Deutsche Telekom puts NVIDIA DSX OS into production

Deutsche Telekom has begun using vCluster automation and components of NVIDIA DSX OS in its Industrial AI Cloud. The deployment marks the first large-scale industrial AI cloud to put NVIDIA DSX OS into production.

The system is designed to give customers a private AI cloud environment within hours, rather than after several days of manual work across network, platform and infrastructure teams. Each customer receives a dedicated URL, a separate cloud console, observability tools and an isolated network environment operated in Germany.

vCluster, founded by German engineers, worked with Deutsche Telekom and the NVIDIA DSX team on the service architecture. The setup takes customers from bare metal infrastructure to private cloud environments through automated provisioning and management.

The Industrial AI Cloud is aimed at organisations that want to keep sensitive industrial data within Germany while meeting sovereignty, security and compliance requirements. It is intended for companies that want dedicated control and isolation rather than a shared public cloud interface.

How it works

At the core of the system are several elements of NVIDIA DSX OS, including NVIDIA NVSentinel for GPU fault detection and remediation for Kubernetes, the NVIDIA GPU Operator and the NVIDIA Network Operator. vCluster says it orchestrates these components into a repeatable process that takes customers from raw NVIDIA GPUs to a private cloud environment.

The platform also creates a dedicated management control plane for each customer. Users sign in through their own Deutsche Telekom-branded portal and can provision Kubernetes, Slurm, GPU virtual machines or bare metal systems through the portal or an application programming interface.

According to the companies, the platform can manage different hardware generations through a single interface. vMetal is being used for bare metal provisioning across NVIDIA RTX PRO Servers and NVIDIA DGX B200 systems, with NVIDIA DGX B300 due to be added. Future systems, including NVIDIA Vera Rubin NVL72, are also expected to join the platform.

Industrial focus

Deutsche Telekom says the Industrial AI Cloud was built for industrial intellectual property and for enterprises seeking a European-operated environment for AI applications. The service is designed to meet Germany's C5 cloud security criteria and other customer requirements around private infrastructure and isolated operations.

The first phase of the project brought what Deutsche Telekom describes as Germany's first sovereign NVIDIA AI factory online in Munich. The current phase focuses on extending that model so industrial customers can receive standardised, isolated environments without having to build and operate GPU cloud infrastructure themselves.

According to the companies, a leading German automotive manufacturer is the first customer to use its own private AI cloud through the service. Deutsche Telekom plans to scale from 500 vCluster-supported NVIDIA GPUs today to more than 10,000 NVIDIA Blackwell and NVIDIA Rubin GPUs on the platform.

Growth plans

The companies also outlined plans to add more NVIDIA software over time. Deutsche Telekom is incorporating additional software, including NVIDIA Omniverse libraries and frameworks and the open-sourced NVIDIA Run:ai, delivered to customers as vCluster Stacks.

For Deutsche Telekom, the arrangement provides a consistent software stack across its infrastructure fleet as hardware changes. For customers, it offers a way to access new GPU capacity and newer chip generations without changing how they use the service.

Andreas Falkner, Head of Industrial AI Cloud at Deutsche Telekom, said the architecture is intended to simplify customer onboarding while keeping environments separate. "With vCluster orchestrating our NVIDIA DSX stack, each customer gets its own private AI cloud, on one platform that spans every GPU generation we run," said Andreas Falkner, Head of Industrial AI Cloud at Deutsche Telekom. "We value the team's deep expertise with NVIDIA hardware and open-source software."

Lukas Gentele, Co-Founder and CEO of vCluster Labs, said the project represents a production deployment of the NVIDIA software stack in an industrial setting. "NVIDIA is giving AI factory operators the best hardware in the world and now the software building blocks to run it. Our job is to turn those building blocks into a production-grade cloud," said Lukas Gentele, Co-Founder and CEO of vCluster Labs. "As a German-founded company, we're proud that Deutsche Telekom's Industrial AI Cloud is where NVIDIA DSX OS first went into production at scale, automated end to end, so any German company can get its own private AI cloud. Any NVIDIA Cloud Partner can follow the same path."

Nico Caprez, Vice President, AI Infrastructure Growth at NVIDIA, linked the deployment to wider efforts to commercialise AI factory infrastructure. "NVIDIA AI factories turn energy into valuable intelligence, and tokens into revenue. vCluster helps Deutsche Telekom put AI to work, producing and delivering tokens at scale. Built on NVIDIA DSX OS, this enables every NVIDIA AI factory to be more productive and fungible across customers and workloads," said Nico Caprez, Vice President, AI Infrastructure Growth at NVIDIA.