TL;DR
- Less robots, more data: many of the published AI results in factories come from systems that predict, inspect and plan, using sensor readings, images and production data rather than new machines.
- Measured results exist: factories in the World Economic Forum’s Global Lighthouse Network report double-digit cuts in defects, lead times and energy use, with AI behind many of the top use cases.
- Adoption is uneven: about 20% of EU companies used AI in 2025, according to Eurostat data. Among large firms it is more than half, among small firms 17%. In Bulgaria the overall figure is 8.5%.
- People and data are the main obstacles: the most common reason EU firms give for not using AI is a lack of expertise, ahead of legal and data protection concerns.
- Small manufacturers can start small: one focused problem, ready-made software and “test before invest” services through European Digital Innovation Hubs lower the entry barrier.
Contents
- What AI in manufacturing actually means
- Where AI already works in factories
- Results from real factories
- Who is using it, and who is not
- Why most projects stall
- How smaller manufacturers can start
- Rules to know in the EU
- FAQ
When people picture AI in a factory, they usually picture robots. Robots matter, and their numbers keep growing, but many of the AI results companies publish are quieter. A camera spots a defect a person would miss. A model warns that a spindle will fail in the next day and a half. A forecast tells the planner how much to produce next week. This explainer looks at where AI is already used in factories, what results companies have published, and why many projects never get past the pilot stage. For the broader context of the sector, see What Is High-Tech Manufacturing.
What AI in manufacturing actually means
Industrial robots follow programmed movements. AI adds the ability to learn patterns from data and act on them: recognising a scratch in an image, predicting a failure from vibration data or finding the best production sequence. The two are often combined, but they are not the same thing.
| Classic automation | AI in manufacturing | |
|---|---|---|
| How it works | Fixed rules and programmed movements | Learns patterns from data |
| Typical input | Sensor thresholds, set paths | Images, vibration, temperature, orders, maintenance logs |
| Strength | Speed and repeatability | Spotting what rules cannot describe |
| Weak point | Breaks when conditions change | Only as good as the data it learns from |
| Example | A robot arm welding the same seam | A camera model rejecting a part with a hairline crack |
The International Federation of Robotics counted 542,000 new industrial robots installed worldwide in 2024, more than double the figure ten years earlier. Asia accounted for 74% of them and Europe for 16%. Yet several of the results in the examples below come from software running on top of existing lines.
Where AI already works in factories
| Area | What AI does | Typical data |
|---|---|---|
| Quality inspection | Finds defects in images or sounds faster and more consistently than manual checks | Camera images, X-ray, acoustic signals |
| Predictive maintenance | Warns before a machine fails, so repairs can be planned | Vibration, temperature, power use, maintenance history |
| Planning and forecasting | Predicts demand and sets production sequences | Orders, sales history, capacity |
| Energy | Finds where energy is wasted and adjusts settings | Meter and machine data |
| Process control | Tunes settings to reduce scrap and cycle time | Process parameters, quality results |
| Generative design | Proposes part shapes that use less material | Design constraints, simulations |
| Industrial copilots | Help engineers write machine code, find instructions and document work | Manuals, code, work orders |
Two examples show how this looks in practice. At the Siemens plant in Amberg, Germany, an AI model predicts whether a circuit board needs an X-ray test, as an alternative to buying an additional X-ray machine for about €500,000. Predictive maintenance on a milling spindle there warns 12 to 36 hours before a failure. At Bosch, as of 2023, AI-based anomaly detection runs in around 50 plants on more than 2,000 production lines, according to the company.
Results from real factories
The World Economic Forum’s Global Lighthouse Network recognises factories that show measurable results from advanced technology. By its September 2025 announcement it included 201 sites, and the WEF said AI was enabling up to half of the top use cases implemented. The figures below are reported by the companies through the programme.
| Site | What changed | Reported result |
|---|---|---|
| Lenovo, Monterrey (Mexico) | Over 60 digital solutions, more than half AI-based | Lead time −85%, quality losses −56%, productivity +58% |
| Siemens Numerical Control, Nanjing (China) | 50+ AI use cases | Lead time −78%, field failures −46% |
| Schneider Electric, Évreux (France), distribution centre | Circularity programme | Energy use −18%, single-use plastic −40% |
| Tüpraş, İzmit (Türkiye) | AI forecasting | Delivery reliability from 85% to 95% |
| Unilever, Hefei (China), distribution centre | 65% of use cases AI or generative AI | Forecast accuracy +39% |
| Kunlene Film Industries, Suzhou (China), SME | 30+ digital use cases built in-house | R&D lead time −45%, defects −31% |
Sources: WEF, September 2025 and WEF, January 2026.
These are leading sites, not typical ones, and not all of them are factories. They show what is possible when data, people and processes are ready, not what an average factory should expect in its first year.
Who is using it, and who is not
About 20% of EU companies with ten or more employees used at least one AI technology in 2025, up from 13.5% a year earlier, according to Eurostat data. Denmark led with 42%. Bulgaria was among the lowest at 8.5%, up from 6.5% in 2024.
Company size matters as much as country.
| Company size | EU, 2025 | Bulgaria, 2025 |
|---|---|---|
| 10–49 employees | 17.0% | 7.2% |
| 50–249 employees | 30.4% | 13.3% |
| 250+ employees | 55.0% | 26.2% |
Source: Eurostat data compiled by the Bulgarian Chamber of Commerce and Industry AI Council.
Even manufacturers that use AI often use it outside the production floor first. An analysis by the Czech National Bank of Eurostat data found that manufacturers most often apply AI to administration, marketing, sales and accounting, with production processes and logistics further down the list.
Why most projects stall
Surveys point less at the technology and more at people and data.
| Barrier | What the data says |
|---|---|
| Skills | 70% of EU firms that considered AI but did not use it cited a lack of relevant expertise; in Bulgaria 73% (Eurostat via BCCI) |
| Legal uncertainty | 54% cited unclear legal consequences |
| Data protection | 53% cited data protection and privacy concerns |
| Talent on the floor | In Deloitte’s 2025 survey of 600 US manufacturing executives, human capital was the least mature area, and 48% reported moderate to significant challenges filling production and operations management roles |
| Data foundations | 78% of the same executives put more than 20% of their improvement budget into smart manufacturing, with data, sensors and cloud as the foundation |
| Scaling | Across all industries, McKinsey’s 2025 survey found 88% of companies use AI somewhere, but only about a third have started to scale it |
The pattern is consistent. A model trained on incomplete or messy machine data gives unreliable answers, and a working pilot that nobody on the shop floor trusts or understands does not spread to other lines. McKinsey also found that the companies with the strongest results were far more likely to redesign workflows around AI rather than add it on top.
How smaller manufacturers can start
Large plants run dozens of AI use cases. A small manufacturer usually needs one that solves a costly, repeated problem. A practical sequence:
| Step | What it involves |
|---|---|
| Pick one painful problem | Scrap, unplanned downtime or slow quoting, with a known cost |
| Check the data | Is the data already collected, is it clean, and who owns it? |
| Prefer ready-made tools | Camera-based inspection, maintenance and planning software, and copilots built into existing systems avoid building models from scratch |
| Test before buying | European Digital Innovation Hubs offer “test before invest” services, training and financing advice to SMEs |
| Involve operators early | The people on the line know where the problems are and decide whether the tool is used |
| Measure before and after | Define the metric (scrap rate, downtime hours) before the pilot starts |
The WEF’s January 2026 cohort included a smaller manufacturer, Kunlene Film Industries, that built more than 30 digital use cases with its own team. Industrial copilots are also spreading: in October 2024, Siemens and Microsoft reported that more than 100 companies were using their Industrial Copilot for engineering tasks such as generating machine code.
Rules to know in the EU
Two sets of rules matter for AI in machinery.
| Rule | What it covers | Key date |
|---|---|---|
| Machinery Regulation (EU) 2023/1230 | Replaces the Machinery Directive; explicitly covers safety functions based on self-learning AI and autonomous mobile robots | Applies from 20 January 2027 |
| EU AI Act, as amended by Regulation (EU) 2026/1744 | High-risk obligations for AI that is a safety component of regulated products | Postponed to 2 August 2028 |
The amendment, part of the EU’s Digital Omnibus, was published in July 2026 and also changes how AI in machinery interacts with the Machinery Regulation, according to a summary by the law firm Cooley. Manufacturers placing AI-enabled machines on the EU market should check the final text with their legal advisers.
FAQ
How is AI used in manufacturing?
Mostly for quality inspection, predictive maintenance, production planning, energy optimisation and engineering support. Robots are part of the picture, but many results come from software that analyses data from existing machines.
What is predictive maintenance?
It uses data such as vibration, temperature and power use to predict when a machine is likely to fail, so repairs can be planned before a breakdown stops the line.
How many manufacturers use AI in the EU?
About 20% of EU companies with ten or more employees used AI in 2025, according to Eurostat data, with large firms far ahead of small ones. Use varies widely by country, from 42% in Denmark to 8.5% in Bulgaria and 5.2% in Romania.
What stops factories from using AI?
The most common reason EU firms give is a lack of expertise, followed by unclear legal consequences and data protection concerns. Poor or scattered data is another frequent obstacle.
Can a small manufacturer afford AI?
Often yes, if it starts with one specific problem and ready-made software. Testing and training services are available through European Digital Innovation Hubs.
Does the EU AI Act apply to factory machines?
AI that serves as a safety component of machinery falls under product safety rules. The high-risk obligations for such AI were postponed to August 2028, and the new Machinery Regulation applies from January 2027.
Vyara’s take
Much of what I see in industrial companies happens before AI ever reaches the factory floor: they rarely explain online what they actually make, for whom and how well. Their websites list machines and certificates, while buyers, and now AI assistants, look for answers to specific problems. AI in manufacturing is less about robots and more about data, and the same is true for how these companies are found. The main obstacles are people and data, not technology, and smaller manufacturers can start with one problem and affordable, ready-made tools. The ones that document their processes and results clearly will gain twice: better decisions inside and better visibility outside.
Sources: IFR World Robotics 2025 · Siemens Amberg · Bosch generative AI in manufacturing · WEF Global Lighthouse Network September 2025 · WEF Global Lighthouse Network January 2026 · Eurostat data via Luxtoday · Sofia Globe on Bulgaria AI use · BCCI AI Council · Czech National Bank analysis · Deloitte 2025 Smart Manufacturing Survey · McKinsey State of AI 2025 · European Digital Innovation Hubs · Siemens and Microsoft Industrial Copilot · EU-OSHA Machinery Regulation · Cooley on the Digital Omnibus
About the Author
👉Want guest posting?🪙Subscribe to the @egorithms Telegram