Kubernetes-native platform aims to simplify the deployment of scalable AI factories and ensure investment security across multiple generations of hardware

As the use of generative AI grows, so too do the demands on the underlying IT infrastructure. Companies and operators of AI data centres face the challenge of rapidly deploying high-performance hardware, efficiently managing complex infrastructures and, at the same time, safeguarding investments across multiple hardware generations. Against this backdrop, Mirantis is further expanding its AI infrastructure platform, k0rdent AI, and has announced end-to-end support for current and future NVIDIA architectures.

The platform is designed to run so-called AI factories – highly scalable data centre environments developed specifically for the training, deployment and operation of large AI models. According to the company, k0rdent AI already supports the current NVIDIA generation, Grace Blackwell, and is also prepared for the upcoming Vera Rubin platform as well as other future architectures.

AI factories place new demands on infrastructure

With the increasing adoption of generative AI, the focus of many companies is shifting from individual AI applications towards industrially operated AI platforms. Such AI factories combine high-performance computers, specialised accelerators, network infrastructure and storage resources to form highly automated data centre environments.

It is not just raw computing power that is becoming increasingly important. Equally crucial are standardised deployment processes, automated lifecycle management functions and the ability to integrate new generations of hardware without making fundamental changes to the infrastructure.

According to Mirantis, k0rdent AI follows precisely this approach. The Kubernetes-native platform helps organisations deploy AI infrastructures at scale and automate operations throughout the entire lifecycle.

Investment security across multiple hardware generations

A key feature of the platform is its ‘Day Zero’ support for new NVIDIA architectures. This is designed to enable operators to deploy current and future GPU generations into production immediately upon their release, without having to fundamentally adapt existing infrastructure stacks or operational processes.

For organisations planning long-term investments in AI data centres, this aspect is becoming increasingly important. As hardware cycles in the field of accelerated computing have become significantly shorter, infrastructure platforms must support the transition to new architectures with as little operational disruption as possible.

According to Mirantis, this enables operators to migrate gradually from today’s Grace-Blackwell systems to upcoming platforms such as Vera Rubin without having to rebuild deployment processes or management structures.

Automation reduces operational overhead

This strategy is underpinned by the tight integration of the NVIDIA Infrastructure Controller (NICo). Mirantis uses the software as a central provisioning and lifecycle management layer within k0rdent AI.

Among other things, NICo automates the detection of new bare-metal systems, the validation of firmware versions, the provisioning of NVIDIA BlueField DPUs via DOCA, as well as network segmentation and security functions for multi-tenant environments. Furthermore, the platform supports procedures for cleaning up tenant-specific data and the automated management of the entire infrastructure lifecycle.

Many of these functions can, in principle, also be implemented using individual open-source components. However, in the day-to-day operation of large AI infrastructures, the integration and maintenance effort increases significantly. Mirantis therefore takes the approach of providing these functions as a validated and fully supported platform to shorten implementation times and reduce the complexity involved in setting up AI factories.

Standardised platform for AI data centres

In addition, k0rdent AI supports integration with NVIDIA DSX OS, the modular open-source platform for AI factories within the NVIDIA DSX architecture. The aim is to provide standardised operating models for AI data centres and to manage infrastructure components via a common platform.

This approach offers several advantages for operators of large AI environments. Standardised processes facilitate the deployment of new clusters, simplify the operation of distributed infrastructures and reduce the administrative burden associated with maintenance and scaling. At the same time, new generations of hardware can be integrated more quickly into existing environments.

Security and tenant isolation are becoming increasingly important

With the growing prevalence of AI clouds and multi-tenant platforms, security aspects are also coming more into focus. Operators must ensure that infrastructure resources are reliably isolated between different tenants and that no residual data remains after hardware has been reallocated.

This is where Mirantis relies on NICo’s security features. These include automated network isolation procedures, the secure removal of tenant-specific data (‘tenant sanitisation’), and controlled lifecycle automation for shared AI infrastructures. For Neocloud providers and operators of large AI data centres in particular, these features form a vital foundation for the secure operation of shared platforms.

Collaboration with NVIDIA

According to the companies, the collaboration between Mirantis and NVIDIA goes beyond the mere integration of existing software. Mirantis is actively involved in the further development of the open-source components surrounding NICo and contributes its own developments to the project. Furthermore, the company is one of the founding partners of the NVIDIA AI Cloud Ready initiative, which certifies and validates infrastructure platforms for use in AI clouds.

This close coordination is intended to give users early access to new technologies and enable them to deploy future hardware platforms more quickly.

AI infrastructure is becoming a strategic success factor

With the increasing industrialisation of artificial intelligence, high-performance infrastructure platforms are becoming a key competitive factor. Companies need solutions that not only provide high computing power but also enable the secure, scalable and cost-effective operation of complex AI environments.

Against this backdrop, open, Kubernetes-native platforms are becoming increasingly important. They create the conditions for flexibly integrating new generations of hardware, automating operational processes and, at the same time, retaining control over their own infrastructure.

With the expansion of k0rdent AI, Mirantis is positioning itself in this field as a provider of a platform that covers the entire lifecycle of modern AI infrastructures – from hardware provisioning and infrastructure management to support for future generations of accelerated computing. For operators of large AI data centres and cloud platforms in particular, this approach could help safeguard investments in the long term and accelerate the development of standardised AI factories.