AI reaches 20% of EU firms, but the figure does not measure intensive use

European business statistics place the proportion of companies that reported using at least one artificial intelligence technology in 2025 near 20%. This figure refers to EU companies with ten or more employees and comes from the Eurostat EU-ICT-Firm survey, analyzed by Funcas.

The definition is broad: a company only needs to indicate that it uses at least one technology included in the questionnaire to be counted as an adopter. Therefore, the percentage records the declared presence of AI, not the frequency, scope, or economic weight of use.

This difference has consequences for executives and analysts. A company that employs a tool in a small team and another that integrates it into many operations are counted the same way in the aggregate rate. The data serves to track diffusion, but it does not allow deducing by itself how much value AI generates or how many processes have been transformed.

The Funcas series places the proportion of adopting companies in the European Union at around 7.6% in 2021 and near 20% in 2025. The increase shows considerable diffusion in that period, although it does not equate to a direct measurement of productivity or technological maturity.

The European average hides substantial differences. In 2025, Denmark, Finland, and Sweden exceeded 35%, while Romania, Poland, Bulgaria, and Greece were below 10%. Spain recorded 20.3%, practically in line with the EU average; Germany reached 26%, France 18.2%, Italy 16.4%, and Portugal 11.5%.

The Funcas analysis relates adoption differences to structural factors, including the level of economic development, research capacity, and process digitalization. The comparison is not a ranking of business quality: sectoral composition, company size, and country-specific conditions can also influence the results.

To interpret the data, one must look at who is included in the count and what is considered use. The cited rate corresponds to companies with ten or more employees that declare using at least one AI technology. It does not represent all companies, including micro-enterprises that fall outside that threshold.

The survey also does not reveal the different degrees of deployment. An affirmative answer does not indicate whether the tool is used occasionally or continuously, how many people use it, what function it serves, or whether it is connected to critical processes. Funcas notes that the data do not measure intensity through variables such as the number of workers or hours dedicated.

Therefore, percentages from different surveys or reports are not always comparable. The observed population, included sectors, technologies asked about, and the concept measured can vary. Before comparing figures, it is advisable to check population, question, year, and methodology: a measure of AI readiness is not equivalent to a declared use rate.

The adoption rate is not enough to conclude that a company has moved from experimenting with AI to incorporating it into core operations. To assess that transition, data are needed on affected processes, decisions supported by AI systems, the number of employees using them, and the results obtained compared to a comparable reference.

The Accenture AI Progress Barometer offers a different perspective. The study tracks the world's 3,000 largest companies and evaluates capabilities related to readiness to leverage AI, with a European score of 43.1 versus 48.9 in North America. It is a readiness indicator for large companies, not an adoption rate for the entire European business sector.

The two measures answer different questions: Eurostat estimates how many companies in a defined universe declare some use; Accenture evaluates the capabilities of a group of large companies. To decide where to invest, a company should distinguish between initial use, organizational readiness, and effective integration, rather than summarizing them with a single percentage.

Adopting AI may require usable data, infrastructure, skills, and organizational changes. The Funcas study finds that adoption starts earlier and advances faster in countries with higher economic development; it also points to differences linked to scientific and technical capacity and the degree of digitalization of production processes.

Declared adoption and business readiness are indicators observed in the cited studies; a projection, in contrast, expresses an estimate about a possible future evolution. They should not be treated as interchangeable measures: a forecast suggesting more investment or deployment does not prove that this change has already occurred.

The gap can also exist within the same country or company. A company may allow the use of tools without modifying workflows or developing internal capabilities. In that case, the presence of AI does not by itself inform about its contribution: the outcome depends on complementary conditions that the binary rate does not capture.

The European Commission presents the AI Act as a risk-based framework that seeks to promote innovation and, at the same time, safe and reliable artificial intelligence. This approach places risk at the center of the European framework, alongside the objective of boosting adoption and competitiveness.

The Commission also indicates that the European AI Office participates in the application and compliance of the Act in collaboration with Member States. For companies, this makes it relevant to integrate responsibilities regarding AI systems into deployment decisions, in addition to considering data, infrastructure, and skills.

Regulation does not replace the business evaluation of each initiative. A general adoption rate informs about diffusion, but it does not determine how a specific case should be managed or what value it can bring. The aggregate figure orients; the decision requires examining the system and its use.

The European data near 20% is useful as a diffusion reference, with a delimited scope: companies with ten or more employees that declare at least one use of AI. It does not certify that a fifth of companies have transformed their operations, nor does it allow estimating the return on investment by itself.

When comparing countries, providers, or studies, it is advisable to order the question before the percentage: is use, readiness, or intensity being measured? Are small companies included? What year is observed? Does the source study companies or collect responses from executives? Comparing equivalent metrics helps separate observed use from expectations.

For a company, a useful evaluation can follow three planes: identify which tools are effectively in use; locate the processes or decisions that have changed; and review results, costs, risks, and training needs. Thus, a market statistic can orient a specific decision: prioritize cases whose value and management can be evaluated, rather than pursuing adoption as an isolated goal.