Europe Is Building The AI Infrastructure And The Harder Part Is Making It Work

Europe has spent the past year sharpening its AI ambitions. The European Commission’s AI Continent Action Plan has put real weight behind the idea that the region must strengthen its AI capacity through infrastructure, data access, skills and regulatory simplification. On paper, it is the kind of industrial-scale response many have been waiting for.

But for all the focus on compute, there is a less visible problem that may prove more decisive.

According to Milos Rusic, CEO of deepset, Europe is moving quickly to address the infrastructure challenge while many enterprises are still stuck with a connectivity challenge. The issue is whether AI systems can reliably connect to the internal knowledge that makes them useful in the first place, not just whether businesses can access more computing power.

That distinction matters. A company can have access to world-class models and significant compute resources, but if the data those systems need is buried in legacy repositories, split across teams, or restricted by fragmented governance structures, the result is unlikely to be trustworthy at scale. In that environment, AI does not fail because it is slow. It fails because it is disconnected.

That is the part of the market Rusic has been focused on through deepset, a Berlin-based company known for Haystack, its open-source framework for building AI applications, and for its work on enterprise AI systems designed for production use. His perspective reflects a broader challenge many organisations are now facing as they move from experimentation to operational deployment.

Where Europe’s AI Push Could Run Into Trouble

The Commission’s emphasis on AI Factories and broader infrastructure is understandable. Europe wants to strengthen its position in AI and reduce the risk of falling behind in the global race for capability and scale, but infrastructure by itself is only one layer of the problem.

For enterprises, the real test comes later when a system has to retrieve the right internal document, operate within strict access controls, distinguish between what one team is permitted to see and what another is not, or show where an answer came from and provide enough traceability for compliance, oversight, and trust.

These are not side issues, but often the reason promising AI initiatives begin to lose momentum once they leave the pilot stage.

Rusic’s argument is that Europe risks building impressive AI capacity without fully addressing the practical layers that determine whether companies can use that capacity effectively. If organizations cannot safely bridge their own trusted information into these systems, the compute layer may advance faster than enterprise adoption itself.

Why The Data Union Story May Matter More

This is why the conversation around Europe’s new data agenda may be more important than many of the infrastructure headlines suggest.

The emerging focus on a stronger data framework, including efforts to make data more accessible and usable across organisations and borders, points to a more grounded understanding of what AI adoption requires. Enterprises do not just need access to models. They need access to their own knowledge in a way that is secure, governed, and usable.

That is where Europe may have a genuine advantage.

The region already holds enormous volumes of high-value industrial, engineering, scientific, and public-sector data. In many cases, the issue is accessibility, not scarcity. The challenge is making that knowledge available to AI systems without undermining security, compliance, or operational control.

For Rusic, this is where retrieval becomes central. Europe does not necessarily need to win by competing on hype or by chasing every development in the foundation model race. Its strength may lie in helping enterprises connect AI to highly specific, high-quality domain knowledge that already exists inside their own environments.

The Shift From Restriction To Application

There is also a broader shift taking place in the policy and business conversation around AI. For some time, much of the debate in Europe has centered on risk, restriction, and what organizations might not be allowed to do. Those questions were always important, but they were never enough for companies trying to build something practical.

Now the emphasis is starting to move toward implementation, which is a meaningful change. The discussion is increasingly becoming about how AI can truly be deployed inside real institutions with real constraints.

That means asking harder questions:

• How do companies connect private documents to public or external compute environments safely?
• How do they govern access across different teams and systems?
• How do they ensure outputs are grounded in trusted internal sources?
• How do they build AI in a way that can survive procurement, oversight, and internal security review?

These are the questions that will determine whether Europe’s AI ambitions become operational reality.

A Practical View From Inside The Market

In that sense, Rusic’s role in this conversation is not as the story itself, but as a useful voice within a much larger one.

He represents a strand of the AI market focused less on headline-grabbing model performance and more on what it takes to make AI useful inside real organisations. Through deepset, his attention has been on the architecture beneath enterprise AI: retrieval, orchestration, permissioning, and the systems required to make outputs more trustworthy and production-ready.

That perspective is relevant now because it points to the gap between AI capability and AI usability. Europe may succeed in accelerating the first without fully unlocking the second.

And if that happens, the region could find itself in an uncomfortable position with better infrastructure, stronger policy momentum, and still too many enterprises unable to operationalise AI at scale.

The Bigger Risk For Europe

The real risk is not that Europe will underinvest in AI infrastructure, but rather that it will invest heavily in the visible parts of the stack while the less visible barriers remain unresolved.

If governance stays fragmented, if enterprise knowledge remains trapped in silos, and if permissioning models are too inconsistent for AI systems to navigate effectively, then many businesses will continue to struggle with the same core problem of getting AI to work reliably in practice.

That is why the next phase of the conversation matters so much. Europe is building the machines but unless it also makes knowledge more retrievable, governable and usable, many enterprises will still find themselves unable to translate AI ambition into real deployment.

And that may be the issue that determines whether Europe’s AI strategy delivers lasting competitive advantage or simply impressive infrastructure with limited operational impact.