The World Inequality Report 2026, coordinated by Lucas Chancel and his colleagues at the World Inequality Lab, argues that inequalities do not result from inevitability but from political and institutional choices. Early sectoral data on AI adoption offers elements for examining this question in the context of recent technological change. Early sectoral data on AI adoption provides elements of measurement for this question.
The Essentials
- Atlanta Fed/NBER surveys show that productivity gains linked to AI are concentrated in certain sectors in the United States, notably finance and skilled services, and that this gap could result from institutional choices.
- Atlanta Fed/NBER surveys document gaps in productivity gains across sectors between 2025 and 2026, comparing notably finance, skilled services, and low-wage services, though based on statements from executives and broad sectoral categories. Aggregate GDP does not make this sectoral distribution visible.
- Atlanta Fed/NBER surveys show gaps in productivity gains across sectors and firm sizes in the United States, without establishing a single measure of AI’s overall impact on growth between advanced economies and low-income countries.
- The authors advocate for progressive taxation, notably on very large fortunes, and for strengthening labor rights.
- Governments lack a reliable indicator to steer AI economic policy, although American national accounts include sectoral accounts; aggregate GDP alone does not provide this level of detail.
Lucas Chancel and the World Inequality Lab
Lucas Chancel directs the World Inequality Lab at the Paris School of Economics, which he co-directs with Thomas Piketty. His work focuses on measuring global inequalities and their links to climate policy; he notably coordinated the World Inequality Report 2022, which brought the question of wealth concentration into public debate. The 2026 report addresses multiple dimensions of global inequality, including climate, gender, finance, taxation, and political divides.
His co-authors, Ricardo Gómez-Carrera, Rowaida Moshrif, and Thomas Piketty, each bring distinct expertise: labor market econometrics for Gómez-Carrera, political economy of development for Moshrif, and the long-term reading of capital-labor relationships for Piketty. The result is less an advocacy piece than a measurement instrument: the report documents before prescribing, which gives it an evidentiary force that more activist work often loses.
The Central Thesis: AI Inequality Is Institutional, Not Technological
The report’s thesis can be formulated thus: available data show heterogeneous adoption across sectors and firm sizes, without establishing a general rule on the redistribution of benefits. The report analyzes several broad dimensions of inequalities, notably in training, capital, labor regulation, and fiscal systems. The report addresses the role of institutions in distributing income and wealth, within a broader framework than productivity linked to AI. Others stagnate or fall behind.
This is a thesis that has nothing self-evident about it in its measurement. The gap in productivity gains between sectors is documented in the short term by sectoral surveys. Over time, gaps could widen or narrow depending on AI diffusion into less-skilled sectors. Aggregate GDP makes this sectoral distribution of gains difficult to see. Diane Coyle, whose work on measuring progress in the digital age laid the foundations for this type of analysis, has precisely shown that traditional economic statistics render invisible the transformations that matter most for well-being.
The 2026 report provides sectoral documentation of this.
The report addresses general economic and regional inequalities, including fiscal systems and access to human capital, without causally establishing these elements as direct factors in the observed gaps in AI gains. These are choices, not data.
The Blind Spot of GDP: When Measurement Betrays Policy
One thread running through the report deserves attention: the question of indicators. An economic policy based solely on aggregate GDP lacks sectoral detail, even though governments have sectoral accounts and statistics to observe this distribution. These gaps are absent from aggregate GDP alone, but not invisible in the full set of national accounts, which include industrial accounts published by the BEA.
This point directly connects to the work of Daron Acemoglu and Simon Johnson in Power and Progress, a work that traces a thousand years of struggle between technological innovation and the distribution of prosperity. Their central argument is that technology obeys the institutions that deploy it. The report formulates general conclusions about institutional choices and inequalities, without specific hypotheses about sectoral gaps in AI gains.
The concrete question posed to governments is the following: should an indicator of sectoral distribution of AI productivity gains be created, published annually, in the manner of disaggregated labor income data published by Nordic countries? Several OECD economists and the G20 Stiglitz Report from November 2025 have supported this direction. The report makes available data on global inequalities, without prescribing a sectoral indicator of AI gains.
Limitations and Objections to the Report
Any honest reading of a report of this scope requires that one name its limitations.
The first is that of time frame. The sectoral data cited cover a short window; NBER and Atlanta Fed surveys concern 2024-2026, that is, the early years of generalized adoption of generative AI. The history of electrification and automation suggests that productivity gains from generic technologies take ten to fifteen years to diffuse throughout the entire economic fabric. It is therefore possible that the current gap between qualified sectors and low-wage sectors is an effect of temporal sequencing as much as a permanent institutional effect.
The second limitation concerns the counterfactual. The report analyzes fiscal and political institutions within the general framework of inequalities, but does not take a position on a causal direction regarding the distribution of AI gains.
The third concerns the comparative corpus. The report compares advanced economies to low-income countries by treating the former as a relatively homogeneous bloc. Yet data on AI adoption in Europe show concentration in large firms that varies significantly by country and sector. The average masks very different trajectories between, for example, Sweden and Greece, or between manufacturing Germany and the rest of the Union.
These limitations do not destroy the book. They situate it: it is a diagnosis of the current trend, not a law of AI economics.
What Is at Stake Between 2025 and 2040
The report’s forward-looking dimension is its most ambitious part, and the most useful for understanding why this work matters beyond academic circles.
The report examines several possible trajectories of political and institutional realignment in the face of AI adoption.
In the first scenario, the absence of specific policy lets current mechanisms deepen. Qualified sectors capture increasing productivity gains, income gaps widen within advanced economies, and populist pressure on commercial and fiscal institutions intensifies. This is the status quo scenario, neither the most probable nor the most improbable, but the most documented by current trends. Data on industrial automation in Germany show that technology can create qualified jobs in a favorable institutional context, but this context is not the norm.
In the second scenario, taxation adapted to AI productivity gains, a form of tax on sectoral windfall profits linked to technological adoption, directly finances continuing education in less-qualified sectors. Several countries have opened this path without going all the way: Portugal reduced its general corporate income tax rate from 21% to 20% in its 2025 budget, several countries are exploring training mechanisms linked to technological adoption. These signals are insufficient to validate the trajectory, but they indicate that the mechanism is politically feasible.
In the third scenario, labor regulation conditions AI adoption on agreements for gain-sharing with employees, on the model of Scandinavian productivity agreements, extended to gains linked to digital tools. Denmark has begun integrating clauses of this type into certain sectoral collective agreements. This is the weakest signal of the three, it concerns sectoral experiments, not national policies, but it is also the one that would best respond to the precise mechanism the report identifies.
The authors acknowledge that they cannot predict the actual pace of AI adoption in low-wage sectors. If this adoption accelerates in the second half of the 2020s, the productivity gap could narrow through natural diffusion before any public intervention. If it stagnates, the gap widens and the costs of later catch-up increase.
The signals to watch are simple to name: the evolution of the capital-labor ratio in sectors with high AI adoption, the share of training income in collective agreements signed after 2026, and the productivity gap between qualified and non-qualified workers measured by NBER and Atlanta Fed sectoral surveys at the 2028-2030 horizon. These three indicators will say whether the trend documented by the report stabilizes or deepens.
Interest of the Report
The World Inequality Report 2026 is not a book to read in order to know the figures of global inequalities; the World Inequality Lab’s databases are accessible online and more exhaustive than any possible summary. One reads it for the thesis and for the method.
The thesis, that AI inequality is institutional, is defensible because it is falsifiable. The authors do not claim that technology is inherently bad, nor that markets are structurally incapable of distributing the gains from progress. The report documents the concentration of income and wealth and discusses redistributive policies. And they indicate the levers that could change direction.
The report primarily measures income and wealth inequalities according to positions in the distribution and regions, as well as other dimensions such as gender, offering a perspective complementary to national aggregates. It offers policymakers a framework of analysis that national statistics do not yet provide. For anyone working on AI economic policy, from human resources to public finance to vocational training, it is a usable diagnostic tool.
The report is addressed to researchers, public policymakers, and union and employer representatives who negotiate the conditions of AI adoption in their sectors. It is dense, technical in its methodological notes, but readable in its main argument. To anyone who wants to understand why two countries at the same income level can diverge sharply due to AI, it offers a structured and empirically grounded answer. The question that remains open at the end, how institutions that organize work and social protection evolve in the face of these transformations, is exactly the one that this report poses without claiming to answer it alone.
Bibliographic Information - World Inequality Report 2026 - Lucas Chancel, Ricardo Gómez-Carrera, Rowaida Moshrif, Thomas Piketty (eds.) - Polity Press - Published December 10, 2025
Sources
- World Inequality Report 2026, Chancel, Gómez-Carrera, Moshrif, Piketty (Polity Press, 2025)
- Atlanta Fed / NBER, AI productivity surveys 2025-2026 (sectoral data on AI productivity gains, finance vs. low-wage services)
- Daron Acemoglu & Simon Johnson, Power and Progress: Our Thousand-Year Struggle over Technology and Prosperity
- G20 Stiglitz Report, November 2025, indicators for distribution of AI productivity gains
- Diane Coyle, work on measuring progress in the digital age (Bennett Institute for Public Policy, Cambridge)
