BWI GmbH, the German Bundeswehr’s primary IT and digitalization partner, is building the infrastructure needed to bring AI into defence operations and research.
As demand for AI grows, BWI needs to provide GPU infrastructure across central datacenters, edge locations, and disconnected field environments. The challenge was how to make that infrastructure available quickly and securely.
Traditionally, getting access to GPU resources meant individual requests, approvals, procurement, hardware delivery, and manual setup. Teams could wait months before they had the infrastructure they needed. At the same time, BWI needed multiple teams to share expensive GPU hardware without giving them access to each other’s workloads or data. The infrastructure also had to remain sovereign and operate in environments without a permanent connection to a central datacenter.
To address this, BWI worked with SVA and Kubermatic to build xPlatforms (BwXLab): a self-service platform for running GPU-powered AI workloads across BWI’s infrastructure. SVA brought its understanding of BWI’s environment and requirements, and Kubermatic brought the Kubernetes and cloud-native expertise needed to design and build the platform on Kubermatic AI. The resulting platform combines four key components:
- Kubermatic Kubernetes Platform (KKP) for multi-tenant, multi-cluster Kubernetes management, with the NVIDIA GPU Operator integrated by default.
- Kubermatic Virtualization, built on KubeVirt, for GPU access: physical GPUs can be time-sliced across multiple teams for shared workloads, or passed through directly for workloads that need a full GPU.
- KubeLB for load balancing across the cluster fleet, including sites that need to keep operating without a live connection back to the core datacenter.
- Kubermatic Developer Platform (KDP), the self-service catalog layer being rolled out on top of the platform. Teams pick the resources and node types they need from the catalog and get one of KKP’s cluster templates provisioned for them, rather than requesting a cluster manually. BWI is in the process of moving onto KDP from its earlier, CRD-based provisioning approach.
Together, these four components give BWI a shared GPU infrastructure platform: multiple teams can use the same hardware while keeping their workloads and data isolated, and GPUs can be shared across workloads instead of requiring dedicated hardware for every project. Different AI workloads, from LLM inference and computer vision to image generation and GPU-powered training, can run side by side on the same infrastructure, without requiring separate hardware and procurement processes for every project.
Most importantly, the process has changed from months to minutes. Instead of waiting for a new procurement cycle and manually building infrastructure, teams can request what they need and have their environment provisioned through self-service.
The platform is also designed for the realities of defence environments. The same architecture can run in BWI’s central datacenters and at disconnected field sites, allowing AI capabilities to be deployed where they are needed without relying on continuous connectivity to a central environment.
And because the platform runs on BWI-owned infrastructure and is built on open-source Kubernetes and virtualization technology, BWI retains control over its infrastructure and data rather than depending on a proprietary cloud or virtualization stack.
What used to mean a procurement cycle, getting budget approved, buying hardware, and waiting for delivery, now takes minutes through self-service. The same architecture that runs BWI’s core datacenters is also deployed at a disconnected field site, giving BWI one consistent way to run AI infrastructure whether it’s in the datacenter or in the field.
xPlatforms is one of the first sovereign, self-service GPU infrastructure platforms built specifically for a European defence organization, and it gives BWI a way to keep up with AI infrastructure demand without giving up control over its own infrastructure or data.
