In Europe, artificial intelligence is producing measurable productivity gains. A CEPR study concludes that productivity gains have so far benefited employees of adopting firms, without establishing their distribution by profession. A CEPR study covering 12,000 European firms measures an average productivity gain of 4% attributable to AI. According to PwC, the average wage premium associated with AI skills is 62% across all sectors within its global scope; the report does not provide this figure specifically for engineers.
The essentials
- AI generates an average productivity gain of 4% in European firms, according to CEPR 2026.
- PwC measures a global average premium of 62% for AI skills, without a breakdown establishing +62% for engineers nor 0% for workers or operators.
- In Germany, AI-labelled job postings increased from 72 to 288 between 2022 and the first quarter of 2026. Measured gains are concentrated mainly in medium and large firms, not in professional functions identified by the study.
- The European Pay Transparency Directive, which entered into force on 6 June 2023, establishes a harmonized reporting obligation at EU level, with national transposition by 7 June 2026 at the latest, which could make these gaps visible.
- The issue for 2027-2032: whether European collective bargaining institutions can impose the sharing of gains before gaps become entrenched.
Four percent productivity, zero percent premium for operators
The 4% figure may seem modest, but scaled to an economy, it represents tens of billions of euros of value created. The question of who captures this gain remains wide open.
The European Investment Bank published in 2026 “AI adoption, productivity and employment: Evidence from European firms” using data from more than 12,000 non-financial firms in the European Union and the United States. Firms adopting AI see their productivity rise. In the sample studied, AI adoption is associated with higher wages at the firm level; the study does not break down this effect by professional category. According to available data, certain skilled roles in AI benefit from high wage premiums. Premiums associated with certain AI skills are insufficient to demonstrate a gap in wage progression compared with other professional categories.
This result is no surprise for anyone who has followed the trajectory of the German labor market. AI-labelled job postings there increased from 72 to 288 between 2022 and the first quarter of 2026, according to Indeed Hiring Lab data. The multiplication of job titles mentioning AI signals rapid diffusion of AI-related requirements in job postings, without by itself demonstrating actual adoption by firms. Measured gains are more concentrated in medium and large firms; distribution across skill categories remains uncertain.
Rodrik’s thesis tested against European data
Economist Dani Rodrik has developed, notably in his recent work on shared prosperity, a thesis on technology and labor that 2026 European data directly test. His conviction: technology favors labor only if deployed within an institutional framework that distributes its gains. Without this framework, it reproduces, or even amplifies, existing inequalities between skilled and less-skilled workers.
The sources establish heterogeneity measured by qualification, but not a durable bifurcation of European workers. Skilled services benefit from wage premiums for certain AI skills. Available data do not allow us to assert that less-skilled services suffer automation without compensation. Available data do not allow us to assert that European gains from robotization have primarily benefited shareholders and senior management rather than operators, nor that this mechanism is repeating with AI.
Rodrik argues the solution lies through a third type of industrial policy: aiming no longer only at sectoral competitiveness or green transition, but at job quality in labor-intensive services. His recent work, Shared Prosperity in a Fractured World, poses the problem directly: if services become the new engine of growth, they risk reproducing the inequalities of industry.
The data confirm average wage increases in adopting firms, without allowing conclusion about the distribution of gains by profession. On the solution, opinions diverge.
The liberal reading: institutions also brake adaptation
Another reading of the same phenomenon exists, carried notably by liberal economists like Philippe Aghion. Within the framework of Schumpeterian growth theory, wage bifurcation between skilled and unskilled is a transitory phase inherent to any technological shock. Rigid institutions, fixed collective agreements, seniority-indexed wage scales, inflexible labor law, prevent less-skilled workers from negotiating their upskilling and brake firms’ investment in continuous training.
According to this reading, the 62% premium for engineers is a healthy market signal: it attracts talent toward rare and useful skills. The zero premium for workers is a different signal: it indicates that investment in their upskilling lags, partly because firms see no immediate incentive, and partly because European vocational training systems remain compartmentalized.
This reading does not invalidate Rodrik’s. It complements it. Redistribution institutions are necessary, but their design matters as much as their existence. A collective agreement that guarantees salary without guaranteeing training can protect in the short term and weaken in the medium term. Germany offers a glimpse: IG Metall, the metalworking industry union, has negotiated continuous training clauses linked to AI adoption in certain firms.
These clauses constitute a model, still rare, of redistribution of gains through investment in human capital rather than through wage progression alone.
Concrete effects of the Pay Transparency Directive
The European Pay Transparency Directive, which entered into force on 6 June 2023 with national transposition by 7 June 2026 at the latest, introduces a new obligation: firms with 100 or more employees must communicate gender pay gaps by category according to a staggered schedule, without a general obligation for public disclosure of this detailed indicator. Data are expected in 2027 for the first affected firms.
The instrument is limited. The directive does not impose a general reduction of all gaps, but it does impose corrective measures for certain unjustified gender pay gaps. It sets no redistribution floor, no automatic mechanism for sharing productivity gains. But it produces documentary pressure that employers cannot ignore.
To understand why this matters, one must return to a simple mechanism. Today, available productivity data do not measure gains at the level of an individual quality control position, but at the firm level. The directive changes this opacity. It increases transparency on gender pay gaps and can feed negotiation on this subject, without imposing or measuring the sharing of productivity gains.
European unions have understood this. IG Metall in Germany and CFDT in France have already signaled that data from the directive would constitute a new bargaining argument in coming collective agreement cycles. Transparency as a lever for redistribution: this is the logic of the mechanism. Its effectiveness will depend on the ability of union organizations to seize it quickly, and on firms’ willingness to go beyond minimal reporting.
There is an instructive precedent in this area. Industrial robotics has already shown how gaps between territories widen when institutional adaptation mechanisms lag: regions with a solid foundation of training and integration infrastructure fare better than others, regardless of the level of technology adoption.
The tests facing the European labor market between 2027 and 2032
The 2027-2032 horizon is when the sharing of AI gains will or will not be institutionalized. Initial conditions differ sharply by country.
In Scandinavia, sector-level agreements already include productivity-sharing clauses. Sweden has a tradition of wage indexation to sectoral gains. If Scandinavian unions manage to get AI recognized as a documented and negotiable productivity factor, the model could serve as a continental reference.
In Germany, IG Metall has real bargaining power, but the economy is going through a phase of profound industrial restructuring, partial deindustrialization in automobiles, energy transition, which weakens the position of unions. AI training clauses negotiated in 2026 remain experimental; their generalization is not assured.
In Central and Eastern Europe, the situation is structurally different. Countries like Poland, Hungary or Romania host low-cost production sites that adopt AI to compress margins, in a context where unionization rates are low and collective bargaining institutions are underdeveloped. The pay transparency directive will apply there, but its effectiveness will depend on an institutional fabric still under construction.
Two trajectories are emerging for 2027-2032, without current data allowing us to decide with certainty. In the first, collective bargaining institutions—unions, European works councils, sector agreements—seize the data produced by the directive and build a redistribution framework that links AI productivity gains and wage progression or training for the unskilled. Germany and Scandinavia would be its laboratories. In the second, firms optimize their reporting to satisfy the directive without substantially redistributing, unions struggle to translate data into bargaining power, and gaps become entrenched during the five years when the institutional framework still seeks its form.
The signals to watch are precise. The rate of agreement coverage including an AI-training clause in high-unionization countries will give a direct measure of the first trajectory. The number of firms pursued or sanctioned for non-compliance with the directive will say something about states’ capacity to enforce the reporting obligation. Wage progression for operators in sectors with high AI adoption rates—logistics, agrifood, automotive—will constitute the final indicator.
Europe has assets the United States does not
It would be inaccurate to present Europe as defenseless against this challenge. It possesses a structural advantage that the United States, for example, does not: a dense network of collective bargaining institutions that still covers, in Western Europe, the vast majority of workers. The pay transparency directive is itself proof of regulatory capacity: the United States has nothing comparable at the federal level.
Europe’s capacity to regulate digital technologies has already shown it can anticipate risks that other regions discover belatedly, even if translation into effective standards remains an open agenda.
The European challenge lies in the gap between the pace of technology adoption and that of existing institutions. In Germany, a sector agreement is negotiated every two to three years. AI adoption in job positions, by contrast, progresses within months. Adapting institutions to this pace constitutes the central problem.
Concrete avenues exist. European works councils, which bring together worker representatives from several countries in large multinationals, could become sites for rapid reporting of AI productivity data and accelerated negotiation. Certain firms like Siemens or Bosch have already established digital works councils with an expanded mandate for AI tool governance. The scale remains limited; the model is there.
The question posed by 2026 data does not call for a simple answer. It asks European institutions to decide whether AI productivity gains will be a shared good or a captured rent, and to decide it quickly, before gaps become the norm.
Sources
- Improof Luxembourg, AI and the Labour Market: https://www.improof.lu/en/articles/ia-et-marche-du-travail/
- Dani Rodrik, Shared Prosperity in a Fractured World (2025): https://drodrik.scholars.harvard.edu/shared-prosperity-fractured-world-new-economics-middle-class-global-poor-and-our-climate-2025
- CEPR, AI Adoption and Labour Market Outcomes in European Firms (2026), Aldasoro et al. (no certified URL)
- European Commission, Directorate-General for Economic Affairs, Data on AI adoption in firms, June 2026 (no certified URL)
- Juritravail.com, Analysis of the European Pay Transparency Directive (2026) (no certified URL)
- IG Metall, Sector agreements on continuous training linked to AI (2026) (no certified URL)
