AI uptake gains momentum in the EU, but regional gaps and skill shortages persist

The uneven adoption of artificial intelligence (AI) across the European Union remains a critical hurdle to achieving digital competitiveness and innovation parity, according to a new data-driven study that offers one of the most comprehensive cross-country assessments of enterprise-level AI adoption in the EU to date.

Published in the journal Economies and titled “Artificial Intelligence Adoption in the European Union: A Data-Driven Cluster Analysis (2021–2024)”, the research uses Eurostat data and applies principal component analysis (PCA) and k-means clustering techniques to dissect usage patterns, identify adoption barriers, and categorize EU member states into distinct AI maturity clusters. The findings cover firms with at least ten employees across both industry and services, capturing a broad spectrum of enterprise behavior and national readiness for AI transformation.

How has AI adoption evolved across EU economies?

As per the analysis, AI adoption has seen a significant increase from 2021 to 2024, nearly doubling during the period. Despite this upward trend, the absolute levels remain modest, with the majority of countries still in single-digit adoption territory. Northern and Western European nations, including Denmark, Belgium, and Sweden, lead the bloc with adoption rates ranging between 15% and 28%. In contrast, countries like Romania, Bulgaria, and Poland lag significantly, with less than 7% of firms reporting any AI implementation.

A critical insight from the study is the shifting composition of AI technologies being deployed. While early phases saw growth in process automation, machine learning, and text mining, the period between 2023 and 2024 marked a pivot toward natural language generation, text mining, and speech recognition. The growing appeal of generative AI applications, especially following the rise of tools like ChatGPT, is altering enterprise priorities and accelerating experimentation with language-based systems.

Yet this adoption is heavily skewed by market size and policy environment. Larger, more digitally integrated economies have scaled faster and diversified more broadly across AI technologies. Smaller and less digitally mature economies have often pursued niche deployments, excelling in one or two categories but lacking full-spectrum integration. This divergence has led to a persistent three-tier cluster of member states: high-adoption generalists, niche-focused adopters, and low-adoption laggards.

What barriers are slowing down enterprise AI integration?

Despite the surge in interest and use cases, several key bottlenecks continue to impede AI uptake across EU enterprises. The most commonly cited challenge is a shortage of specialized skills. From 2023 to 2024, the proportion of firms identifying this constraint rose from 4.5% to 7.1%, overtaking other concerns such as data availability, legal ambiguity, and trust issues. This indicates a growing mismatch between technological ambition and human capital readiness, particularly in countries with underdeveloped digital education infrastructure.

Cost is another critical obstacle. Although the cost factor was not the most dominant barrier in earlier years, it has steadily gained prominence as firms face rising expenses for integration, training, and software customization. This is especially relevant for small and medium-sized enterprises (SMEs), which often lack both the internal budget and external funding to pursue sophisticated AI projects.

Interestingly, the study finds that only a minority of firms develop AI systems in-house. As of 2024, approximately 84% of AI adopters rely on external solutions, either purchased or open-source, which raises questions about long-term strategic independence and data security. While such approaches can lower upfront costs and accelerate deployment, they also reinforce dependency on global tech vendors and limit domestic innovation ecosystems.

Bias-checking and ethical oversight remain inconsistently applied across the board. Firms relying on external solutions are less likely to conduct rigorous bias assessments compared to those developing AI internally. This introduces potential risks around fairness, transparency, and regulatory compliance, particularly as the EU moves forward with the AI Act and related legislative initiatives.

How should EU policymakers respond to bridge the digital divide?

The study calls for a differentiated, policy-driven response to AI adoption across the European Union. A uniform set of strategies will likely fall short, given the entrenched digital asymmetries between member states. For countries in the high-adoption cluster, such as Finland, the Netherlands, and Denmark, the priority is to strengthen existing capabilities through targeted funding, robust ethical oversight, and regulatory clarity.

On the other hand, countries in the low-adoption cluster, most notably Romania, Poland, and Bulgaria, require foundational investment in digital infrastructure, workforce development, and AI literacy. Financial instruments, including public-private partnerships and EU cohesion funding, can play a central role in enabling SMEs in these regions to participate in the AI economy.

Training programs must be intensified to address the skills shortage, with an emphasis on both technical roles (e.g., data science, model validation) and implementation support functions (e.g., digital project management). Governments and academic institutions have a central role to play in creating scalable and accessible curricula aligned with emerging enterprise needs.

Regulatory environments must also adapt. Ionascu calls for the introduction of flexible policy instruments, such as regulatory sandboxes, that allow firms to test and refine AI systems under supervised conditions. Such measures can help mitigate risk while encouraging innovation, especially in sectors like health, energy, and finance where compliance burdens are high.

The study also points to the role of harmonized data governance frameworks . Enterprises need secure, interoperable, and high-quality data flows to scale AI responsibly. The proposed European Data Act and Data Governance Act are steps in the right direction, but must be supported with localized technical assistance to ensure effective adoption.