EU and US AI Adoption
How many manufacturers actually use AI?
Official data from Eurostat and the US Census Bureau: who adopts AI, what they use it for, how the EU and US compare, and what drives adoption in EU manufacturing.
This research summary was created using AI, but it is based on actual articles and our own research. Read our AI Usage Policy.
AI in EU manufacturing
In 2025, — of EU manufacturers used at least one AI technology, compared with — of all enterprises. Adoption has risen fast since 2023, but manufacturing still trails, and most of the new use is generative AI in the office rather than on the shop floor.
Manufacturing trails, and the gap is widening
Both lines were flat between 2021 and 2023, then took off with generative AI. Chemicals runs well ahead of manufacturing as a whole.
Enterprises using at least one AI technology, EU27
% of enterprises with 10+ employees. No AI survey in 2022. From 2025 the survey also asks about AI that generates images, video or audio, so part of the 2025 jump may come from the wider question.
Process-heavy and high-tech sub-sectors lead
Pharma, electronics, refining and chemicals lead. Food, textiles and metals trail, though every sub-sector has roughly doubled since 2023.
AI use by manufacturing sub-sector, EU27
% of enterprises, 2025 bars, with the 2023 value as a tick.
Country spread is wide, and Belgium is near the top
Benelux, Nordic and Alpine manufacturers lead. Much of Central and Eastern Europe trails far behind.
Manufacturers using AI by country, 2025
% of manufacturing enterprises, with the all-enterprise value as a tick. Finland has no 2025 manufacturing figure.
Generative AI drives the growth; production AI is outgrown, not abandoned
Text mining and content generation now lead in manufacturing. The share of AI users applying it to production fell from — to —, but the share of all manufacturers doing so still rose, from — to —. Office uses simply grew faster.
AI technologies used, 2025
% of all enterprises using each technology.
What manufacturers use AI for, 2023 to 2025
% of EU manufacturers that use AI. The tick shows all enterprises in 2025.
Two waves: classical AI creeps up, language AI surges
The technologies that defined industrial AI before LLMs (machine learning, image recognition, workflow automation, robots) still reach only a few percent of manufacturers. Language and generative technologies overtook all of them within two years. Part of the recent rise in machine learning may itself be LLM-based analysis tools that firms report as machine learning.
EU manufacturers using each AI technology
% of manufacturing enterprises with 10+ employees, EU27. A firm can use several technologies. Image, video and audio generation was first asked in 2025.
Skills, legal uncertainty and data hold manufacturers back
Among firms that considered AI but don't use it, lack of expertise is the top reason by a wide margin. Cost comes well down the list.
Reasons for not using AI, 2025
% of enterprises that considered AI but don't use it.
- Source: Eurostat, Artificial intelligence by NACE Rev.2 activity (isoc_eb_ain2), from the EU survey on ICT usage in enterprises. Downloaded 1 November 2026.
- Covers enterprises with 10 or more persons employed. Manufacturing is NACE section C. "All enterprises" means all activities except agriculture, mining and the financial sector.
- Values are % of enterprises, not % of employment, so a large firm and a small firm count the same.
- Eurostat marks a break in series for France and Sweden in 2023. Some country and sub-sector cells are confidential or unreliable and are left out.
AI adoption in the US
In the two weeks to 4 October 2026, — of US firms used AI, up from — in mid-November 2025. Manufacturing started well below the average and has nearly closed the gap.
Steady growth, with manufacturing climbing fastest
Use has risen steadily since the new question started in November 2025. Expected use runs a few points above current use, so firms expect the climb to continue.
US firms using AI, every two weeks
% of firms that used AI in any business function in the previous two weeks (solid), or expect to within six months (dashed). Dated by the end of each reference period. Each point carries a margin of about ±1–2 points.
Knowledge sectors lead; manufacturing is mid-table but moving up
Information, professional services and finance are far ahead. Manufacturing sits near the national average after one of the largest gains of any sector.
US firms using AI by sector
Bars: average of the six two-week periods from July to October 2026. Ticks: November 2025 to January 2026. Management of companies is left out because many of its values are suppressed.
US firms using AI by size
From Nov 2025–Jan 2026 (grey dot) to Jul–Oct 2026 (colored dot). Ticks show the share expecting to use AI within six months.
- Source: U.S. Census Bureau, Business Trends and Outlook Survey (BTOS), national, sector and employment-size-class files, Response Estimates sheet ("Yes" share). Downloaded 1 November 2026. Covers employer firms of all sizes, firm-weighted.
- Only the new question wording is shown, from November 2025 ("any business function"). The older series asked about AI in producing goods or services and isn't comparable. No data were collected during the October–November 2025 federal shutdown.
- The latest questionnaire refers to "Super Intelligence or SI (formerly referred to as Artificial Intelligence or AI)". The files don't say from which period that wording applies.
- Sectors are NAICS two-digit groups: manufacturing is 31–33, retail 44–45, transportation 48–49.
EU and US AI adoption, side by side
The two surveys ask different questions, so treat the gaps as approximate. At the closest dates, EU readings are a little higher in most sectors and far higher for ICT and large firms. Since late 2025, US manufacturing has caught up fastest.
Both are on a similar climb, at a similar level
The EU survey runs once a year and the US survey every two weeks. The US series starts in November 2025, when Census reworded its question; earlier US figures aren't comparable and are left out.
Share of firms using AI over time
EU: % of enterprises with 10+ employees, plotted at the start of the second quarter of each survey year. US: % of all firms using AI in the previous two weeks, by end of reference period.
Sector by sector, the EU reads higher, with ICT the big exception
The US figures from late 2025 are the closest in time to the EU 2025 survey, and they are lower in every sector shown. By mid-2026, US manufacturing, construction and professional services have passed EU 2025 levels, while ICT is still far below the EU.
Firms using AI by sector
Sectors matched from NAICS (US) to NACE (EU). US values average six two-week periods to smooth sampling noise. The US total includes finance, health and education; the EU total excludes finance.
The size gap is steeper in the EU
US adoption barely changes with size up to 50 employees, and even US micro firms are near the national average. In the EU, adoption climbs steeply with size, and firms with 250+ employees are far ahead of those with 10–49.
Firms using AI by size
US bars: Jul–Oct 2026, all sectors. EU ticks: 2025, the EU size band that contains each US band. The EU doesn't survey firms under 10 employees.
Speed of adoption
Change in the share of firms using AI, converted to percentage points per year. EU: 2024 to 2025 survey. US: Nov 2025–Jan 2026 to Jul–Oct 2026.
Why the numbers aren't a 1:1 comparison
Most differences push the EU figure up relative to the US, so a small EU lead in the data is no proof of a real lead.
| EU: Eurostat ICT usage in enterprises | US: Census Business Trends and Outlook Survey | Likely effect | |
|---|---|---|---|
| Question | Asks about eight named AI technologies (text mining, speech recognition, image recognition and so on); a firm counts if it uses any of them. | One question: did the business use AI in any business function, with a short list of examples. The latest questionnaire calls it "Super Intelligence (formerly AI)". | A list of named technologies tends to draw more "yes" answers, which raises the EU figure. |
| Time window | Use at the time of the annual survey, generally the first half of the year. | Use in the last two weeks. | A two-week window misses occasional users, which lowers the US figure. |
| Firms covered | 10+ employees. Excludes agriculture, mining and finance. | All employer firms, including 1–9 employees. All sectors, including finance, health and education. | Small: US micro firms adopt at about the national average. |
| Frequency | Annual (2021, 2023, 2024, 2025). 2026 figures not yet published. | Every two weeks since September 2023. New wording from November 2025; no data during the October–November 2025 shutdown. | The two series only overlap in time when EU 2026 figures come out. |
| Weighting | % of enterprises. | % of firms. Employment-weighted figures also exist and are much higher. | Comparable when both are firm-weighted, as here. |
- Source: U.S. Census Bureau, Business Trends and Outlook Survey (BTOS) national, sector and employment-size-class files, Response Estimates sheet, question on current AI use ("Yes" share). Downloaded 1 November 2026. Two-week standard errors are about 1 percentage point for manufacturing.
- Background and wording change: Bonney et al., "The Microstructure of AI Diffusion", CES Working Paper 26-25R, October 2026.
- US sector codes: manufacturing NAICS 31–33, information 51, professional services 54, wholesale 42, retail 44–45, construction 23, transportation 48–49, accommodation and food 72.
What drives AI adoption in EU manufacturing
Where a manufacturer is based explains more than what it makes, and most of the country effect follows national cost and income levels. Within countries, digital maturity and skills explain the gaps between sub-sectors far better than size on its own; capital intensity, profit margins and workforce age explain almost nothing. Belgium adopts more than any of these factors predict.
Country comes first, then skills, then sub-sector
Each bar is that factor's share of the variation in AI use across country × sub-sector combinations, split fairly between overlapping factors (a Shapley decomposition). Skills combines average wages with the share of firms employing ICT specialists. Digital intensity is Eurostat's 2022 index, measured before the survey asked about AI, so it doesn't count AI use itself.
Share of variation in AI use explained, 2025
AI use 2025 on a log-odds scale. Business structure 2023; digital intensity and ICT specialists 2022. Only the 147 country × sub-sector cells with every measure available.
The country effect follows cost and income levels
Countries where manufacturers pay higher wages and prices are higher adopt more AI. Once the price level is known, wages in purchasing-power terms add little, so the pattern fits labour costs and national income rather than skills alone. The age of the workforce makes no difference. Belgium sits well above the line: its lead isn't explained by being a high-wage economy.
Country effect on AI use against national price level
Vertical: each country's AI use in manufacturing, holding sub-sector fixed (log-odds, 0 = average country). Horizontal: price level of GDP, EU = 100, 2023. Line: best fit.
How much of the country effect each factor explains
R² of the 27 country effects on each factor alone, and on all of them combined.
Sub-sector gaps shrink once size and skills are accounted for, but don't vanish
Pharma and transport equipment lose the most of their lead once structure is controlled for, because they have many large, high-wage firms. Electronics and chemicals keep most of theirs, which suggests something specific to those industries, such as data-rich processes or a digital culture.
Odds of using AI relative to food manufacturers
Grey dot: country controls only. Colored dot: adding firm size, capex intensity, margin, productivity and wages. Refining has very few firms per country and is left out of this chart.
Actual AI use against what firm size alone predicts, EU27 2025
Tick: expected rate if each sub-sector's mix of 10–49, 50–249 and 250+ firms adopted at the EU-wide rate for that size. Bars past the tick adopt more than their size predicts.
Classical AI needs scale and digital maturity; language AI much less so
Robots, image recognition, machine learning and workflow automation rise steeply with the share of large firms. For machine learning and workflow automation that size effect disappears once digital intensity is accounted for: large firms adopt because they are more digitally mature. Robots still need scale on top. Text mining and language generation depend far less on size.
| Technology | Wave | Doubling share of large firms | Sector pays 10% more | +10 points highly digital firms | Size, once digital intensity is known |
|---|
Change in the odds of using each technology, 2025, with country fixed effects; above 1 means more use. Faded values are not significant at 5%. Sector wage is relative to the country's average. Highly digital firms: high or very high on the 2022 digital intensity index.
Belgium leads within sectors, not because of them
Belgium has more than twice the EU share of chemical and pharma firms (— of manufacturers with 10+ employees, against —) and somewhat larger firms. But because Eurostat counts firms rather than output, those groups are too small to move the total much.
From the EU rate to Belgium's rate, manufacturing 2025
Sub-sector mix: Belgian sub-sector weights at EU sub-sector rates. Size mix: Belgian size weights at EU size-class rates. The rest is Belgian firms adopting more than EU firms of the same sub-sector and size.
How robust each driver is
The analysis was re-run 64 ways: AI year 2024 or 2025, business data 2022 or 2023, log-odds or percentages, with or without refining, with or without the three smallest countries, and with or without weighting by number of firms. Firm size is a positive driver in every variant across sub-sectors, and in about half when comparing only countries within the same sub-sector, because part of the size effect is the sub-sector itself. Capex intensity is never a positive driver.
| Driver | Positive, significant (without sub-sector) | Negative, significant (without sub-sector) | Positive, significant (within sub-sector) | Negative, significant (within sub-sector) |
|---|
Each number counts the variants, out of 64, where the driver's coefficient is significant at 5%, with country-clustered standard errors. "Within sub-sector" adds sub-sector fixed effects, so the driver is identified only from differences between countries in the same sub-sector. Productivity and wages are highly correlated, so their separate signs are unreliable.
- AI use: Eurostat isoc_eb_ain2 (enterprises with 10+ employees using at least one AI technology). Size-class AI rates: Eurostat isoc_eb_ai, all enterprises.
- Business structure: Eurostat structural business statistics, sbs_sc_ovw (enterprises by size class) and sbs_ovw_act (value added, gross operating surplus, turnover, gross investment in tangible assets, employment, wages), 2023. Divisions are summed to the AI survey's sub-sector groups; a group is left out when any of its divisions is confidential.
- Drivers: share of 10+ firms with 250+ employees (log), investment in tangible assets as % of value added, gross operating surplus as % of turnover, value added per person employed (log), wages per employee (log), all standardised.
- Extra measures: digital intensity index (isoc_e_diin2, 2022 version without AI), enterprises employing ICT specialists (isoc_ske_itspen2, 2022), price level index of GDP (prc_ppp_ind, 2023), employment by age in manufacturing (lfsa_egan2, 2024). Business R&D intensity (rd_e_berdindr2, 2021–2023, as % of value added) predicts AI use strongly on its own, but adds nothing once digital intensity is included, because the two move closely together.
- These are associations between country-sector averages, not causal effects on individual firms. Size, skills and sub-sector overlap heavily, and the analysis can't tell, for example, whether skilled workforces drive AI use or AI-using industries pay more.