In the 2025 European survey on ICT usage in enterprises, carried out in Luxembourg by STATEC, 65.4% of Luxembourg companies with 10 or more employees report using no artificial intelligence technology, against 79.1% in the European Union. Among them, 10.1% of all companies considered AI without putting it in place, and the survey asks them why. A lack of skills comes first, cited by 7.9% of companies, ahead of uncertainty about the legal consequences (5.6%), fear of breaching data protection (5.5%) and cost (4.0%). On 17 October we published what the survey says about adoption; here is what it says about the companies that have not adopted.
The facts
- Eurostat's isoc_eb_ai table (updated 15 June 2026, consulted 17 October) counts 65.4% of Luxembourg companies with no AI technology in 2025, after 75.2% in 2024; the EU average is 79.1%. The share of companies that considered AI without using it rises from 7.9% to 10.1% in one year (11.1% in the EU).
- Barriers are measured as a percentage of all companies, multiple answers, among those that considered AI (STATEC table for the structure of the questionnaire and the 2024 breakdown by sector). In Luxembourg in 2025: lack of skills 7.9% (EU 7.8%), uncertainty about the legal consequences 5.6% (5.9%), data protection 5.5% (5.8%), availability or quality of data 4.6% (4.8%), incompatibility with existing systems 4.0% (4.6%), cost 4.0% (4.2%), ethical considerations 3.4% (2.7%), technologies judged not useful 1.8% (2.0%).
- Companies that use AI obtain it first as ready-to-use commercial software: 19.3% of all Luxembourg companies in 2025, against 6.5% for systems developed or modified by external providers, 5.5% for open-source software modified in-house, 5.1% for in-house developments and 4.9% for commercial software modified in-house (STATEC table for 2024). The survey is carried out by STATEC among companies with 10 or more employees in the non-financial business economy (methodological note): 2,504 companies surveyed, 2,077 responses used.
One company in ten considered AI without putting it in place, and cites skills first
Set against the 10.1% of companies that considered AI, the barriers read differently: nearly eight in ten cite a lack of skills, just over one in two legal uncertainty and data protection, four in ten cost. These shares are calculated by us from Eurostat's two figures, and a company can tick several reasons. The Luxembourg ranking is the same as the European one, with one difference: ethical considerations weigh more in Luxembourg (3.4% against 2.7%), cost and technical incompatibility slightly less.
| Barrier | Luxembourg | European Union |
|---|---|---|
| Lack of skills | 7.9% | 7.8% |
| Uncertainty about the legal consequences | 5.6% | 5.9% |
| Data protection and privacy | 5.5% | 5.8% |
| Availability or quality of data | 4.6% | 4.8% |
| Incompatibility with existing systems | 4.0% | 4.6% |
| Cost judged too high | 4.0% | 4.2% |
| Ethical considerations | 3.4% | 2.7% |
| Technologies judged not useful | 1.8% | 2.0% |
Companies that took the step buy AI ready to use
Of the 33.6% of Luxembourg companies that use AI, 19.3% of all companies use it as ready-to-use commercial software, a subscription or a feature built into an existing tool. In-house developments (5.1%) and systems built by providers (6.5%) remain a minority. The most widespread technologies confirm it: 25.0% of companies analyse text with AI, 14.2% generate images, video or sound, 13.9% generate text, speech or code, 13.0% convert speech to text. Machine learning for data analysis (6.6%) and image recognition (4.4%) come far behind.
The purposes reported follow the same logic. AI serves first the organisation of administrative processes or management (9.2% of companies, against 6.2% in the EU), ICT security (5.9%), accounting and finance (5.7%), research and development (5.6%). Marketing or sales, at 5.7%, is the only purpose where Luxembourg sits below the European average (6.9%), as we wrote in our article of 17 October.
What the survey measures, and what it leaves out
The question on barriers is only put to companies that considered AI. The 55% of Luxembourg companies that do not use it and did not consider it say nothing about their reasons: the survey cannot tell whether they have no use for it, had no time to look, or were never asked. Barriers are reported, multiple choice, with no ranking requested from the company: the order comes from frequencies, not from a stated priority.
The scope remains companies with 10 or more employees in the non-financial business economy: neither very small companies nor the financial sector appear. Finally, the rise in the share of companies that considered AI without adopting it, from 7.9% to 10.1%, can be read two ways, more companies looked, or more gave up; the survey does not settle it.
Four reference points for a company that considered AI without putting it in place
The reference points below come from our own practice; the survey offers none.
- Skills. The most cited barrier is dealt with in a few days of training on the teams' real tasks, with the tools open in front of them; State co-funding covers 15% of training expenses and 35% of the salary costs of certain participants.
- Law and data. Two of the four main barriers concern the framework: a written usage rule, stating which data must never go into a tool and who validates an output, answers most situations; data that must stay in Europe calls for a chatbot hosted in Europe.
- Cost. Companies that use AI buy it ready to use: subscriptions already paid for, office suites included, often contain the features needed for a first use, before any development.
- Usefulness. For a company that judges the technologies not useful, two tasks tested over a month, with time spent and quality measured before and after, give a surer answer than a comparison with the European average.
These reference points do not say that a company must use AI. They say that the four most cited barriers each have a known answer, and that the answer depends on the company, its data and the tasks it wants to hand to a tool.
AIxH's view
The companies that cite a lack of skills are the ones we meet in training: the tools are often already there, on subscription or in the office suite, and nobody has taken a day to learn to use them on their own files. Our AI training courses take place in person, in Differdange or at the client's; when the barrier is legal or concerns data, our ChatGPT and Claude integration service configures assistants with the excluded data written down in black and white, and the sovereign European chatbot answers organisations whose documents must stay in Europe. Cost, the third barrier, is discussed once the first two are lifted: it is rarely the one the company imagined.
