According to the AI Chamber 2025 survey, a majority of SMEs surveyed in Central Europe claim to use artificial intelligence. 7% of enterprises in the eurozone are classified as intensive users; no measured rate of effect on productivity is established. The gap between these two figures sums up a problem the region has not yet named clearly: adopting a tool and extracting value from it are two operations that demand radically different skills. Countries that understood this distinction—South Korea, Germany—built precise policies to bridge it. The survey suggests still-limited integration into strategic functions, without demonstrating the absence of transformative effects.
The essentials
- According to AI Chamber 2025, more than three-quarters of enterprises in the sample claim to use AI; the OECD describes uneven strategic integration, without quantifying a productive share.
- National Eurostat 2025 data shows significant gaps between countries: Romania is at 93% according to AI Chamber, Czechia is at 17.6% in Eurostat 2025 or 90% in the AI Chamber survey; the AI Chamber 2025 report provides no rate for Denmark or Finland. The observed gaps do not, by themselves, allow attribution of causality to differentiated investments in digital training.
- According to the OECD survey on the Western Balkans, a significant share of SMEs declares resorting to external expertise; the effect on the location of value created is not measured.
- Some countries, such as South Korea and Germany, show that a deliberate policy of shared platforms and practical training can help reduce this dependency, but only on condition of starting before the gap widens further.
- The intra-European digital divide could threaten the cohesion of the single market in a manner comparable to trade fractures with the outside.
Declaring is not adopting
This figure of declared adoption needs nuancing. The OECD finds that, in its international sample, most respondents stick to off-the-shelf tools—a text generator for emails, a ready-made customer service chatbot, sometimes an automatic translation tool. The OECD describes uneven strategic integration, without quantifying these uses to a European SME rate.
This distinction is not anecdotal. Tyler Cowen, an American economist whose recent work on AI raises precisely this problem, argues that AI amplifies human intellect on condition that users maintain real cognitive effort, that they know what to ask, how to evaluate the answer, how to integrate the tool’s output into a broader process. Without this effort, AI produces an illusion of competence: it generates content, but no one in the organization can judge its quality or transform it into durable competitive advantage.
Many Central European SMEs know this situation. They often consume tools designed elsewhere, by teams that master the deep layers of these systems, without internal capacities always being sufficient to adapt or challenge them. According to the AI Chamber survey, 44% declare mobilizing their internal resources and 25% resort to consulting. The region imports technology as it once imported machinery, with a risk that mastery remains concentrated outside.
The skills deficit as structural constraint
Eurostat 2025 data shows highly variable national levels depending on countries covered by the survey. These gaps reflect historically differentiated investments in vocational training and digital education.
This gap is not an inherited fate of geography. It corresponds to variations in public policies for digital training and education. Nordic countries treated the rise in digital skills as a strategic state investment from the 1990s onward. In Central Europe, vocational training approaches varied by country and period.
When external support ends, acquired competence does not always inscribe itself into durable organizational practices. Daron Acemoglu and Simon Johnson analyze this mechanism in their work on the distribution of technological gains: the benefits of automation concentrate where complementary skills exist, not where tools are simply deployed.
This ties moreover to the broader question of who controls the levers of advanced technology: when critical capacities remain concentrated outside a region, dependency perpetuates itself.
The distinct choices of South Korea and Germany
Two models merit attention, for different reasons. South Korea supports SME use of cloud services through vouchers and separately offers practical AI training. Korean SMEs had access to shared resources to test integrations. They had access to sufficiently powerful technical environments to test real integrations, with trainers capable of explaining the link between the tool and the business process.
Germany followed a partially different logic, anchored in its Mittelstand network, those mid-sized enterprises that constitute the industrial backbone of the country. The Mittelstandszentren played the role of intermediaries between university research, large technology companies, and SMEs. The adoption of AI in German SMEs did not happen through external consultants, but through sectoral networks where enterprises that had already integrated the tools passed experience to those getting started. Learning remained in the ecosystem.
These two models share shared infrastructure that reduces the cost of entry and a mechanism for skills transmission. This approach exists in Central Europe through digital hubs and clusters, even if their coverage and capacities vary by country. Policies in place combine national subsidy programs for tool purchases with European initiatives, some of which SMEs judge have overly complex access procedures. The European Union has moreover had the painful experience of this disconnect between the announcement of funding and its real absorption in contexts of low administrative capacity. Regulations designed upstream of adoption, as European experience on other technology segments shows, can add extra friction for the most fragile actors.
The risk of permanent dependency
Tyler Cowen, in a 2026 text on long-term risks to American competitiveness facing AI, formulates a warning transposable to Central Europe: a society that delegates cognitive effort to external systems without building competencies to supervise them risks creating structural dependency difficult to resolve. His argument targets the United States, but the diagnosis applies equally to Central Europe: when an economy consumes AI without understanding how it works, it risks seeing a share of its innovation capacity progressively concentrate among those who build the systems.
For Central Europe, this risk takes concrete form. SMEs that become accustomed to using off-the-shelf tools, without developing competencies to adapt or challenge them, expose themselves to the risk of becoming captive customers. In case of change in supplier conditions, pricing or service model, they often lack the means to migrate to an alternative or develop an in-house solution. Technological dependency reproduces, in the digital realm, the vulnerabilities the region has known in energy or certain industrial supply chains.
A competing reading exists, and it deserves to be posed honestly. Economists like Dani Rodrik argue that participation in global value chains, even in a position of initial dependency, can create progressive learning that eventually reduces the gap. Countries that integrated technologies developed outside sometimes, after a catch-up period, developed their own capacities. Taiwan with semiconductors, Korea with consumer electronics: both examples show trajectories where initial dependency did not prevent the emergence of local mastery.
But this trajectory supposes a condition that Rodrik himself states clearly: there must be deliberate policies for skills development, not merely access to tools. Without this explicit investment, dependency risks perpetuating itself without producing durable organizational learning.
New pressures by 2030
Pressure will intensify before policies have time to produce their effects. By 2030, AI models currently accessible to SMEs through consumer interfaces will become significantly more powerful, and more complex to use productively. Enterprises that will have developed real integration competencies during this transition window will be positioned to extract exponentially larger gains than those that will have remained at the stage of superficial consumption.
For Central European SMEs, this sketches two plausible trajectories by 2030-2035. In the first, public policies align with what South Korea and Germany have built: mutualized cloud platforms whose cost is partly borne by public or European funds, practical training programs anchored in existing sectoral networks, and mechanisms for validating acquired competencies that allow SMEs to signal their level of real integration to customers and partners. In this trajectory, SMEs could progress significantly in their productive integration, provided training policies begin now.
In the second trajectory, programs remain fragmented, European funding continues to be absorbed by large enterprises that have teams to build dossiers, and SMEs remain stuck in tool consumption without organizational learning. The gap with Northern Europe widens, and the region de facto specializes in value chain segments where technology counts less, which, in an increasingly automated economy, means segments with lower margins and less skilled jobs.
What would allow distinguishing between these two trajectories in coming years: the evolution of the digital skills rate in the 25-40 age bracket, the generation managing SMEs today and that will lead them in 2035. If this rate progresses by more than five points per year in countries of the region, it is a sign that training policies are beginning to take hold. If the rate stagnates, the second trajectory becomes more likely.
It should be noted, on this point, that some available data rest on declarations. Some Eurostat statistics rest on declarations, but Eurostat also measures the use of skills at work through the EU-LFS module on professional skills. A longitudinal monitoring program, on the model of what the OECD does with the PIAAC survey on adult skills, would give a much more precise picture of what SMEs can actually do with the tools they say they use.
Precise levers exist
The picture is not entirely dark. Several Central European countries have begun experimenting with more targeted approaches. In 2024, regional Polish programs funded SME training inscribed in the Baza Usług Rozwojowych, which records service providers and their certifications, a mechanism that shifts the subsidy from tool to competency. Czechia has developed partnerships between its technical universities and industrial clusters so that SMEs can test AI integrations in real environments, with academic support, without having to bear the cost of experimentation alone.
These initiatives remain small-scale and their impact on aggregate figures is not yet measurable. They show that the region identifies the necessary levers and that public and private actors are willing to invest in them. The sharing of technological gains between regions and between economic actors will become a more acute issue as AI establishes itself as a source of differential productivity: SMEs that will have built real mastery will accumulate advantages that those remaining on the surface cannot offset by simply declaring they use AI.
The cohesion of the European single market depends partly on this dynamic. A Europe where SMEs of the North and North-West capture the gains of AI while those of the Center and East remain in a position of dependency reproduces the asymmetries that European integration aimed to reduce. Targeted policies, well-designed funding, and networks for skills transmission can modify this trajectory. This supposes treating the digital skills deficit as a priority of industrial policy, not as a side effect to correct at the margins.
Decision-makers in the region will need to build the learning infrastructure before the gap becomes too large to be bridged at reasonable cost.
Sources
- AI Chamber, How do SMEs in CEE find their way in the world of AI?, 2026, https://aichamber.eu/report/how-do-smes-in-cee-find-their-way-in-the-world-of-ai/
- Tyler Cowen, A Doomsday Scenario for American AI, The Free Press, 2026, https://www.thefp.com/p/tyler-cowen-ai-doomsday-scenario-american
- Eurofound / Cedefop, SME Digitalisation in the EU, 2025
- OECD, SME AI Adoption Report, 2026
- Eurostat, ICT skills of individuals, data 2024-2025
