In 2025, Eurostat’s annual survey on the use of ICT technologies in businesses - conducted among 157,000 companies with at least ten employees in EU Member States - found that 20% had used at least one AI technology, compared with 8% in 2023. Growth is rapid, but there are numbers further down the line that bring attention to the division: 55% of large firms (250 or more employees) have adopted at least one AI technology, compared with 17% of small firms (10-49 employees). The gap in adoption rates was already wide in 2023 - 24 percentage points - but in two years it has risen to 38. Large enterprises not only start from a higher adoption rate, they are also increasing it more rapidly. Between 2024 and 2025, the proportion of large enterprises using AI grew by 13.9%; that of small enterprises by 5.8%.
The most obvious reason points to resources: large companies not only have more capital but also more infrastructure and more specialised technical staff. However, Eurostat data suggest a more nuanced interpretation. Among companies that considered adopting AI but did not proceed, 70.3% cite a lack of relevant skills as the main obstacle - the highest figure among all the reasons identified. This is followed, some way behind, by uncertainty over legal consequences (53.6%), privacy concerns (52.7%) and costs deemed too high (38.4%). Interestingly, however, the distribution of this obstacle varies by company size: it is cited by 70.9% of small enterprises, 69.2% of medium-sized enterprises and 65.1% of large enterprises. The variation is minimal, and the issue of skills, therefore, does not appear to be a characteristic of SMEs, but rather is distributed fairly evenly across the entire productive structure.
An analysis by the McKinsey Global Institute of the US market provides a useful interpretative framework, although it may not be directly comparable with European data. The report notes that 90% of organisations invest in AI, but fewer than 40% report a tangible impact on results, and explains this gap not by technological factors, but by organisational ones: value is unlocked by redesigning workflows so that people and systems each perform the tasks for which they are best suited, rather than superimposing tools onto existing processes. Doing so requires what the report terms ‘AI fluency’. If this mechanism were also applied to European companies, the Eurostat data on the lack of relevant skills would take on a more precise meaning: it would indicate not only an obstacle to initial adoption, but a limitation on the ability to transform adoption into a tangible productive advantage.
The gap in AI adoption between large and small businesses is well-documented and widening. Eurostat data suggests that this gap is not primarily driven by access to tools: the bottleneck (which cuts across company size, sector and country) is the ability to integrate AI into processes in a structural manner. If the mechanism identified by MGI also applies to the European context, this capacity for organisational integration would be one of the variables determining whether adoption translates into real productivity or remains superficial. It is therefore plausible that the observed gap in adoption rates reflects, at least in part, a deeper gap in the organisational capacity to exploit it. And a gap of this kind, within a production system that shares the same competitive conditions, tends to widen over time rather than narrow.
Sources: Eurostat, 'The Use of Artificial Intelligence Technologies in the European Union: Key Results', 2026 edition (ISOC_EB_AI). McKinsey Global Institute, 'Agents, robots, and us: Skill partnerships in the age of AI', November 2025. ICT ENT survey 2025: approximately 157,000 EU enterprises with 10 or more employees.