A new policy brief from the Luxembourg Institute of Socio-Economic Research (LISER), published on 10 March 2026 by Julio Garbers, Christina Gathmann and Terry Gregory, gives the cleanest cross-country read so far on AI adoption in the Greater Region: Luxembourg leads the four-country sample, but the lead is largely a story about the country's industrial composition rather than a Luxembourg-specific advantage in AI uptake.
The headline numbers are striking. 23% of Luxembourg firms use AI, against 16% in Germany, 10% in France, and 8% in Belgium. Across all four countries combined, the share has grown twelvefold over recent years to roughly 12% on average. Luxembourg sits well above that mean.
The methodology that matters
What separates this brief from the older Eurostat ICT Usage in Enterprises survey is the underlying data: the LISER team scraped and analysed text from more than three million firm websites across Luxembourg, Belgium, France and Germany, classifying AI use through a large language model trained on the European Commission's official AI definition.
The trade-off is familiar — websites do not capture every back-office deployment, and they may pick up some AI-marketing puffery — but the upside is a panel that is finer-grained, cross-country comparable, and updateable in something close to real time. For policy work it is, for now, the best available substitute for survey-based statistics with their year-or-more lag.
Why Luxembourg leads — and where the lead vanishes
Once industries are held constant, Luxembourg's headline 23% drops to a sector-adjusted 14.3% — within a hair of Germany's 14.8%. About nine percentage points of Luxembourg's lead come from sector composition. Two industries do most of the work:
- Finance and insurance. Almost 40% of Luxembourg firms in the sector report AI use — predictably, given the prediction-heavy nature of trading, risk, anti-money-laundering, and fund administration workflows.
- Information and communication technology. A similarly high share, in line with the sector that develops, integrates, and resells AI tools.
Below that, the picture flattens fast. AI adoption in Luxembourg is below 10% in health, construction, real estate, and manufacturing — sectors where the country's structural weight is smaller, where the data foundation for productive AI use is thinner, and where the workforce has not yet built up the AI-adjacent skill base that the LISER team identifies as the single strongest predictor.
Skills, not countries
Across all four countries in the sample, workforce skills — concentrations of data analysts, engineers, and roles requiring computational thinking — emerge as the dominant explanator of AI adoption. Country fixed effects matter much less than firm-level skill mix once the regressions are run with controls. The policy implication is sharper than the headline number: Luxembourg's adoption ceiling is set less by the national framework and more by how fast firms in lagging sectors can build, attract, or partner for the relevant skills.
This aligns with the IMF's parallel recommendation, restated in this year's Article IV concluding statement: the Grand Duchy needs to expand reskilling programmes in STEM and ICT skills if its productivity story is going to broaden beyond finance and IT.
What the brief leaves open
The LISER work is descriptive and cross-sectional. Three questions sit just past the boundary of what it can answer:
- Productivity payoff. Adoption is observed; productivity gains are not yet measured at firm level. Whether the 23% headline translates into measurable output per worker remains an open empirical question.
- Generative versus predictive AI. The brief uses the EU's broad AI definition. Disentangling the share of adoption that runs through general-purpose generative tools (which has surged since 2023) from the longer-running predictive-AI footprint would change how policymakers weigh skills versus capital investment.
- Cross-border dynamics. Luxembourg's labour market is exceptionally cross-border. The brief does not yet ask how cross-border worker flows propagate AI skills — a question of practical importance to firms recruiting from Trier, Metz, or Arlon.
Why this brief lands now
March 2026 is an interesting moment for the data to arrive. The EU AI Act's high-risk obligations crystallise in August, AIFMD II is reshaping the Luxembourg fund industry through April, and the cabinet's productivity discussion has moved from "should we" to "where do we focus." The LISER brief reads as a precise answer to that last question: the country's AI lead is real but concentrated, and broadening it is principally a workforce-skills problem — not a regulatory or industrial-policy one.
Expect this to be cited in three places over the next quarter: in the productivity chapters of the Frieden government's mid-year economic communication, in the trade-and-investment narrative used by Luxinnovation and the AI Factory, and in the next iteration of every major consultancy's Luxembourg AI-readiness benchmark.
