Data is Becoming the Center of AI Strategy

Artificial intelligence is developing faster than the infrastructure, regulatory frameworks and organisational processes required to support it. Europe is talking intensively about data sovereignty, regulation and the need for its own capacity, but a significant gap remains between debate and implementation. “We talk a lot, but we do not do that much; that is essentially the premise at the moment,” Marko Dagelić, CEO of Datum, tells ICTbusiness Media - ICTbusiness.info. For him, control over processes and data is one of the foundations of serious AI deployment.

Dagelić argues that the data-centre debate cannot be separated from the business case supporting an investment. Building infrastructure simply because global demand for AI capacity is rising is not enough; investors need to know who will use it, how efficiently it will be utilised and what economic rationale supports the project. “The question is the business case and the logic behind it, even at European Union level,” he says.

To illustrate the infrastructure gap, he compares the United States and Europe. According to the figures he cites in the interview, the US has more than twice as many data centres as the EU, while the difference in computing power is even greater. “When you put it in the context of computing power, Europe has only around five percent,” Dagelić says, adding that southeastern Europe has an even smaller place on the global map. At the same time, he believes Croatia’s current capacity is sufficient for present domestic needs.

The challenge lies in the mismatch between development cycles. AI models, tools and use cases can change from day to day, whereas serious infrastructure takes years to plan and build. “For serious infrastructure, we are not talking about days or weeks; we are talking about years,” he says. Long-term capacity therefore requires broader coordination between the state, institutions, researchers, industry and citizens.

One possible route involves public-private partnerships or joint ventures with a clear commercial rationale. “We all need to come together, agree on possible public-private partnerships or a joint venture and move in a direction that has business logic,” Dagelić says. Data must sit at the centre of such projects: where they are located, who controls them and under what conditions they are processed.

Corporate awareness is rising alongside the practical use of AI. Larger companies are already testing different models more intensively and learning through real projects where the technology can help. “Companies increasingly use different models and see in which segments AI can genuinely help,” he says. Once experimentation begins, control becomes the next question: if a company wants its own agent for a particular process, it wants to know exactly what environment the agent runs in and where the data go.

These requirements are particularly strong in regulated industries such as healthcare, telecoms, insurance and banking. Yet less regulated sectors are also starting to understand the need for controlled environments. Data sovereignty is not the only reason. “It is also about cost control,” Dagelić explains, because broad and unmanaged use of AI services can steadily increase total expenditure even when an individual service initially appears inexpensive.

The next phase will bring wider use of agents in business processes. Dagelić expects agents to take over specific tasks and processes, but he does not see human oversight disappearing. “There will always be a human in the loop, more in some areas and less in others,” he says. The amount of human control will depend on the type of process, risk, data sensitivity, and consequences of an error.

Regulatory uncertainty comes with deployment. The AI Act establishes a new framework, but companies still need to understand what specific obligations mean for their systems and processes. Dagelić says the industry itself is still looking for answers and that practical issues are gradually being clarified through expert workshops and discussions with European institutions. “There will be plenty of questions and a search for answers,” he expects.

Another major issue is the security of the agents themselves. As AI receives more authority over processes, organisations have to ask whether an agent can be compromised, whether it can move outside defined boundaries, and how such scenarios can be contained. “Once broader deployment starts, two important issues will be how we regulate it on paper and how we defend against possible attacks or an agent escaping its boundaries,” Dagelić concludes. AI strategy is therefore evolving into a combination of infrastructure, data governance, regulation, security and business economics rather than a standalone technology project.