How Kubermatic Is Powering an AI Infrastructure Platform in Defence

BWI is the German Bundeswehr’s digitalization partner, tasked with turning its Software Defined Defence strategy into something real: moving military capability away from fixed hardware and toward software that can be updated, scaled, and deployed wherever it’s needed. Part of that mission means giving Bundeswehr units and research partners access to modern AI tools — language models, computer vision, geospatial analysis, and more.

None of that is straightforward in a defence environment. GPUs are expensive, and getting new hardware approved, purchased, and deployed can take months. At the same time, different Bundeswehr units and research partners need to share that hardware without compromising security or stepping into each other’s environments. Everything has to run on sovereign infrastructure, without relying on foreign cloud providers. And the platform can’t stop at the datacenter. It also has to work in disconnected field environments, where losing connectivity can’t mean losing the capability itself.

Kubermatic and BWI built the solution using Kubermatic’s platform: the Kubermatic Kubernetes Platform (KKP), Kubermatic Virtualization, KubeLB, and the Kubermatic Developer Platform (KDP). Together, they form the backbone of BWI’s experimental platform, known internally as xPlatforms, or BwXLab.

KKP handles the day-to-day work of running Kubernetes at scale across many tenants, with GPU scheduling and monitoring built in from the start. Kubermatic Virtualization, running on KubeVirt, gives teams access to NVIDIA GPUs — some dedicated, others shared across multiple users — without locking BWI into one hardware vendor. KubeLB keeps load balancing running even when a site loses connection, which matters for field deployments. And KDP turns what used to be a slow, ticket-based request process into something closer to ordering off a menu: a data scientist can spin up a fully configured, GPU-ready environment in minutes instead of waiting weeks.

On that foundation, Kubermatic helped BWI stand up real applications. One platform serves live language model inference and a full set of data science tools for analysts. Another combines computer vision, image generation, and 3D rendering for a different set of use cases. Bundeswehr data analysts now train on GPU-powered notebooks running on the same infrastructure: no separate systems, no special exceptions.

The platform is already running in production. As of April 2026, it supports 9 AI models across 27 physical GPUs, spread over three datacenters, with five use cases already live and several more on the way, including work tied to geospatial reconstruction and a shared language-model gateway for the wider organization.

For BWI, the payoff shows up first as speed: what used to take months of hardware procurement now takes minutes of self-service provisioning. Multiple units can share the same GPU pool safely, which keeps costs down and utilization up. And because the platform isn’t tied to any single vendor’s proprietary tooling, BWI keeps control over where this goes next.

For the Bundeswehr, it means AI capability that stays entirely on its own infrastructure, with nothing routed through a foreign cloud. That capability is now reaching toward the tactical edge, too. GPU hardware is already in place at a disconnected field site, ahead of the applications that will run on it. Meanwhile, a first group of Bundeswehr analysts is already training hands-on with GPU-powered tools, building the internal skills this kind of platform depends on long-term.

Taken together, it’s a working example of what Software Defined Defence is supposed to look like: the same AI infrastructure, built once, running consistently from headquarters all the way out to the field.