Mirantis and Netris Unify Kubernetes Orchestration and Network Automation at Scale

AI infrastructure has reached a turning point.

Neoclouds and other operators are racing to turn GPU capacity into differentiated, push-button services. Enterprises are moving beyond pilots into production inference. Telcos are building sovereign AI platforms that have to operate across regions, jurisdictions, and distributed footprints.

The demand is obvious. The hardware is here. Kubernetes is everywhere.

And yet, most AI infrastructure efforts still break down in the same place: the network.

Today, Mirantis and Netris are announcing a new integration and strategic partnership that removes that blocker and helps GPU operators deliver AI infrastructure services as repeatable, multi-tenant cloud products rather than one-off projects. Read about it in our press release.

The Real Bottleneck Is Still the Network

At this point, buying GPUs is table stakes.

Kubernetes is still hard. But Kubernetes, at this point, is also multiple domains: services, clusters, infrastructure, and the networks tying them together. Of these, networks are arguably the hardest part: east-west fabric performance, RoCE and InfiniBand tuning, north-south connectivity, and tenant isolation that holds up under real multi-tenancy.

Most of these pieces are still handled manually. Networking is still ticket-driven, fragmented across teams, and bolted on after the cluster is already standing.

That’s where AI clouds stall.

When networking is an afterthought, every new tenant becomes a custom project. Every expansion introduces fragility. And scaling multiplies operational cost.

That doesn’t produce an AI cloud business.

It produces infrastructure debt.

Networking as Part of Cluster Delivery

The Mirantis + Netris integration introduces a fundamental shift: networking becomes a first-class, Kubernetes-native part of cluster delivery itself.

Through k0rdent AI, Mirantis orchestrates the infrastructure lifecycle from bare metal through Kubernetes, automating cluster provisioning and lifecycle management. Netris makes the physical network programmable and intent-based, turning switching, fabric configuration, and tenant segmentation into something that can be delivered automatically through Kubernetes APIs.

By embedding Netris networking automation into a Kubernetes operator model, networking is no longer something configured manually after the fact. It becomes declarative, automated, lifecycle-managed, and delivered alongside compute, policy, and ongoing cluster operations.

The result is that AI infrastructure becomes something operators can deliver repeatedly, safely, and at scale.

What This Means for Neocloud Providers

For Neocloud operators and Telcos, the impact is straightforward: better economics and faster delivery.

Profitability in AI infrastructure comes down to utilization. The real question isn’t how many GPUs you have. It’s how many tenants you can safely run per rack without isolation risk or performance collapse.

With automated switching and hardware-enforced isolation using NVIDIA BlueField DPUs, providers can achieve higher tenant density and improved GPU utilization while maintaining strong trust boundaries. That means more revenue per rack and less capacity left idle for safety reasons.

Tenant onboarding improves as well.

Most Neocloud providers still onboard customers through a mix of manual network provisioning, custom fabric configuration, and cross-team ticket queues. That doesn’t scale.

With Mirantis and Netris, tenant networking becomes part of the cluster lifecycle. No VLAN sprawl. No manual switch work. No fragile overlays. Clusters and tenant connectivity are delivered together, enabling standardized AI infrastructure SKUs with predictable timelines.

And because AI workloads are extremely sensitive to east-west performance, fabric stability matters. By delivering AI-optimized networking from day one, tuned for RoCE and InfiniBand, providers can offer predictable throughput and latency even as tenant count grows. This strengthens SLAs and reduces noisy-neighbor risk.

The operational upside is just as important. Network automation removes one of the biggest scaling bottlenecks in AI infrastructure operations, lowering cost per cluster and making growth far more predictable.

What This Means for Enterprises

For enterprises (and Neoclouds and Telcos too), value centers on trust, repeatability, and control.

Many AI workloads cannot rely solely on software-based segmentation. They require hard guarantees. With isolation enforced across switches and DPUs, tenant boundaries become architectural rather than aspirational. That unlocks regulated, sovereign, and higher-value workloads that demand stronger separation.

Performance is another requirement.

Enterprises moving into production inference cannot tolerate unpredictable latency or fabric instability. Embedding AI-optimized networking into cluster delivery ensures consistent east-west communication and performance that scales with workload growth.

Operationally, enterprise platform teams are already stretched across compute, networking, security, and lifecycle management. Making networking part of Kubernetes-native delivery reduces silos and increases repeatability.

Expanding AI infrastructure into new regions or environments becomes predictable instead of a reinvention exercise.

AI Infrastructure Delivered as a Cloud Product

This integration is about turning GPU clusters into a real cloud service operators can deliver.

Mirantis brings Kubernetes-native infrastructure lifecycle automation through k0rdent AI. Netris brings intent-based networking automation integrated directly into that lifecycle.

Together, the companies enable automated, multi-tenant AI infrastructure with hardware-enforced isolation, AI-optimized fabrics, and networking delivered as part of provisioning, not as an afterthought.

The Bottom Line

AI infrastructure doesn’t fail because of compute or clusters.

It fails because networking was never designed to operate within cloud delivery.

Mirantis and Netris are changing that.

By making networking a first-class, Kubernetes-native component of AI cluster delivery, operators can scale AI platforms with higher tenant density, improved GPU utilization, faster tenant onboarding, lower operational cost, stronger isolation guarantees, and predictable fabric performance.

AI infrastructure is no longer about standing up GPU clusters.

It’s about operating them reliably and economically at cloud scale.

This integration delivers the foundation to do exactly that.