Mirantis, delivering Kubernetes-native infrastructure for AI, today announced that k0rdent AI has been certified through the NVIDIA-Certified Hypervisors program, validating its performance for high-performance GPU virtualization for enterprise AI factories and cloud service providers.

For organizations building AI infrastructure, this validation can help reduce the engineering effort typically required to design, benchmark, and operationalize production-ready GPU virtualization.

Mirantis demonstrated through an extensive suite of benchmarking tests that k0rdent virtualization delivers predictable, near bare-metal performance – on NVIDIA GB200 NVL72 and NVIDIA HGX-based validation tracks – while providing the operational flexibility and tenant isolation required for large-scale AI infrastructure. Across validation scenarios, the k0rdent AI virtualization stack achieved performance within 5% of comparable bare-metal deployments.

Mirantis is one of the inaugural independent software vendors (ISVs) participating in the NVIDIA-Certified Hypervisors program. The certification builds on Mirantis’ earlier participation in the NVIDIA AI Cloud Ready validation initiative, extending its validated AI infrastructure software with enterprise-ready GPU virtualization capabilities.

“As AI infrastructure evolves into shared, multi-tenant AI factories, organizations need virtualization that delivers flexibility without sacrificing performance,” said Kevin Kamel, vice president of product management, Mirantis. “This certification demonstrates that k0rdent AI enables enterprises and cloud providers to securely virtualize NVIDIA accelerated infrastructure while maintaining the performance required for AI training and inference workloads.”

Built for enterprise AI factories and cloud providers

k0rdent AI virtualization was engineered specifically for modern AI infrastructure that requires flexible GPU allocation, secure workload isolation, and efficient utilization across multiple teams and tenants.

Its topology-aware scheduler understands how GPUs, CPUs, network interfaces, and storage are physically connected within each server and across the rack. By intelligently placing workloads across NVIDIA NVLink fabrics, NUMA domains, and NVIDIA InfiniBand/RoCE networking, k0rdent AI preserves high-performance communication while avoiding the bottlenecks that can reduce AI training and inference performance.

GPU virtualization enables organizations to allocate precisely the compute resources required for each workload rather than dedicating entire GPU servers to individual users or projects. The result is a virtualization platform capable of supporting demanding AI workloads without introducing significant performance penalties.

For cloud providers and enterprise AI platforms, this enables:

  • Flexible GPU allocation;
  • Higher GPU utilization;
  • Secure tenant isolation;
  • Faster onboarding;
  • Improved infrastructure efficiency.

Virtual machines serve as secure tenancy boundaries, allowing networking, GPUs, and storage to be isolated for each customer or workload while providing the operational model expected from modern hyperscale AI cloud environments.

The k0rdent platform also simplifies operations by managing both virtual machines and Kubernetes workloads through a unified control plane. Organizations can run AI models, traditional enterprise applications, databases, and supporting services from a single management platform without maintaining separate virtualization and container orchestration stacks.

Extending into the NVIDIA AI ecosystem

Certification through the NVIDIA-Certified Hypervisors program further expands Mirantis’ collaboration with NVIDIA around AI factories and enterprise AI infrastructure.

Earlier this year, Mirantis participated in the NVIDIA AI Cloud Ready validation initiative, validating k0rdent AI for reliable AI infrastructure deployment. Together, these validations give customers a stronger basis for deploying AI training, inference, and multi-tenant AI cloud environments with confidence.

k0rdent AI Virtual Machine as a Service (VMaaS) is currently available as a technical preview and complements the platform’s broader infrastructure capabilities, including Bare Metal as a Service (BMaaS), Kubernetes orchestration, and integrated AI infrastructure management.

Organizations interested in deploying secure, high-performance GPU virtualization for NVIDIA AI infrastructure can contact Mirantis to speak with a solutions architect.

About Mirantis

Mirantis delivers the fastest path to scalable, efficient GPU cloud infrastructure for neoclouds and enterprise AI factories, with full-stack AI infrastructure technology that removes complexity and streamlines operations across the AI lifecycle, from Metal-to-Model. Through k0rdent AI and strategic partnerships, Mirantis helps organizations maximize GPU utilization, improve cloud economics, and accelerate time to revenue with production-grade multi-tenancy, intelligent workload orchestration, and automated operations.

Mirantis serves many of the world’s leading enterprises, including Adobe, Ericsson, Inmarsat, MetLife, PayPal, and Societe Generale. Learn more at www.mirantis.com.

Mirantis is a registered trademark of Mirantis, Inc. Metal-to-Model is a trademark of Mirantis, Inc. All other trademarks are the property of their respective owners.

Media gallery

About The Author
Business Wire