The rise of artificial intelligence is increasingly raising questions about control over data, infrastructure and business processes, rather than simply choosing the most advanced model. “The real value of AI does not come when a model answers a question, but when it reliably completes a task and delivers a result that justifies the investment,” Marko Dagelić, CEO of Datum, tells ICTbusiness Media – ICTbusiness.biz. Datum is developing a model that combines computing resources, cloud environments, an AI platform and agents, with its partnership with DataBox providing a foundation in domestic infrastructure.
In this interview, Dagelić explains the difference between storing data in Croatia or the European Union and achieving genuine operational independence from a particular technology provider. He also discusses practical applications of agentic automation, from WhatsApp bookings to business document processing, alongside security and human accountability. The conversation addresses GPU system costs, energy availability, capacity utilization and the conditions under which investment in AI data centers can be sustainable. For Croatian technology companies, he sees an opportunity in combining infrastructure, systems integration expertise and regional demand, rather than trying to compete head-on with the world's largest platforms.
Datum is developing a model that connects computing infrastructure, an AI platform, models and business applications. What does sovereign AI actually mean for customers in Croatia and across Europe?
I think we need to distinguish between two things: using artificial intelligence and retaining control over the systems on which it is built.
Today, any company can use global AI models, and that is a good thing. We do not need to develop everything ourselves. But once AI becomes part of critical business processes, we need to know where data is processed, who can access it, who operates the infrastructure and what happens if a provider changes its prices or terms of service.
That is why I do not see sovereign AI simply as a question of where a data center is located. It means controlling data, access, processing rules and the way AI is embedded in business operations.
At Datum, we are developing a fully controlled approach that brings together infrastructure, cloud services, AI models, a platform and agents. Our ambition is to enable customers to deploy AI in a controlled environment without having to integrate every part of the technology stack themselves.
I do not believe Croatia should try to replicate the American technology giants. We need to build capacity that makes economic sense for our market and connect it to regional and European demand.
What is the difference between controlling where data is stored, managing AI models and achieving genuine independence from external providers?
Data location is only the first layer. The fact that data is physically stored in Croatia or the European Union does not automatically mean we have full control over how it is processed, who can access it or the technology that uses it.
The second layer is model governance: can we choose the model, decide where it runs, control its permissions and usage policies, and monitor how it processes data? The third layer is operational independence. Can we switch models or service providers without having to fundamentally redesign our business processes? Can we continue operating if a platform's prices, terms or availability change?
That does not mean we have to be completely independent of global technology companies. That is neither realistic today nor always economically justified. The goal is to avoid dependence on a single provider and retain control over what is critical to the business. Having options is what matters.
In which areas can domestic infrastructure compete with global platforms on price, performance and service quality?
We cannot compete with global platforms in every area, particularly when it comes to the scale of investment in the largest models and computing capacity. But we do not need to win in every category to be competitive.
We can differentiate ourselves through integration quality, continuous local support, controlled environments, knowledge of regulatory and business requirements, and the ability to connect AI to specific customer processes.
For business customers, the price of an individual model is not the only deciding factor. What matters is the total cost of the solution, its reliability and security, and the results it delivers. I see an opportunity for Datum and other Croatian technology companies in bringing those elements together.
Datum is working with DataBox to integrate data center infrastructure, GPUs and an AI platform. What does this integrated approach offer customers, and how far has the model progressed?
AI is often presented as software that can simply be switched on. In reality, every serious AI system depends on computing capacity, power, cooling, networking, data, models and the software that connects everything.
Our partnership with DataBox gives us a physical infrastructure foundation in Croatia, while Datum develops the platform and business solutions on top of it. Our approach is to connect infrastructure, models and agents with integration into business processes.
The important thing for customers is that they do not have to bring together fifteen different suppliers and take responsibility for every integration themselves. They can start with a specific business problem and then select the right infrastructure, model and deployment approach.
The development of this model should be viewed as the integration of several layers that create value together. Infrastructure alone is not enough; you also need a platform, a business application and customers who are willing to pay for the service.
Which services and capacities can you offer today, and which are still being planned?
Our business focus is on connecting computing infrastructure and cloud environments with an AI platform, models and business solutions. In addition to infrastructure, an important part of our approach is deploying our own AI agents, integrating them with existing systems and tailoring them to customers' specific needs. There are far too many AI companies today that are essentially “wrappers” or resellers. We develop and deploy our own solutions on our own infrastructure. That provides both a competitive advantage and a measure of protection for every customer.
Examples of applications we have already implemented include a WhatsApp agent and agents that assist with processing and checking business documentation.
With infrastructure services, it is important to distinguish between existing capabilities and capacity that is still under development or being planned. Specific GPU configurations, available quantities, timelines and commercial terms should reflect what has actually been confirmed at the time of publication. I would not announce capacity that is not yet operational.
In the long term, we want to offer customers an integrated service, from computing infrastructure through to the use of AI in a real business process.
At the same time, customers retain full control over their data and agent training, as well as their costs.
What are the biggest challenges in financing, power supply, GPU utilization and day-to-day operations?
The biggest challenges are the cost of GPU hardware, the availability and price of electricity, cooling, skilled professionals, capacity utilization and rapid technological obsolescence.
With investments of this kind, buying equipment and expecting demand to appear is not enough. We need to know who will use the capacity, for which workloads, how frequently and at what price.
That is why I believe infrastructure investment must be tied to the development of business applications. If we have GPU capacity but no customers or revenue-generating services, we have an expensive asset, not necessarily a sustainable business model.
Operational management is another challenge: ensuring reliability, security, efficient capacity utilization and service quality. Connecting infrastructure with a platform and specific use cases can help us manage these challenges more effectively.
One of Datum's development priorities is a platform for managing AI agents. Which business processes are best suited to agentic automation today?
I see the greatest potential in repetitive processes that involve high volumes of requests and require information from multiple sources to be brought together.
It is important to understand the difference between a chatbot and an AI agent. A chatbot provides an answer. An agent can be assigned a task, verify data, apply business rules, communicate with other systems and complete work within defined permissions.
At Datum, we are not building AI simply to give employees another chat window. We are building it to get work done with less manual effort, while maintaining control and the ability to verify results. The most promising areas include handling inquiries and bookings, customer support, document verification, invoice preparation, and gathering and comparing data across business systems.
Even so, not every process should be automated just because the technology makes it possible. The starting point should be a problem that occurs frequently enough and is costly enough to justify automation commercially.
Can you give an example where the impact on processing time, cost or service quality can be measured?
The first example is Datum Booking Agent on WhatsApp.
A customer requests an appointment or reservation through WhatsApp. The agent understands the request, collects the necessary information and links the conversation to the booking process. This enables a business to respond outside normal working hours while reducing the burden of repetitive questions on employees.
The impact can be measured by the time between the initial inquiry and confirmed booking, the proportion of successfully completed bookings, the number of requests handled and the share of cases requiring human intervention. The agent is also available at any time, including holidays, and can communicate in the customer's preferred language. That alone can improve productivity and service quality.
The second example is an AI agent for checking and preparing invoices. It compares information from business documents and ERP systems, flags discrepancies, identifies missing information and prepares entries for further processing. This reduces manual checking and the risk of errors.
In both cases, success should be demonstrated by comparing results before and after deployment. It is important to measure not just speed but also quality, reliability and the overall cost of the process.
How do you manage agent permissions, audit trails, data protection and human accountability for actions taken by AI agents?
Security cannot be an add-on introduced at the end. It has to be part of the architecture, from infrastructure and data access to AI agent permissions and activity logs.
An agent should have only the permissions it needs for a specific task. Answering a question about appointment availability is very different from changing records in a business system, sending a binding offer or initiating a payment. That is why access controls, activity logs, auditable decisions and clear rules for when an agent must seek human approval are essential. For critical or financially sensitive actions, human oversight must remain clearly defined.
AI can take over the execution of some tasks, but accountability cannot simply be transferred to a model. The owner of the business process must know what the agent is allowed to do, where the limits of its autonomy lie and who steps in when the system makes a mistake.
Building AI data centers requires substantial investment and brings the risk of rapid technological obsolescence. How do you assess the profitability and long-term sustainability of such infrastructure in Croatia?
AI infrastructure demands substantial capital investment, and the technology is changing extremely quickly. Investment decisions therefore cannot be based solely on forecasts of market growth.
At Datum, we want to connect infrastructure with our platform and specific business solutions. That means we are not relying exclusively on selling computing resources; we are building the ability to provide a service that solves a real customer problem. For every investment, we need to assess expected demand, energy costs, financing and maintenance expenses, the pace of technological change and whether the equipment will remain commercially viable. It is particularly important to avoid building capacity before there is sufficient demand.
Croatia can participate in this market, but investments must make economic sense. Market size alone does not guarantee profitability, just as the technological appeal of an investment does not mean it will deliver an adequate return.
One avenue I personally see potential in is public-private investment in infrastructure.
When does it make more sense for a business to purchase its own GPU systems, and when is using computing capacity as a service the better option?
There is no universal answer. Owning GPU systems can make sense when a customer has stable, predictable workloads, clear security requirements, a skilled team and sufficient utilization to justify the capital investment.
Using capacity as a service is often more rational when demand fluctuates, for scale-up companies, when AI projects are still being developed or when a customer does not want to tie up substantial capital in equipment that becomes obsolete quickly.
The comparison needs to be based on total cost of ownership, not just the purchase price of the hardware. That includes electricity, cooling, maintenance, staffing, utilization and the eventual replacement of equipment.
For many businesses, the greatest value will not come from owning GPUs but from having access to reliable computing services that can be turned into business results.
Datum can cover all the options: we can install our own GPUs and technology at the customer's premises, or connect directly to the customer through a dedicated link. The outcome is the same.
What level of demand do you expect in Croatia and the wider region, and which utilization and return-on-investment indicators matter most?
I expect demand to grow as AI moves beyond experimentation and becomes part of everyday business processes. But demand should not be estimated solely from interest in AI or the number of projects launched. What matters is how many customers are prepared to pay for a service and how much they will actually use it.
That is why the market needs to be viewed regionally. Croatia alone may not generate sufficient demand for every type of infrastructure, but bringing together customers from several countries can expand the addressable market and improve investment economics. A good example is the recently signed Prague Declaration, in which Central and Eastern European countries agreed on a regional approach and cooperation.
I would monitor GPU utilization, revenue per unit of capacity, energy costs, gross margin, contracted demand and return on invested capital. The structure of contracts also matters: how much revenue comes from long-term commitments and how much depends on variable demand.
Investment discipline is essential. Capacity should not be built simply because market growth is expected; investment must be aligned with demonstrable demand and realistic return expectations.
Can Croatia become a regional hub for sovereign AI infrastructure, or will it remain primarily a customer of major international platforms?
I believe Croatia has an opportunity, but not if we think only in terms of the size of our domestic market.
We cannot expect to finance every layer of AI infrastructure ourselves or compete with global companies in every field. But we can bring together existing infrastructure, technical expertise, domestic systems integrators and the needs of customers across the region.
Croatia can build a position in segments where the combination of infrastructure, business integration, security and an understanding of the local market matters. We do not have to develop everything ourselves, but we do need the ability to bring technologies together and turn them into services the market needs.
The partnership between Datum and DataBox is part of that approach: physical infrastructure connected with a platform and AI solutions. The next step is to build sufficiently large and stable demand to make the model sustainable beyond the domestic market.
What is the biggest constraint on development: capital, electricity, talent, regulation or market size?
There is no single constraint. Capital is needed for infrastructure, energy shapes part of the operating economics, and skilled people are essential to developing and running the systems. At the same time, the domestic market may not be large enough to justify every investment on its own.
Regulatory requirements also matter, particularly in sectors that process sensitive data or use AI in critical processes. But regulatory compliance can also become a competitive advantage if security is built into the service from the outset.
In my view, the biggest challenge is bringing all these elements together in a sustainable business model. We can secure funding and build infrastructure, but without customers, high-quality services and sufficient capacity utilization, the investment will not be sustainable.
How should domestic data centers, systems integrators and AI service providers work together to deliver a competitive offering in the European market?
They should collaborate around creating value for customers rather than each trying to build the entire technology stack independently.
Datum, for example, was established as a joint venture between several technology companies to offer faster, better and more accessible services. Datum is also the first Croatian company to join the CEE AI Chamber. Initiatives of this kind provide a valuable foundation for cooperation and ultimately competitiveness.
Data centers can provide the physical infrastructure and computing capacity. Platform providers can connect infrastructure, models and agents, while systems integrators contribute their understanding of business systems, data and processes in different industries.
Under this model, customers receive a single integrated service rather than a collection of disconnected technology components. This simplifies deployment, clarifies accountability and makes it easier to move from a pilot project into production.
Competing in the European market requires standardized services, strong security, reliability and competitive economics. It is equally important to connect domestic supply with regional demand, as this allows us to operate at greater scale.
Our ambition is not to create a Croatian copy of American technology companies. It is to combine infrastructure, technology and business applications into a competitive service for Croatia and the wider European market.
AI is not just about models. It is about infrastructure, control and business outcomes.
We may have access to the best models in the world, but without high-quality data, connected systems, appropriate infrastructure and people who know how to apply the technology, we will not realize its full potential.
That is why Datum is building an approach that brings these layers together. The real value of AI does not come when a model answers a question, but when it reliably gets the job done and produces a result that justifies the investment.
