforms are finally here. The focus is no longer just on possessing the hardware, but on mastering the strategy to operate it.