en.Wedoany.com Reported - Cloud-native infrastructure platform provider Mirantis officially announced on May 14 in Campbell, California, the addition of three core capabilities to its k0rdent AI platform: Model Registry, Inference Mesh, and Inference Runtime. For the first time, this integrates the secure distribution of AI models, governance policy enforcement, inference workload routing, and efficient GPU resource utilization into a unified operational plane covering the journey from development to production. This release directly addresses the fragmentation dilemma currently faced by GPU cloud operators and enterprise AI platform teams when moving AI workloads from experimentation to production.

Kevin Kamel, Vice President of Product Development at Mirantis, pinpointed the core issue in the official announcement: "As organizations transition AI projects from the experimental phase to production environments, infrastructure teams are increasingly facing operational and governance challenges in areas such as model distribution, inference visibility, compliance enforcement, and GPU economics. Enterprises and GPU operators are forced to piece together fragile workflows and disparate tools to operate AI." He further emphasized that AI models are fundamentally different from containers—models have unique governance, sovereignty, compliance, and lifecycle requirements, and the container-centric operational paradigm of the cloud-native era cannot simply be applied wholesale.

The k0rdent AI Model Registry is optimized specifically for the storage and distribution workflows of large language models and their derivatives. This component provides a secure, native registry compliant with the Open Container Initiative (OCI) standard, capable of managing base large language models, fine-tuned variants, quantized builds, and related AI artifacts across distributed infrastructure environments, directly reducing operational complexity and supply chain risks in the secure distribution of AI models. Model versioning, provenance tracking, and permission management are embedded into the registry, enabling enterprises to implement consistent CI/CD processes for AI models just as they do for container images.

The k0rdent AI Inference Mesh assumes the responsibility for cross-cluster routing and governance of inference workloads. This component enables intelligent routing, access control, and usage metering of inference requests across federated computing resources, abstracting the reverse proxy, load balancing, and API gateway logic previously manually configured by different teams into a unified, policy-driven layer. Organizations can thereby transform raw GPU infrastructure into a governed AI inference platform, while gaining centralized visibility into model call volumes, latency distribution, and consumption costs. For operators running multiple GPU clusters or using a mix of on-premises data centers and public cloud GPU instances, the Inference Mesh provides a unified control point across environments.

Released in conjunction with the Inference Mesh, the k0rdent AI Inference Runtime focuses on the execution efficiency of inference workloads. Designed with the goal of maximizing the number of tokens generated per second per GPU, this runtime enhances GPU infrastructure utilization through model quantization, batch processing optimization, and dynamic resource scheduling. In a market environment where GPU supply remains tight and computing costs stay high, marginal improvements in inference efficiency can directly translate into significant reductions in operational costs—a demand that is particularly urgent in the current AI infrastructure market.

The three components launched by Mirantis this time are not standalone products but functional layer extensions of the k0rdent AI platform. k0rdent itself is positioned as a cloud-native infrastructure management platform for the AI era, supporting unified orchestration across bare metal, virtual machine, and container environments, and is compatible with various accelerators such as NVIDIA GPUs and AMD GPUs at the underlying level. Through this platform, Mirantis is attempting to extend its enterprise-grade infrastructure management capabilities, accumulated in the OpenStack and Kubernetes domains, to the full lifecycle of AI workloads.

Headquartered in Campbell, California, USA, Mirantis was founded in 1999. Formerly a cloud infrastructure company long maintaining OpenStack, its business focus has now shifted to providing cloud-native infrastructure solutions for AI/ML workloads. The MOSK 26.1 version released by the company in April this year has already added an AI assistant to the OpenStack platform, utilizing technical documentation and knowledge bases to provide automated operational guidance for high-performance computing and AI workloads. From OpenStack to k0rdent AI, Mirantis's transformation path reflects a clear strategic intent: deeply coupling traditional cloud infrastructure management capabilities with an AI-native toolchain to establish its position during the rapid expansion cycle of the AI infrastructure market.

The market for AI governance and inference tools is entering a period of accelerated consolidation. Enterprises are no longer satisfied with isolated access to GPU computing power but demand full-chain platform support from model storage, secure distribution, and compliance governance to inference deployment and cost management. NVIDIA's AI Enterprise suite, Google's Vertex AI, and AWS's SageMaker are all advancing similar integrations at different levels. Mirantis has chosen to start from the infrastructure layer and extend upward into model governance and inference management, attempting to find a differentiated position between cloud providers and AI platform vendors. As more enterprises embed generative AI into core business processes, the compliance of model governance, the observability of inference pipelines, and the economics of GPU resources will become the three core indicators determining platform competitiveness.

This article is compiled by Wedoany. All AI citations must indicate the source as "Wedoany". If there is any infringement or other issues, please notify us promptly, and we will modify or delete it accordingly. Email: news@wedoany.com