Deutsche Telekom launches sovereign industrial AI cloud in Munich
Deutsche Telekom and T-Systems have switched on a sovereign, NVIDIA-powered AI factory in Munich’s Tucherpark, positioning Germany as a serious contender in industrial AI infrastructure.
Specs: 10,000 NVIDIA Blackwell GPUs and 0.5 exaFLOPS
The new facility brings nearly 10,000 NVIDIA Blackwell GPUs online, including DGX B200 systems and NVIDIA RTX Pro Server GPUs, delivering up to 0.5 exaFLOPS of AI compute for training, fine-tuning, and large-scale inference. Operated by T-Systems on German soil, the platform targets industry, research, startups, and the public sector with strict controls for data protection, security, and availability. Early customers include Agile Robots, which is combining vision, robotics, and foundation models, and PhysicsX, which applies AI to technical simulation. The deployment is already operating at more than one-third capacity.
Architecture: NVIDIA stack, Polarise data center, sovereign operations
T-Systems partnered with NVIDIA for the accelerated computing stack and Polarise for the data center footprint. The result is a GPU-rich, high-bandwidth environment connected by high-performance fiber that can support demanding industrial workloads such as multi-physics simulation, generative design, robotics perception, and autonomous systems testing. The emphasis on sovereignty is clear: compute, data residency, and operations are anchored in Germany to meet European requirements.
Germany Stack: T-Systems infrastructure, SAP BTP, Siemens Simcenter
Deutsche Telekom and SAP are bundling infrastructure, platform, and applications into a cohesive “Germany stack.” T-Systems provides the infrastructure and platform layer, including its T Cloud, while SAP contributes the Business Technology Platform and AI capabilities to embed intelligence into core business processes. Siemens will bring its Simcenter portfolio and digital twin tooling to the cloud, enabling GPU-accelerated simulation, virtual commissioning, and AI copilots for engineering and manufacturing. Together, these partnerships move beyond raw GPU access to deliver industry-grade platforms with services and reference architectures.
European industry impact of sovereign AI cloud
The launch aligns European compute capacity, industrial software, and data sovereignty at a time when AI is moving from experimentation to production in factories, labs, and public services.
Compliance and data residency for regulated AI workloads
Enterprises in automotive, aerospace, healthcare, and the public sector face strict data residency and governance rules, as well as rising expectations under the EU’s evolving AI regulatory landscape. A sovereign AI cloud gives them a compliant path to train domain models, run RAG pipelines, and serve copilots without moving sensitive data to non-European jurisdictions. Notably, the SOOFI project will train a roughly 100-billion-parameter, open-source European language model focused on EU languages and industrial use cases—hosted end to end within the Munich facility.
Faster digital twins, simulation, and robotics development
Industrial AI is now converging with physical simulation and operations technology. By integrating Siemens Simcenter and related digital twin tooling, the cloud can cut simulation times, speed product validation, and enable “software-defined hardware” approaches on the factory floor. Robotics firms such as Agile Robots can iterate foundation models and deploy AI to inspection, assembly, and logistics tasks more rapidly, closing the loop from design to operation.
Sustainable data center: renewable power, cooling, heat reuse
The site repurposes a 10,700-square-meter facility in the Tucherpark redevelopment with 100 percent renewable power, advanced cooling that draws from the nearby Eisbach, and plans to channel waste heat to the surrounding district. For industrial buyers under pressure to decarbonize, embodied reuse, clean energy, and heat recovery improve the sustainability profile of large-scale AI adoption.
Strategy for telcos: from connectivity to AI platforms
The project underscores a broader shift in telecom from connectivity providers to end-to-end AI platform operators with sector depth and sovereign assurance.
Telco move to full-stack AI platforms
T-Systems is packaging compute, storage, platform services, and consulting into a full-stack offer, with SAP adding enterprise integration and application logic. This moves telcos into higher-value territory versus pure colocation or bandwidth, creating vertical solutions for manufacturing, public services, and midmarket German enterprises.
Competing on GPU capacity, SLAs, and tenancy
Securing nearly 10,000 Blackwell-class GPUs signals procurement strength in a constrained market and provides a credible alternative to hyperscale clouds for customers requiring European jurisdiction. Expect capacity reservation models, dedicated tenancy, and high-touch SLAs to become key differentiators as organizations lock in scarce accelerator supply for multi-year AI roadmaps.
Edge-to-core AI: private 5G, MEC, and GPU cores
As industrial AI pipelines mature, demand will grow for low-latency inference at the edge—on private 5G, campuses, and factories—backhauled to sovereign core training and fine-tuning in Munich. Telcos are uniquely positioned to integrate private networks, MEC, and data pipelines with GPU cores, enabling closed-loop quality, predictive maintenance, and real-time digital twins.
What to track next: adoption, models, ecosystem, ESG
Execution will hinge on ecosystem momentum, capacity ramp, and transparent performance and sustainability metrics.
Utilization growth and training vs. inference mix
Track how quickly utilization advances beyond one-third capacity, the balance between training and inference, and the diversity of industrial workloads onboarded. Pricing transparency and QoS commitments will influence adoption by SMEs and regulated buyers.
SOOFI milestones for European foundation models
Monitor milestones from SOOFI, including model checkpoints, evaluation on European-language benchmarks, and licensing that encourages commercial use while preserving openness and security objectives.
Ecosystem growth and MLOps interoperability
Watch for new ISVs, simulation partners, and robotics vendors joining the platform; support for standard MLOps stacks; and integration patterns that ease portability across sovereign clouds and on-prem estates.
Transparent energy, cooling, and water metrics
Expect more detail on energy mix, energy reuse factors, cooling efficiency, and water use, which will shape procurement decisions and ESG reporting for industrial buyers.
Next steps for CIOs and CTOs
Enterprises should use this launch to accelerate AI industrialization while de-risking compliance, sustainability, and GPU supply.
Pilot-to-production paths with MLOps and governance
Move targeted use cases—simulation acceleration, visual inspection, multilingual copilots—into pilots on the Munich cloud with clear MLOps pipelines, model governance, and shadow operations plans.
Design for sovereignty, portability, and auditability
Design deployments for data residency and auditability, while keeping models, features, and services portable across sovereign cloud and on-prem clusters to avoid lock-in.
Leverage Siemens Simcenter and digital twin tooling
Combine Siemens Simcenter, GPU-accelerated solvers, and enterprise data on SAP’s platform to shorten design cycles and enable continuous validation in production environments.
Reserve Blackwell GPU capacity and optimize cost-to-value
Evaluate reserved instances and dedicated clusters for predictable access to Blackwell-class GPUs, align budgets to training and inference profiles, and instrument usage to control cost-to-value.
Deutsche Telekom’s Industrial AI Cloud is more than another GPU farm—it is a sovereign, industry-aligned platform built to translate Europe’s engineering strength and data assets into competitive AI at scale.
