Digital sovereignty in the age of AI: what European leaders need to know
AI is changing how organisations work. It also raises hard questions about who controls your data, your infrastructure and your supply chain. In our first episode, host Federico Winer talked these through with Dr Thomas Wollmann, Managing Director of Merantix Momentum and a lecturer at the Hasso Plattner Institute.
AI is moving from tools to coworkers
Thomas sees this as the year autonomous systems really take off. Picture a development team of five with a sixth member: an AI agent. It checks code quality, writes patches and handles releases in the background. Your team stays in charge, and the routine work happens on its own.
Sovereignty means having a choice
Sovereignty isn’t one single thing. It’s the ability to choose. For one organisation that means keeping data in-house. For another it means a supply chain that trade bans or geopolitics can’t cut off. For a third it means a model that reflects European values. Your priorities decide what sovereignty means for you. What matters is that you aren’t locked in.
You don’t always need a frontier model
The headlines go to the biggest models, but many everyday business tasks don’t need them. Soofi takes a different approach. Thomas compares it to a motorway: publicly funded infrastructure that the wider economy builds on.
Unlike many “open” models, Soofi is fully open source, not just open weight. You get the data and the training recipe as well as the finished model. Its first model has 30 billion parameters. It’s strongest on German-language documents: extraction, summarisation and retrieval. It also handles short coding tasks and agent workflows. Larger models built on the same foundation are already planned.
Hardware is the hardest layer
European cloud providers already give you an alternative to the US hyperscalers. Energy and chips are harder problems that will take years to solve. In the meantime, compression techniques such as quantisation and pruning let models run on smaller, easier-to-find hardware. That gives you more freedom, lower costs and faster performance.
The AI Act matters less than copyright
Thomas isn’t worried about the AI Act. Highly regulated sectors like finance and medical devices already benefit from clear rules. He sees European copyright law as the bigger barrier. Without fair-use rules like those elsewhere, training large models in Europe is far more complicated.
Security is now about speed
AI speeds up attackers and defenders alike. Attackers can gather information at scale, find weak spots and personalise phishing in minutes. Defenders can spot unusual traffic and trigger contingency plans automatically. Thomas suggests one question for reviewing every policy you have: “Am I prepared for the speed?”
Build or buy? Follow the uniqueness of the problem
If many organisations share your problem, buy. Intrusion detection, for example, gets better because the vendor learns from all its customers. If the process is unique to you and costly to get wrong, a tailor-made solution often pays off. Revisit these decisions regularly, because what the technology can do keeps changing.
Watch the full session on demand right here on this page.
Let’s talk possibilities!
Every project on this page started with a conversation about a real operational challenge. Bring us yours, and let’s explore what a working prototype could look like.