Telekom's AI Bet: Salvation for SMEs or Costly Delusion?

KI & Automatisierung · 8. Februar 2026 · Manuel Krapf

Telekom is building an AI fortress in Munich. But is it the long-awaited game-changer for German SMEs, or just a prestige project for large corporations?

Last week, I was visiting a mechanical engineering company in the Black Forest, a classic hidden champion. We were discussing AI in production control. Suddenly, the CEO, a man whose family has run the business for three generations, leaned across the table and said in an almost whispered voice: "Mr. Müller, I'd love to do more. We have the data. But should I really upload the plans for our new gearbox – our crown jewels – to a US server? With all due respect, I wouldn't sleep soundly at night."

Boom. There it was again. This deep-seated unease, this conflict within German industry. On the one hand, the absolute will to be at the technological forefront, and on the other, the sheer fear of data leakage and the loss of sovereignty. Everyone knows that AI is the next big lever, but no one wants to hand over control to American hyperscalers, who are, in case of doubt, subject to the laws of the Patriot Act. This is precisely where a project comes in that is either a brilliant move or an incredibly expensive misunderstanding: Deutsche Telekom has built an "Industrial AI Factory" in Munich. And it aims to be nothing less than the digital declaration of independence for Europe's industry.

The AI Fortress on the Isar: What's Behind It?

Let's be honest: The numbers initially sound like something out of a science fiction novel written in the Chancellery. In just six months, Telekom has erected a data center in Munich's Tucherpark – equipped with almost 10,000 of the latest NVIDIA Blackwell GPUs. This is not just any server basement; it's a high-performance monster with a computing power of 0.5 ExaFLOPS. To put that into perspective: This could theoretically run an AI assistant simultaneously for all 450 million EU citizens. This thing is a statement.

The whole thing has been christened "Deutschland Stack." An alliance of Telekom as the infrastructure operator, SAP for business software, and NVIDIA as the supplier of the horsepower under the hood. The idea is simple and tempting: a secure, data-sovereign platform that plays entirely by EU rules – GDPR and the upcoming AI Act send their regards. Data remains in Europe, under German control. And because we are in Germany, there is, of course, the green fig leaf: It is cooled with water from the Eisbach (no joke), the electricity comes from renewable energies, and the waste heat is eventually intended to feed the district heating network. Sustainability as a selling point – clever.

But this is not future music. Cedrik Neike, who is responsible for Siemens' digital business, put it succinctly: "This is not a promise for the future. In Munich... it is already a reality." Siemens is one of the first customers and uses the computing power to massively accelerate the development of digital twins. Instead of waiting weeks for simulation results, it now happens – so the promise goes – in hours. In addition to Siemens, exciting smaller players such as the robotics specialist Agile Robots and the simulation experts from PhysicsX are also on board. Utilization was already over a third at the start. This shows that the demand is there. The only question is: for whom?

A Comparison: Sovereign AI vs. American Hyperscalers

The opening of Deutsche Telekom's AI factory is good news for Germany and Europe. We need such lighthouse projects to strengthen Europe's technological sovereignty.

— Prof. Antonio Krüger, Managing Director German Research Center for Artificial Intelligence (DFKI)

Between Workbench and World Power: The Geopolitical Context

This project cannot be viewed in isolation. It is a piece of a much larger game. While we in Germany discuss the right heat pump manufacturer, the USA and China have long since turned their tech giants into geopolitical instruments. There is a fierce vertical integration happening: from the chip factory to the cloud platform to the AI application – everything from a single source, under national control. Europe was threatening to become a digital recipient of alms, dependent on the whims of other powers for better or worse.

The Munich AI factory is therefore also a political statement. Finance Minister Lars Klingbeil spoke of an "important pillar for the German and European AI ecosystem." This is no exaggeration. It's about regaining control over the most important resource of the 21st century – computing power for AI. This also fits with other initiatives, such as the German-French-Italian plans for joint stockpiling of critical raw materials such as lithium or rare earths. It has been understood: Without its own robust technological and material basis, "Made in Europe" will soon be nothing more than a nostalgic memory. There's no getting around that.

But Beware: Pitfalls Lurk Here

So far, so good. Champagne corks are popping in Munich and Berlin. Tim Höttges, Telekom's CEO, proclaims: "Many can talk. Telekom acts." That may be true, but applause alone doesn't fill order books for SMEs. In my experience, practical hurdles are often swept under the rug in such lighthouse projects. And there are quite a few of them.

First: The costs. Sovereignty is not cheap. I bet the first price lists for computing time on this super cluster will bring tears to the eyes of many SME COOs. Telekom has to recoup its multi-billion euro investment. Can a 200-person company from Westphalia really afford this, or is it ultimately just a club for Siemens, SAP, and the German automotive industry? Second: The complexity. A bunch of GPUs is not yet a functioning AI application. The biggest bottleneck in German industry is not the lack of ExaFLOPS, but the lack of qualified employees who know what to do with them. The problem is not the hardware; it's the "wetware" in the engineers' heads. And third: The focus. The "Deutschland Stack" is strongly geared towards the SAP ecosystem and large industrial applications. What about the countless companies that work with other software or have very specific niche problems? The danger is that a golden, but also closed, world will emerge here, while open innovation takes place elsewhere. One must not throw the baby out with the bathwater and run into a sovereign but isolated dead end.

What You as an SME Must Do Now – and What Not To Do

1. Step 1: Conduct a data audit. Forget the AI factory for a moment. Do your homework. Do you even know what data you have, where it is, and what quality it is? Identify a specific dataset that could solve a real business problem (e.g., machine data for predictive maintenance, scrap rates from quality control).
2. Step 2: Define a manageable pilot project. Don't start by trying to make the entire production autonomous. That's nonsense. Start with a small, clearly defined project. Example: AI-supported optimization of energy consumption for a single production line. The learning effect here is more important than the immediate ROI.
3. Step 3: Ruthless competence check within your own company. Do you have someone on your team who speaks the language of data scientists and manufacturing engineers? No? Then that is your most urgent task. Look for external support, train your own people, but don't think you can simply delegate this topic to the IT department. This is a CEO issue.
4. Step 4: Actively gather information. Don't drive blindly. Talk to providers like Telekom, but also to their competitors. Visit Hannover Messe and specifically ask about use cases for companies of your size. Join working groups at VDMA or ZVEI. Listen, learn, and then decide – with a cool head.

My Conclusion: A Wake-Up Call, Not a Cushion

So, what's the bottom line? Is the AI factory on the Isar a big hit? Yes – and no. Yes, it is a damn important and correct signal. A wake-up call. It shows that Germany and Europe are willing not just to watch, but to play along in the technological race. The political and symbolic significance of this project can hardly be overestimated. But – and this is the big but – it is only the provision of a racetrack. The companies themselves have to win the race. And there I still see far too much hesitation, too much "we've always done it this way" mentality, and a blatant lack of skilled workers.

Telekom has done its homework. Now it's the SMEs' turn. This infrastructure is an opportunity, but not a guarantee of success. Anyone who now thinks they can lean back and wait for Telekom to serve them the finished AI solution on a silver platter is greatly mistaken. The horse will ultimately still have to be saddled by the companies themselves. I bet that in three years we will look back and see who really got going. And I can tell you now: The list of winners will be significantly shorter than many hope in their euphoria today.