Average of the panel (2025)
9.6
percent of enterprises
+3.5pp vs 2024
Highest
30.8
Denmark
Lowest
2.7
Turkey
Countries ranked
32
with a published figure

Data

Generative AI Use by Businesses in Europe by Country (2025) — source data. Eurostat.
# Country % of enterprises (10+ employees) Change vs 2024
1 Denmark 30.8 +12.3pp
2 Finland 20.1 +7.1pp
3 Sweden 19.7 +7.8pp
4 Netherlands 16.3 +3.8pp
5 Belgium 15.9 +3.2pp
6 Austria 15.3 +7pp
7 Norway 15.1 +4.8pp
8 Estonia 15.1 +9.3pp
9 Luxembourg 13.9 +5.1pp
10 Czechia 12.8 +6.8pp
11 Slovakia 9.3 +5.4pp
12 Ireland 9 +1.8pp
13 Germany 9 +2.4pp
14 Spain 8.9 +4.7pp
15 France 7.8 +4.6pp
16 Croatia 7.4 +3.5pp
17 Lithuania 6.7 +3.1pp
18 Malta 6.3 +2.6pp
19 Italy 6.2 +2.5pp
20 Slovenia 6.1 -5.6pp
21 Portugal 5.3 +2.2pp
22 Poland 5.2 +2.9pp
23 Montenegro 4.8 +0.8pp
24 Latvia 4.7 +2.5pp
25 Cyprus 4.5 +1.5pp
26 Hungary 4.2 +1.4pp
27 Greece 3.7 +0.5pp
28 Bulgaria 3.2 +0.9pp
29 Serbia 3.2 +1.7pp
30 Romania 3.1 +1.6pp
31 Albania 2.9 +0.4pp
32 Turkey 2.7 +0.8pp

About this Dataset

30.8% of enterprises in Denmark use AI to generate written or spoken language, against 2.7% in Turkey. The panel average is 9.6%, +3.5pp on 2024 — growth of roughly 58% in one survey year, which is not a rate any other indicator in Eurostat's ICT module comes close to.

This is the nearest thing Eurostat publishes to a measure of business adoption of large language models. It asks specifically about AI that generates text or speech, not about AI in general.

Denmark, Finland and Sweden take the top three places, and the Nordic and Benelux lead here maps almost exactly onto the broader AI-adoption ranking. That correspondence is the useful finding: generative AI is not redistributing digital capability across Europe, it is compounding what was already there. Denmark alone added +12.3pp over the year, more than the 7.1% the median country has reached in total, and the spread between top and bottom widened to 28 percentage points.

Firm size matters as much as geography. Adoption is far higher among large enterprises than among those with ten to forty-nine employees in every country surveyed, so a national score partly reports the shape of the business population rather than the sophistication of its leading companies — Italy at 6.2% and Poland at 5.2% both sit below what their large-firm sectors alone would produce. And Slovenia is a reminder of how young the series is: it fell -5.6pp over the year, which is unlikely to describe firms abandoning the technology and more likely reflects survey sampling. Four survey points are enough to establish an ordering and to size the annual moves, not to fit a trend.

Every row is the figure published for 2025; no value is carried over from an adjacent year to fill a gap. North Macedonia has not published a 2025 figure and is left out rather than carried forward from an older year. The chart tracks a fixed panel of 31 countries: those reporting in every year from 2021. The line therefore moves because the countries moved, not because the sample changed. The indicator is Eurostat ISOC_EB_AI, code E_AI_TNLG: enterprises with ten or more employees using AI technologies that generate written or spoken language. Eurostat first ran the module in 2021 and skipped 2022, so the chart covers four survey years rather than a continuous annual run.

Country pages: Denmark · Finland · Sweden · Netherlands · Belgium · Austria · Norway · Estonia · Luxembourg · Czechia · Slovakia · Ireland · Germany · Spain · France · Croatia · Lithuania · Malta · Italy · Slovenia · Portugal · Poland · Montenegro · Latvia · Cyprus · Hungary · Greece · Bulgaria · Serbia · Romania · Albania · Turkey

Frequently Asked Questions

Denmark, where 30.8% of enterprises with ten or more employees reported using AI to generate written or spoken language in 2025, ahead of Finland, Sweden and Netherlands. Denmark is well clear of the field: it gained +12.3pp over the year, more on its own than the 7.1% that the median European country has reached in total.

Eurostat asks enterprises whether they use AI technologies that generate written or spoken language — the survey code is E_AI_TNLG within the ICT usage module. In practice this is the closest published measure of business adoption of large language models and related text and speech generation tools. It is a subset of the broader AI-adoption question, which also covers machine vision, robotics, process automation and machine learning for data analysis, and a firm counts as a user if any part of the business uses the technology at all.

The panel average reached 9.6% in 2025, +3.5pp on 2024 — growth of about 58% in a single survey year, faster than any other indicator in Eurostat's ICT module. The history is short by design rather than by omission: the module ran in 2021, 2023, 2024 and 2025, with no 2022 round, so there are four survey points in total.