AI in manufacturing in Europe is growing fast from a low base. In 2025, 17.3 percent of EU manufacturers with ten or more staff used at least one AI technology, up from 6.8 percent in 2023.1 The average across all sectors was 20.0 percent.2 This briefing shows where that AI sits, which countries lead, how factories buy it and which EU rules apply to it until 2028.

Summary

  1. Growth came from the office, not the line. Language tools reached 7 to 9 percent of manufacturers. AI in production processes reached 3.5 percent.
  2. Adoption varies twelvefold by country. It ranges from 39.8 percent of manufacturers in Belgium to 3.2 percent in Romania. Germany stands at 24.4 percent, France at 16.9.
  3. Five EU laws set nine dates for factory AI. They run from February 2025 to August 2028. Five of the dates are still ahead, four of them by December 2027.

How widely is AI used in manufacturing in Europe?

Eurostat's enterprise survey is the only measure harmonised across all 27 member states. It counts a firm as an AI user if it runs at least one of eight technologies, from text mining to autonomous robots.1 Manufacturing adoption rose from 6.8 percent in 2023 to 10.6 percent in 2024 and 17.3 percent in 2025. It still trails the economy as a whole, and the gap is widening: 1.3 percentage points in 2023, 2.7 in 2025.

Exhibit 1 splits that growth by technology.

Language AI spread; machine AI did not

Text mining rose from 1.8 to 9.4 percent of manufacturers. Autonomous machines stayed near 1.5 percent.

Text mining and language generation grew fivefold in EU factories in two years; autonomous machines barely moved.

Manufacturing enterprises using each AI technology, EU-27, % of enterprises

2023 2025
0 2 4 6 8 10 1.8 9.4 1.2 7.1 1.4 5.3 1.7 3.7 2.9 3.9 2.2 3.1 1.5 1.6 Text mining Language generation¹ Speech recognition Machine learning² Process automation³ Image recognition Autonomous machines⁴

¹ Generation of written or spoken language or programming code. ² Machine learning, such as deep learning, for data analysis. ³ AI-based robotic process automation of workflows or decisions. ⁴ Autonomous robots, self-driving vehicles or drones. Share of all manufacturing enterprises (NACE C) with 10 or more persons employed, EU-27. Image, video or audio generation was first surveyed in 2025 (7.5%) and is not shown.

Source: Eurostat, Artificial intelligence by NACE Rev. 2 activity (isoc_eb_ain2), database updated 15 June 2026; accessed 29 September 2026

The technologies that grew are the ones inside chat assistants and document tools. The shop-floor technologies, image recognition, machine learning on process data and autonomous machines, each gained less than two percentage points.

Data on purpose tells the same story. The share of manufacturers using AI in production processes rose from 2.6 to 3.5 percent. Use in marketing or sales rose from 1.0 to 5.3 percent, and in administration or management from 1.1 to 4.7 percent.1 Among manufacturers that use AI, production fell from 38.2 percent of users in 2023 to 20.2 percent in 2025. The early adopters were factory specialists. The new ones mostly bought office tools. Quality inspection, predictive maintenance and process control are still rare in European plants.

Manufacturers that considered AI and held back cite missing expertise first, then legal uncertainty and data protection. The manufacturing and industry page charts those reasons and adoption by sub-sector.

Which European countries lead AI in manufacturing?

The gap between member states is wide. Belgium, Denmark and Luxembourg lead, each near 40 percent. Spain, France, Italy and Poland sit below the EU average. Exhibit 2 ranks selected countries for 2025.

Belgian factories use AI twelve times as often as Romanian ones

Germany is above the EU average. Spain, France, Italy and Poland are below it.

In 2025, 39.8% of Belgian manufacturers used AI, against 17.3% across the EU and 3.2% in Romania.

Manufacturing enterprises using at least one AI technology, 2025, % of enterprises

EU-27 average and below Above the EU-27 average
0 10 20 30 40 50 Belgium 39.8% Denmark 38.6% Luxembourg 37.6% Sweden 32.9% Austria 32.5% Netherlands 28.6% Germany 24.4% EU-27 17.3% Spain 17.1% France 16.9% Italy 14.7% Poland 7.7% Romania 3.2%

1 Share of manufacturing enterprises (NACE C) with 10 or more persons employed that used at least one of eight AI technologies, reference year 2025. Selected member states. Finland's 2025 figure is flagged as low reliability and not shown.

Source: Eurostat, Artificial intelligence by NACE Rev. 2 activity (isoc_eb_ain2), database updated 15 June 2026; accessed 29 September 2026

Belgium also leads on production use: 10.2 percent of its manufacturers apply AI to production processes, nearly three times the EU rate.1 The laggards are moving. France more than doubled its rate in one year, from 7.4 to 16.9 percent. Italy rose from 8.0 to 14.7 percent, Poland from 5.1 to 7.7 percent.

For a group with plants in several countries, the figures mean uneven starting points. Skills, data quality and supplier readiness differ by site. A rollout plan built on the head office's maturity will stall at the slower plants.

How do European manufacturers source their AI?

Eurostat last asked in 2024 how firms obtain their AI. Most manufacturers that use AI buy it ready-made. Multiple answers were possible:1

  • 58.4 percent use commercial software that is ready to use;
  • 34.8 percent have AI developed or modified by external providers;
  • 24.5 percent adapt commercial software with their own staff;
  • 20.3 percent adapt open-source software with their own staff, against 27.0 percent across all sectors;
  • 17.1 percent develop AI with their own staff.

Ready-made tools are quick to start. They usually process data on the vendor's servers. In a factory that data is drawings, recipes, process settings and customer specifications. EU trade secret law protects such know-how only if its holder takes reasonable steps to keep it secret.11 In 2025, about four in five manufacturers using AI also bought cloud computing services.1

The Commission sees the same tension. Under its Apply AI Strategy, announced on 8 October 2025, it plans to pool industrial data through trusted third parties, so models can be trained while intellectual property stays protected. It will also strengthen a Testing and Experimentation Facility that validates and certifies AI-driven machinery.3

Open-weight models offer a third route. They run on the manufacturer's own servers or in a European cloud, so design data stays inside. Lindstead's model selection tests such models on a manufacturer's own specifications and test reports. The self-hosted LLM guide explains what running one in-house involves.

Which EU rules apply to AI in manufacturing, and when?

No single law governs factory AI. Five EU laws set dates between 2025 and 2028: the AI Act, the Machinery Regulation, the Data Act, the Cyber Resilience Act and the new Product Liability Directive. The table lists each date that changes what a manufacturer may do with AI.

EU rules for AI in manufacturing, 2025 to 2028

Date Law What changes for AI
2 Feb 2025 AI Act, Art. 5 Emotion recognition at work banned
12 Sep 2025 Data Act Machine data use only under user contract
11 Sep 2026 Cyber Resilience Act Report actively exploited vulnerabilities
12 Sep 2026 Data Act, Art. 3 New connected products: data access by design
9 Dec 2026 Product Liability Directive Software and AI count as products
20 Jan 2027 Machinery Regulation Notified body for self-learning safety parts
2 Dec 2027 AI Act, Annex III AI that manages workers becomes high-risk
11 Dec 2027 Cyber Resilience Act Main security requirements apply
2 Aug 2028 AI Act, Annex I Sec. B Machinery delegated acts apply

Applies now Upcoming

Sources: Regulation (EU) 2024/1689 and 2026/1744; Regulation (EU) 2023/1230; Regulation (EU) 2023/2854, Article 50; European Commission, Cyber Resilience Act; Directive (EU) 2024/2853; accessed 29 September 2026. Status as of that date.

The AI Act reaches factories in three ways

First, bans already apply. Since 2 February 2025, AI that infers the emotions of people at work is prohibited, except for medical or safety reasons.4 Camera systems that read the mood of line workers fall under this ban.

Second, AI that manages workers becomes high-risk. Annex III covers AI that allocates tasks based on behaviour or personal traits, or that monitors and evaluates staff. After the Digital Omnibus, these duties apply from 2 December 2027.5

Third, AI in machines. The Omnibus moved the Machinery Regulation to Section B of Annex I. AI Act requirements reach machinery through Commission delegated acts that must apply by 2 August 2028.12 Earlier, from 20 January 2027, the Machinery Regulation requires a notified body to assess safety components with self-evolving, machine-learning behaviour.6,13 The AI Act timeline lists every date after the Omnibus.

Liability and cyber security now cover software

The new Product Liability Directive treats software, AI systems and digital manufacturing files as products. It applies to products placed on the market after 9 December 2026.9 A manufacturer that adds an AI function to a machine is liable for its defects without proof of fault.

Under the Cyber Resilience Act, makers of products with digital elements have had to report actively exploited vulnerabilities since 11 September 2026. The main security requirements apply from 11 December 2027.8 NIS2 already covers medium and large makers of machinery, electronics, electrical equipment, vehicles and chemicals.10

The rules stack up on one product. A machine builder that ships a machine-learning safety function in 2027 needs a notified body under the Machinery Regulation. It carries strict liability for the software under the new directive and reports exploited vulnerabilities under the Cyber Resilience Act. AI Act requirements follow through the machinery delegated acts. Lindstead advises one technical file that covers all four regimes.

Does the Data Act limit AI training on machine data?

Yes. Since 12 September 2025, a manufacturer that holds data from a customer's connected machine may use readily available non-personal data only under a contract with the user.7 It may not use that data to learn about the user's economic situation or production methods in ways that harm the user's commercial position.

Connected products placed on the market after 12 September 2026 must also let users access their data by design. A predictive maintenance model trained on fleet data therefore needs contract terms that allow it. Personal data in machine logs, such as operator IDs, also falls under the GDPR.

What should manufacturers do before 2027?

Lindstead advises five steps, in this order:

  1. List every AI tool in use, including public chatbots, and the data each one touches.
  2. Classify each against the AI Act: banned practice, Annex III worker management or machinery safety function.
  3. Check customer contracts on machine data and confidentiality before training any model on field data.
  4. Move work on drawings, recipes and customer specifications to models the company controls.
  5. For products shipped after 9 December 2026, add AI functions to the liability, cyber and machinery files.

Steps one to three need no new infrastructure. Steps four and five need budget and a decision on where models run. The sector page sets out the main risks for manufacturers. Lindstead maps AI Act, Data Act, NIS2 and machinery obligations to each AI system and deploys a first use case on servers the company controls. Whether owning the hardware pays off depends on volume; the break-even analysis runs the numbers.

Frequently asked questions

  • In 2025, 17.3 percent of EU manufacturers with ten or more staff used at least one AI technology, according to Eurostat. That is up from 6.8 percent in 2023 but below the 20.0 percent average across all sectors. Only 3.5 percent of manufacturers used AI in production processes.

  • Belgium, where 39.8 percent of manufacturers used AI in 2025. Denmark (38.6 percent) and Luxembourg (37.6 percent) follow. Germany stood at 24.4 percent and Romania was lowest at 3.2 percent.

  • It can be. After the Digital Omnibus on AI, the Machinery Regulation sits in Section B of Annex I, and AI Act requirements reach machinery through delegated acts that must apply by 2 August 2028. From 20 January 2027, the Machinery Regulation already requires notified body assessment for safety components with self-evolving, machine-learning behaviour.

  • Yes. Directive (EU) 2024/2853 treats software, AI systems and digital manufacturing files as products. It applies to products placed on the market after 9 December 2026. A manufacturer is liable for a defective product without proof of fault.

  • No, apart from narrow exceptions. Since 2 February 2025, Article 5 of the AI Act bans AI that infers the emotions of people in the workplace, except for medical or safety reasons. Fines reach €35 million or 7 percent of worldwide turnover, whichever is higher.

Sources

  1. Eurostat, Artificial intelligence by NACE Rev. 2 activity (isoc_eb_ain2), database updated 15 June 2026. ec.europa.eu/eurostat/databrowser/view/isoc_eb_ain2/default/table
  2. Eurostat, 20% of EU enterprises use AI technologies, 11 December 2025. ec.europa.eu/eurostat/web/products-eurostat-news/w/ddn-20251211-2
  3. European Commission, Apply AI Alliance: AI for manufacturing, engineering and construction flagship actions, 8 October 2025. futurium.ec.europa.eu/en/apply-ai-alliance/news/ai-manufacturing-engineering-and-construction-commission-launches-flagship-actions-under-apply-ai
  4. Regulation (EU) 2024/1689 (Artificial Intelligence Act), Official Journal, 12 July 2024. Article 5 and Annex III. eur-lex.europa.eu/eli/reg/2024/1689/oj
  5. Regulation (EU) 2026/1744 (Digital Omnibus on AI), Official Journal, 24 July 2026. eur-lex.europa.eu/eli/reg/2026/1744/oj/eng
  6. European Commission, Machinery Regulation (EU) 2023/1230, accessed 29 September 2026. single-market-economy.ec.europa.eu/sectors/mechanical-engineering/machinery_en
  7. Regulation (EU) 2023/2854 (Data Act), Articles 3, 4 and 50. eur-lex.europa.eu/eli/reg/2023/2854/oj/eng
  8. European Commission, Cyber Resilience Act, accessed 29 September 2026. digital-strategy.ec.europa.eu/en/policies/cyber-resilience-act
  9. Directive (EU) 2024/2853 on liability for defective products, Official Journal, 18 November 2024. eur-lex.europa.eu/eli/dir/2024/2853/oj/eng
  10. Directive (EU) 2022/2555 (NIS2), Annex II, sectors manufacturing and chemicals. eur-lex.europa.eu/eli/dir/2022/2555/oj/eng
  11. Directive (EU) 2016/943 on the protection of trade secrets, Article 2. eur-lex.europa.eu/eli/dir/2016/943/oj/eng
  12. artificialintelligenceact.eu, Digital Omnibus on AI: amended text on machinery and Annex I, 2026. artificialintelligenceact.eu/ai-act-explorer/digital-omnibus/
  13. EU-OSHA, Regulation 2023/1230/EU on machinery, accessed 29 September 2026. osha.europa.eu/en/legislation/directive/regulation-20231230eu-machinery

Method: adoption figures read from the Eurostat API for NACE section C, enterprises with 10 or more persons employed; shares of users derived from the same table. Legal dates from the Official Journal texts and Commission pages. All sources accessed 29 September 2026. This page is general information, not legal advice. Corrections: contact@lindstead.com.