Mirantis today launched the industry’s first comprehensive reference architecture for AI infrastructure, aimed at helping enterprises deploy and operate scalable, secure, and sovereign AI environments across any infrastructure.

Built on Mirantis k0rdent AI, the new AI Factory Reference Architecture enables AI workloads to be deployed within days, accelerates model development, and offers curated integrations for AI/ML tools, observability, CI/CD, and security — all leveraging open standards.

Addressing challenges in high-performance computing, it supports advanced networking (RDMA), GPU allocation and slicing, performance tuning, and Kubernetes scaling. It integrates AI Platform Services such as Gcore Everywhere Inference and the NVIDIA AI Enterprise ecosystem.

Unlike typical cloud-native workloads, AI deployments often require treating many GPU servers as a single supercomputer with ultra-fast networking. The architecture supports AI workload types — including training, fine-tuning, and inference — across dedicated or shared servers, virtualized environments (KubeVirt/OpenStack), public, hybrid/multi-cloud, and edge locations.

Designed for flexibility, the architecture uses reusable templates to configure compute, storage, GPU, and networking layers — with support for NVIDIA, AMD, and Intel accelerators.

“Enterprises need sovereign, scalable AI infrastructure that can be deployed quickly without requiring developers and data scientists to master complex infrastructure,” said Shaun O’Meara, CTO at Mirantis.

The solution also addresses key AI operational needs: data sovereignty and compliance, GPU resource sharing, hard multi-tenancy, management of distributed infrastructure, and simplified deployment for non-specialist teams.