Sovereign AI: Europe's Billion-Dollar Deal from Munich's Vault

KI & Automatisierung · 17. Februar 2026 · Joseph Flesh

Telekom and Nvidia open Europe's most powerful sovereign AI cloud. What does this mean for German SMEs? Find out your next steps here.

I recently stood in Munich's Tucherpark, where heavy steel doors once protected a bank's gold and securities. Today? Today, it hums and buzzes down there. Instead of gold bars, server racks are stacked, cooled by the water of the nearby Eisbach. A strange sight. An old vault, repurposed as the epicenter of a new industrial era. One inevitably asks: What is more valuable – yesterday's gold or tomorrow's data?

Tim Höttges from Telekom and his partners from Nvidia answered that question quite unequivocally at the beginning of February. Because this vault is no ordinary data center. No, this is the new "Industrial AI Cloud." Worth a billion euros. Equipped with almost 10,000 of the latest Nvidia Blackwell GPUs. A computing monster that delivers almost half an ExaFLOP of performance. Honestly: These are numbers that raise eyebrows even in Silicon Valley. The thing is: This isn't just about flexing computing power. This is about a strategic declaration of war. It's about sovereignty. More precisely: about sovereign AI for German and European industry.

What the new Industrial AI Cloud means for manufacturing

Let's be clear. What does this tech giant in the basement mean for you as the CEO of a medium-sized mechanical engineering company or automotive supplier? First and foremost, an alternative. Until now, the equation was simple: If you wanted to seriously do AI – for predictive maintenance, for optimizing your supply chains, or for digital twins of entire production facilities – you had to go to the Americans. To AWS, to Microsoft Azure, to Google Cloud. There's no getting around that. But that also meant: Your own, often highly sensitive production data, the crown jewel of every manufacturer, was on US servers. And thus potentially within the access range of the notorious CLOUD Act. A risk that many CEOs I've spoken to recently have reluctantly taken.

This new facility in Munich, operated by T-Systems, approaches the problem differently. The magic word is "Deutschland Stack." A combination of T-Systems infrastructure, the T-Cloud, SAP's Business Technology Platform, and Siemens' simulation tools. The promise: Your data remains in Germany. Under German sovereignty, according to EU law. No US law can just sneak in through the back door here. For Industry 4.0, which is fundamentally data-driven, this is – and I rarely use this word – a real game-changer. Suddenly, digital twins (Siemens is on board) or complex robot controls (Agile Robots, also a launch customer) can be developed and operated on a European platform, without selling your soul, i.e., your data.

The bare figures in comparison

To put this into perspective: This is not a small, nice niche solution. Telekom and Nvidia have really gone all out here. And the demand seems to be there. Over a third of the capacity was already booked at the opening. And not with state-funded research projects, but with paying industrial customers at commercial rates. This shows: The need for a sovereign AI solution is real. But how does the offering compare directly?

Voices from practice: A necessary, but expensive step?

I spoke on the phone last week with the CEO of a South German automotive supplier (the name doesn't matter here, you know the drill). He told me: "Mr. Müller, we've been simulating our crash tests and manufacturing processes in the cloud for years. Every time we upload the data, it feels like we're handing over the blueprint for our best racehorse to the competition. This Munich cloud is the first option where I can sleep soundly again at night." That pretty much sums up the mood in SMEs. It's a mix of relief and concern.

For German SMEs, the availability of a high-performance and sovereign AI infrastructure is not an option, but a matter of survival. We cannot afford to build the next wave of industrial value creation on platforms that we do not control. The Industrial AI Cloud is a crucial puzzle piece to secure the competitiveness of German manufacturing.

— Prof. Dr.-Ing. Anja Weber, Head of the Fraunhofer Institute for Production Technology and Automation IPA

The concern? The costs. Höttges emphasizes "commercial rates," but what that means for a company with 200 employees is another matter. The investment in AI is not just the rent for GPU time. You need people who can train models, prepare data, and program interfaces. That's the real chunk. The new cloud solves the availability problem – not the competence problem.

The European context: A lonely lighthouse or the beginning of a fleet?

This project should not be viewed in isolation. It is part of a larger European effort to regain digital sovereignty. Initiatives exist all over the continent. Think of the European Processor Initiative with chips like the SiPearl Rhea1 or the DARE consortium working on European RISC-V chiplets. The problem is – and we have to be honest – these projects are years away from being able to compete with Nvidia. They are important, absolutely. But they are future music.

Telekom and Nvidia are taking a pragmatic approach here. They are saying: Instead of waiting for the perfect European chip and losing ground, we are now building a sovereign European platform with what is best (Nvidia hardware). The data stays here, the software structure (SAP, Siemens) is European, and operations are in German hands. This is a compromise. But perhaps the only realistic one at the moment. Telekom is already eyeing the next steps, an "AI Gigafactory" that is applying for funds from the EU's 20-billion-euro funding pot. So Munich is probably just the beginning.

The critical view: How sovereign is truly sovereign?

Now let's get down to brass tacks. Is an AI cloud, whose heart consists of 10,000 American high-tech chips, truly "sovereign"? It's a matter of definition. If Washington decides tomorrow to restrict the export or support of these GPUs to Europe – what then? Then Germany's most expensive and quietest vault will stand in Munich. That is the Achilles' heel of the entire concept. We are exchanging a software dependency on US hyperscalers for a hardware dependency on Nvidia. Whether that is a good trade will be shown by the geopolitics of the coming years.

I also doubt whether this will be so easily accessible for the broad SME sector. Siemens, Agile Robots, EY – these are the financially strong initial customers. No SME from the Sauerland region among them. T-Systems will have to prove that they can also put together smaller, flexible, and above all affordable packages. Otherwise, we will end up with a wonderful AI highway where only corporations are allowed to drive, while SMEs continue to trundle along on the country road. That would be fatal.

Your roadmap: 5 concrete steps for getting started with sovereign AI

Sitting around and waiting is not an option. The competition – in Asia and the USA – is not sleeping. Here are five pragmatic steps that you as a CEO or COO should initiate now:

1. 1. Conduct a potential analysis: Where in your company lies the greatest AI potential? Is it quality control in production (keyword: computer vision)? Predictive maintenance of your machines? Or optimizing setup times? Identify a concrete, painful use case.
2. 2. Inventory your data: Do you even have the necessary data? And what is its quality? AI is only as good as the data it is fed. Start a project for the systematic collection and cleaning of your most important production data.
3. 3. Define a pilot project: Choose a small, manageable process. The goal is not to immediately revolutionize the entire company, but to learn. Start a pilot project for predictive maintenance for a single, critical machine. The risk is limited, the learning effect enormous.
4. 4. Conduct informational interviews: Pick up the phone. Talk to T-Systems, to Siemens, to SAP. Ask about entry-level packages, test installations, and reference customers from your industry. Let them show you what is concretely possible – and what it costs.
5. 5. Qualify employees: Send your best engineers and IT people to training courses. Build internal know-how. The best infrastructure is useless if no one in the company can operate it. Invest in minds, not just computing time. That is the most sustainable investment of all.

Frequently Asked Questions about the Industrial AI Cloud

Is this AI cloud only for large corporations?

Officially, no. T-Systems has emphasized its intention to create offerings for SMEs as well. Reality will have to show. In my experience, such platforms are often optimized for financially strong large customers at the beginning. SMEs should team up or exert pressure through industry associations to obtain suitable and affordable access.

Do I have to abandon my existing cloud strategy now?

No, absolutely not. The cart is often put before the horse here. It's not about an 'either-or'. A hybrid strategy is probably the best way. Non-critical applications can continue to run on the cheaper US hyperscalers. But for the crown jewels – i.e., your production, R&D, and customer data – you should seriously consider a sovereign option like the one in Munich.

Your Compass for AI Investment: The ICP Playbook Before investing in new AI technology: Do you know exactly which customer problem you are solving? Our ICP Playbook helps you define your ideal customer and their most urgent needs. The basis for every successful technology strategy.

My Conclusion: An expensive, but indispensable wake-up call

At the end of the day, the opening of this facility in the old bank vault is more than just a press release. It is a symbol. A sign that Europe does not want to give up in the global AI race. Tim Höttges' exclamation "Europe can do AI!" may sound a bit like whistling in the dark, but it has a kernel of truth. We can do it – if we finally get to work. The solution is not perfect, the dependence on Nvidia hurts. But it is a pragmatic, powerful step forward. Better a sovereign platform with American chips than no platform at all.

I bet that in three years we will see a dichotomy in the market: On one side, companies that operate their critical processes on sovereign platforms like this, thereby protecting their data and know-how. On the other side, those who, for cost reasons or convenience, have remained with US providers and thus have a permanent strategic risk on their balance sheets. The question that every CEO in the DACH region must ask themselves today is simple: On which side of this line do I want to stand in 2029?