Deutsche Telekom Opens Europe's Largest Sovereign AI Factory in Munich, Deploying 10,000 NVIDIA Blackwell GPUs Under German Law
Germany's first industrial AI cloud delivers 0.5 ExaFLOPS from a converted bank building cooled by the Eisbach river, marking Europe's boldest bid for AI independence from US hyperscalers.
Overview
Deutsche Telekom has opened what it calls Europe’s largest sovereign AI infrastructure in Munich, deploying nearly 10,000 NVIDIA Blackwell GPUs capable of delivering up to 0.5 ExaFLOPS of computing power. The Industrial AI Cloud, which officially went live on February 4, was built in roughly six months from concept to operation and represents a one-billion-euro partnership between Deutsche Telekom, NVIDIA, and data center partner Polarise, according to a Deutsche Telekom press release.
The facility increases Germany’s total available GPU capacity by approximately 50 percent and is designed to offer European companies, research institutions, and public-sector organizations high-performance AI compute without sending their data outside EU jurisdiction.
What We Know
Hardware and Location
The AI factory houses more than 1,000 NVIDIA DGX B200 systems and NVIDIA RTX PRO Servers across roughly 10,700 square meters of floor space in Munich’s Tucherpark district, according to the NVIDIA blog. The facility occupies a converted former Hypovereinsbank building and runs entirely on renewable energy. Cooling comes from water drawn from the nearby Eisbach river, and waste heat is fed back into the surrounding Munich district heating network, according to Euronews.
Deutsche Telekom describes the 0.5 ExaFLOPS capacity as sufficient to let all 450 million EU citizens run an AI assistant simultaneously, as noted in its press release.
The Deutschland-Stack
The Industrial AI Cloud serves as the foundation for what Deutsche Telekom and SAP are calling the “Deutschland-Stack” — a vertically integrated technology platform intended to give German and European organizations a sovereign alternative to US hyperscalers. Deutsche Telekom’s subsidiary T-Systems provides the infrastructure and platform layer through T Cloud, while SAP contributes its Business Technology Platform and AI applications on top, according to the NVIDIA blog.
Deutsche Telekom says its T Cloud Public platform already delivers 80 percent of the core functionality offered by leading US hyperscalers and is targeting full feature parity by the end of 2026, according to its T Cloud expansion announcement. The platform holds C5 certification, operates under a zero-trust architecture, and claims three times the certification density of competitors. It currently serves more than 4,000 enterprise customers, including DAX-listed corporations and public-sector organizations.
Early Customers and Use Cases
The facility was already operating at more than one-third utilization at launch, according to Deutsche Telekom. Early customers include Munich-based Agile Robots, which is training robotic foundation models on the platform, and PhysicsX, which uses it for engineering simulation to shorten product development cycles.
Siemens is integrating its SIMCenter simulation portfolio with the Industrial AI Cloud, with board member Cedrik Neike stating at the launch that the setup “drastically reduces” simulation times and calling it “not a promise for the future” but “already a reality,” as reported by Euronews. Mercedes-Benz and BMW are also running complex simulations with AI-driven digital twins on the platform, according to the NVIDIA blog.
Open-Source Language Model
Alongside the industrial platform, Deutsche Telekom announced the SOOFI project in partnership with Leibniz Universität Hannover — an effort to build a sovereign, open-source language model of approximately 100 billion parameters, covering European languages and trained entirely within Europe, according to Deutsche Telekom.
What We Don’t Know
Deutsche Telekom has not disclosed pricing for the Industrial AI Cloud, making it difficult to assess whether the sovereign premium will deter cost-sensitive customers accustomed to the aggressive pricing of AWS, Azure, and Google Cloud. The T Cloud platform’s target of 100 percent feature parity with US hyperscalers by year-end is ambitious, and the gap between 80 percent and full parity often contains the most complex enterprise features.
The facility’s 0.5 ExaFLOPS is substantial for a single European deployment but remains a fraction of the total capacity that US hyperscalers are building globally. Whether a single data center in Munich can meaningfully shift European AI workloads away from US providers — which operate dozens of regions worldwide — is an open question.
The SOOFI language model project, at 100 billion parameters, would enter a crowded field of open-source models. Details on its training data, timeline, and how it will compete with established alternatives from Meta, Mistral, and others remain unclear.
Analysis
The Munich facility represents the most concrete step yet in Europe’s long-running effort to build AI infrastructure that does not depend on American cloud providers. Deutsche Telekom CEO Tim Höttges framed the launch in blunt terms: “We are proving here that Europe can do AI,” as quoted by Euronews. German Vice-Chancellor and Finance Minister Lars Klingbeil reinforced the political stakes, calling the facility one that “strengthens digital sovereignty.”
NVIDIA CEO Jensen Huang, speaking at the original November 2025 announcement, described the concept as “factories of intelligence” and predicted that “every manufacturing company will have two factories — the factory for the car, and the factory for the AI,” according to the NVIDIA blog.
The timing is notable. Europe’s AI Act entered full enforcement in 2025, creating regulatory requirements around data handling that are easier to meet when infrastructure sits on European soil under EU law. The EU Chips Act is channeling billions into semiconductor manufacturing on the continent. And the political pressure to reduce technological dependence on the US has only intensified as American hyperscalers pour hundreds of billions into their own AI infrastructure — Meta’s $10 billion Indiana data center, Alphabet’s $20 billion bond sale, and Amazon’s $35 billion cloud expansion in India are all recent examples.
Whether the Deutschland-Stack can evolve from a compelling national project into a commercially competitive alternative for the broader European market will depend on execution, pricing, and whether other EU member states adopt or replicate the model.