Use of artificial intelligence in enterprises (Eurostat, Statistics Explained)
TL;DR
Eurostat’s standing article on AI use by EU firms, from the EU survey on ICT usage and e-commerce in enterprises. This snapshot carries the 2025 data. The page was last edited on 4 November 2026; its data were extracted in December 2025, and the next update is planned for 15 December 2026.
- One in five EU firms uses AI. In 2025, 19.95% of EU enterprises with 10 or more employees and self-employed persons used at least one of eight AI technologies, up 6.47 points on 2024 (13.48%).
- Size drives use. 17% of small, 30.36% of medium and 55.03% of large enterprises used AI.
- Countries range eightfold. From 5.21% in Romania to 42.03% in Denmark. The Netherlands is at 33.21% (2024: 23.06%), and Eurostat marks the Dutch 2025 figure as a break in the time series.
- No technology dominates. Text mining leads at 11.75%, then generating pictures, video or sound (9.55%) and generating text, speech or code (8.76%).
- Firms use AI mostly for marketing and administration. Among AI users, 34.70% use it for marketing or sales and 31.05% for business administration or management.
- Expertise is the main barrier. Among firms that considered AI but did not use it, 70.89% cite lack of expertise, 52.52% unclear legal consequences and 48.83% data-protection concerns. Only 20.68% say AI is not useful to them.
Key data
| Measure (EU, enterprises with 10+ persons) | 2024 | 2025 |
|---|---|---|
| Use at least one AI technology: all | 13.48% | 19.95% |
| small (10–49) | 11.21% | 17.00% |
| medium (50–249) | 20.97% | 30.36% |
| large (250+) | 41.17% | 55.03% |
| Netherlands, all (break in series 2025) | 23.06% | 33.21% |
| Did not use AI but considered it (share of non-users) | 12.19% | 14.21% |
By technology, 2025 (all / large): text mining 11.75% / 35.04% · pictures, video or sound generation 9.55% / 27.92% · text, speech or code generation 8.76% / 31.68% · speech recognition 7.22% / 26.13% · workflow automation or decision support 5.35% / 24.42% · machine learning for data analysis 5.10% / 25.05% · image recognition 3.78% / 16.66% · autonomous movement of machines 1.39% / 8.55%.
By sector, 2025: information and communication 62.52% · professional, scientific and technical 40.43% · real estate 24.82% · manufacturing 17.27% · construction 10.79%.
Method and definitions
- Survey. National statistical authorities ran the survey in the first months of 2025. 157,000 of the EU’s 1.53 million enterprises were surveyed; 83% of the population is small, 14% medium and 3% large.
- Unit and coverage. The enterprise, with at least 10 employees and self-employed persons. Covered: manufacturing, energy and water, construction, trade, transport, accommodation and food, information and communication, real estate, professional and scientific activities, administrative and support services, and computer repair. Financial services, health care, education and public administration are outside the scope.
- What counts as use. At least one of eight listed technologies. CBS notes that the eighth, generating pictures, video or sound, was added in 2025. The page itself does not say so.
- Aggregation. Unreliable national data are included in EU aggregates.
Scope and reliability
- Official statistics with a common EU questionnaire and a large sample, unlike the vendor and consultancy surveys most adoption figures in the corpus come from.
- Breadth only. A firm counts if it uses any one technology. Nothing is measured about depth, scale within the firm, or value.
- Firm-weighted. Every firm counts once, and 83% of the population is small, so the EU total is close to the small-firm rate.
- No hospitals. Health care is outside the survey’s scope, so this page gives no figure for hospitals.
- Breaks and definition changes. The Netherlands’ 2025 figure is flagged as a break in series without explanation on the page. CBS’s StatLine notes a new weighting method from 2025. The added eighth technology affects every country.
- A living page with loose ends. The “Data sources” text says the data are “based on the results of the 2024 EU survey” while every figure is labelled 2025, and the highlight box links a file titled for 2023 and 2024. The figures in this snapshot match the page’s own source-data spreadsheet.
Dynamic-capabilities reading
digital-sensing/digital-scouting: a yearly official scan of which AI technologies EU firms use, by country, size and sector.contextual/internal-barriers: the reasons non-users give (lack of expertise, unclear legal consequences, data-protection concerns, data availability and quality at 44.37%, incompatible systems at 41.39%) are barriers inside the firm and its environment, measured across all EU countries.
How it connects
- enterprise-ai-adoption and the thread ai-maturity-measurement-comparison: the official EU baseline for firm adoption.
- Rendtorff-Smith (OECD): 20% of firms, 50%+ of big firms and about 17% of small ones in official statistics.
- AI Index 2026: 88% organisational adoption from an executive survey.
- Stanford DEL Adoption Monitor: 68.6% of firms in an executive survey of the US, UK, Germany and Australia, and a table of adoption figures by instrument.
- CBS, ICT-gebruik bij bedrijven 2025: the Dutch part of the same survey, with size and sector detail.
- RaboResearch (Rabobank): Dutch AI use by business function against the EU average, from the same survey.
- The page cites the EU Digital Decade target that three out of four EU companies use cloud computing, big data or AI by 2030.
- Federal Reserve FEDS Note: the US counterpart in official statistics: about 18% of firms in late 2025, after a change of question wording.
What was actually ingested
The full article text as served on 7 November 2026 (last edit 4 November 2026, revision 723073) and its source-data spreadsheet, which holds the data behind all charts. Eurostat’s database (dataset isoc_eb_ai) was queried once to check the Netherlands and EU series for 2021 to 2025 and the break flag. That check is appended to the raw file and is not part of the page.
Linked entities and concepts
- enterprise-ai-adoption: concept.
Dangling (single-source mention, deferred): Eurostat, European Commission.