VMware vSphere 9.1 in VMware Cloud Foundation Earns NVIDIA-Certified Hypervisor Status with Near Bare-Metal AI Performance

# What happened VMware announced that vSphere 9.1 (and all future vSphere 9 releases) is an NVIDIA-Certified Hypervisor. The certification shows that VMware Cloud Foundation (VCF) can run representative AI and accelerated computing workloads with near bare-metal performance while preserving the management and operational benefits of virtualization.

# Why it matters Agentic and generative AI workloads produce non-linear inference and distributed compute patterns that can strain infrastructure. Certification gives enterprises a clear signal that VCF meets NVIDIA's performance thresholds across a set of behavior‑based tests relevant to production AI deployments. That makes VCF a candidate platform for on-prem AI where performance, operational control, and cost matter.

# What was tested and validated NVIDIA's validation focused on representative, performance-critical behaviors rather than a single benchmark. vSphere 9.1 met the thresholds set by NVIDIA across these categories and the corresponding use cases:

• GPU collective communication: multi-GPU distributed training, multi-GPU inference, communication-intensive HPC.
• GPU, CPU, and memory bandwidth: data-intensive training, high-throughput or batch inference, preprocessing, analytics.
• Compute performance: AI training and fine-tuning, HPC simulation, GPU-accelerated analytics.
• AI inference: generative AI, high-throughput model serving, large-scale LLM and multimodal inference, agentic AI workloads.
• RDMA data path: multi-node training, distributed large-scale inference, MPI and network-bound HPC.
• Topology and device-mapping evidence: cross-cutting validation for multi-GPU/multi-node training and scale-out inference.

The blog states vSphere in VCF met NVIDIA's performance thresholds for all categories while preserving virtualization benefits.

# What this enables for enterprises VCF positions itself as a unified private cloud platform that combines cloud-like scale and agility with private cloud security and resilience. With the NVIDIA-Certified Hypervisor designation:

• IT teams can consider VCF for performance-sensitive AI deployments without committing to bare-metal-only architectures.
• Organizations can run mixed workloads—traditional enterprise applications alongside AI—under a single operational model.
• The certification supports operational consistency for data centers that want to centralize management and reduce TCO while running AI workloads.

# Ecosystem and hardware coverage VCF is certified across NVIDIA's major GPU architectures cited in the announcement: Blackwell and Hopper. The post also references a partnership between Broadcom and NVIDIA aimed at expanding choice and operational flexibility for AI infrastructure.

# Where to go next The announcement links to an NVIDIA technical whitepaper for test details and points readers to VMware resources (VCF AI/ML page and contact form) for more information and engagement.

# Bottom line This certification indicates VCF can meet NVIDIA's defined performance thresholds for a broad set of AI and HPC behaviors while keeping the operational advantages of virtualization, and it covers current NVIDIA server GPU architectures to support enterprise on-prem AI deployments.