In two years, the salary premium linked to artificial intelligence skills has more than doubled in major economies. It now stands at 62 percent on average according to the PwC Global AI Jobs Barometer 2026; the 25 percent figure reported in 2024 is not directly comparable because the analysis then relied on a limited observation scope. Ravio estimates an average European premium of 12 percent for professional AI/ML roles, and employment remains dense.
The essentials
- In 2026, PwC estimates an average AI salary premium of 61.9 percent across 16 sectors, based on more than one billion job postings in 27 countries and territories; the 25 percent figure reported in 2024 is not directly comparable because the analysis then relied on a distinct observation scope.
- In Germany, where codetermination frames technological decisions, industrial employment remains dense and continuing education is funded by companies within the framework of collective bargaining.
- Daron Acemoglu gave his Nobel Lecture on December 8, 2024, addressing the role of institutions in prosperity.
- Without institutional reform in countries lacking it, the risk of increased market segmentation between an AI segment and other segments with distinct wage dynamics persists.
- The roadmap exists—codetermination, retraining funds, continuing education—but the signals to monitor remain absent from most political agendas in 2026.
The 62 percent premium: what the figure actually measures
The PwC Global AI Jobs Barometer 2026 covers 27 countries and several tens of millions of job postings analyzed. Its measure indicates how much more a worker in a position that incorporates AI skills earns on average compared to a comparable position without that dimension. PwC measures the evolution of this premium between 2024 and 2026, across observation scopes that changed between the two dates.
This progression indicates a widening gap in announced salaries for positions with AI skills. A minority of positions is associated with higher salaries, while most positions do not benefit from this premium. The momentum is accelerating.
Two readings of this figure coexist. The first, liberal, sees in this premium the normal remuneration of a rare and productive skill: the market signals what it needs, workers and companies adapt, and the premium eventually spreads as AI skills become widespread. This reading is consistent with Philippe Aghion’s analysis of creative destruction: each technological wave first creates rents for early entrants, then dissolves them as competition and diffusion do their work.
The second reading, the one Daron Acemoglu developed in his 2024 Nobel Lecture, is darker. Acemoglu distinguishes technologies that augment labor, making each worker more productive, from those that substitute for labor, replacing tasks without creating equivalent jobs. His argument: if labor institutions are absent or insufficient, a substitution technology would concentrate its gains among capital holders and a small segment of workers. The 62 percent premium measures an announced remuneration gap associated with AI skills.
Ravio data published for its 2026 Compensation Trends report show a 12 percent premium for professional AI/ML roles and 3 percent for management positions.
The German exception demonstrates that institutions precede technology
Germany is adopting AI. Its large industrial companies are deploying automated processing tools, logistics optimization, and design assistance. Ravio data, which aggregates salaries in the European technology sector across tens of thousands of positions, show an AI salary premium of 12 percent in this sample; PwC reports a 62 percent premium across job postings in its 27-country scope, with an average calculated across sectors.
This configuration coexists with the existence of a German codetermination framework. In establishments with at least five eligible employees, a works council can be elected; the employer must inform it and consult it on procedures including AI, with specific codetermination rights in certain areas. Employee representatives hold one-third of seats in some companies with more than 500 employees, but representation is parity-based in companies subject to the codetermination law with more than 2,000 employees. This oversight right does not block technology: German companies automate. But it frames certain employer decisions; German codetermination does not inherently impose negotiation of all productivity gains.
In Germany, industrial employment remains dense and companies often finance continuing education; some collective agreements can frame its financing or training time. The two estimates are not directly comparable due to distinct populations, geographic scopes, and methods.
The Journal’s article on German automation documents the mechanism by which institutional regulation can accompany technological transformation. Germany presents a contained AI salary premium in the technology sector, compatible with a labor market endowed with a codetermination framework.
Acemoglu argues that institutions frame the distribution of technological gains. The German case provides a useful point of comparison for examining this proposition.
The liberal reading is not wrong, but it is incomplete
The thesis of spontaneous diffusion deserves to be taken seriously. It rests on a real historical observation: the technology premiums from previous waves—computing in the 1980s, the internet in the 1990s—did indeed decline as skills became widespread. If the AI premium follows the same trajectory, institutional urgency is reduced.
Several labor economists, including researchers at the Federal Reserve Bank of Dallas in a February 2026 note, stress the role of experience and tacit skills in the effects of AI on employment and wages. In this reading, the relevant reform is not codetermination but massive investment in skills.
The distinction leads to different policies: regulation of collective bargaining, on one hand, and financing of training, on the other. These two orientations are compatible, but their priorities, their actors, and their timelines diverge.
The 2026 data show a high premium associated with AI postings, without allowing us to conclude on the existence or absence of spontaneous diffusion of skills. The premium measured in PwC’s 2026 edition remains high compared to earlier measures, although the observation scopes varied between years. Actual AI adoption in small and medium enterprises remains very weak, which rules out any imminent generalization of skills. Between the two most recent editions, PwC reports an increase in the average premium for AI skills, without establishing the relative timing of concentration and diffusion of skills.
Concrete risks for workers outside the AI segment
The 62 percent premium has a downside. Positions that do not incorporate AI do not see their salary decline immediately. However, their wage trajectory can be affected by the presence of alternative technologies in the labor market.
The Dallas Fed notes in its 2026 data heterogeneous wage effects depending on occupation characteristics. Workers in the residual segment do not lose their jobs today, but their wage trajectory diverges from that of the AI segment.
A worker in a position without an AI component generally has limited means to access training or participate in technological decisions affecting their employment.
AI adoption in large European companies is concentrated precisely where workers are least represented in decision-making bodies: platforms, logistics, food service. It is in these sectors that the split threatens to be fastest and most enduring.
Today’s choices determine the situation in 2035
Several trajectories are possible depending on the institutional choices adopted in the coming years.
The first is institutionalized bifurcation. Several labor market trajectories could emerge depending on the institutional framework: an AI segment with high remuneration and segments with distinct wage dynamics.
In this scenario, workers outside the AI segment would face increased difficulties in accessing the gains associated with technological deployment.
The second trajectory is partial convergence through institutional diffusion. If several economies adopt mechanisms for worker representation in technological decisions, the AI premium could shrink. In Germany, the maintenance of industrial employment density coexists with advanced adoption of automated processing and optimization technologies. This trajectory requires difficult political reforms, and its adoption would likely remain partial and heterogeneous across countries.
The collective needs that remain in both scenarios are identical, even if their urgency differs. In the first, maintaining a sufficient level of employment so that the majority remains integrated into the market becomes an explicit political priority, requiring appropriate institutional mechanisms. In the second, the transition remains costly in terms of training and adaptation: workers outside the AI segment need access to genuine skills upgrading, not merely declarative upgrading.
Two concrete paths emerge from available data. The first is extending worker representation in decision-making bodies of companies deploying AI, based on the codetermination model, adapted to national structures. The second is a contribution on productivity gains linked to AI, allocated to a sectoral retraining fund managed on a parity basis. These two instruments already exist in various forms in Europe: the issue is their adaptation to the dynamics specific to generative AI.
The signals that would allow distinguishing between the two trajectories by 2027–2028 are identifiable. The annual evolution of the wage premium gap between countries with codetermination mechanisms and those without. The share of workers outside the AI segment who have accessed certified training financed by their employer. The legislative agenda regarding representation on the supervisory boards of technology companies. These signals are measurable.
They are not yet measured systematically.
The roadmap exists, but the political agenda is behind
Labor institutions must adapt to AI. The speed at which they do so, compared to the speed of technology deployment, is the central issue. In 2026, deployment is rapid. Institutional adaptation is slow.
The European Union has partial instruments. The platform workers directive, adopted in 2024, creates a presumption of employment status for digital platform workers. The AI regulation imposes transparency obligations for high-risk systems used in employment. But neither addresses the question of distributing productivity gains: they regulate working conditions and risks, not sharing mechanisms.
The European Commission has no legislative proposal at this stage for extending codetermination to decisions about AI deployment. Several member states—France, Italy, Spain—have works councils whose information and consultation rights could be extended to automation decisions, but these extensions have not yet been the subject of targeted reforms.
Germany is not an exportable model without adaptation. German codetermination rests on a culture of collective bargaining and union density that do not exist in all countries. Its direct transposition is neither possible nor desirable everywhere. But its principle—that workers have a say in technological decisions affecting their trajectories—is adaptable to diverse institutional architectures.
The 62 percent premium measures a salary gap associated with AI skills. Institutional adaptation to the challenges of distributing technological gains remains a central task in most economies.
Sources
- PwC Global AI Jobs Barometer 2026, https://www.success.com/ai-skills-wage-premium-demand-2026
- Daron Acemoglu, Nobel Lecture: Institutions, Technology, and Prosperity (2025), https://www.aeaweb.org/articles?id=10.1257/aer.115.6.1709
- Ravio European Tech Compensation Data, Q2-Q3 2026, Ravio (quarterly report, no public URL available)
- Federal Reserve Bank of Dallas, note on employment and AI, February 2026, Federal Reserve Bank of Dallas (no public URL available)
