Does the Rise of Artificial Intelligence Widen Economic Inequality?

Arasmita Mondal · Zenodo (CERN European Organization for Nuclear Research) · 27 Oct 2026

Where this sits in the debate

Will AI actually boost productivity and growth? · its finding: it depends (why studies differ)
Will AI widen inequality? · its finding: it depends (why studies differ)

Direction of the finding as read from the abstract by an AI model (AI-written).

Key takeaway AI-written

A review of recent evidence on whether frontier AI widens inequality, breaking inequality into wage polarization, labour share, capital, market and wealth concentration and cross-country capability gaps, plus a cross-country look at R&D spending and the Gini index for 131 countries. It finds a mixed picture: firm-level productivity gains that can narrow novice-expert gaps alongside concentration of AI skills, capital and compute in certain workers, firms and nations.

The paper in three steps AI-written

Written by an AI agent from the paper's abstract, which is the primary source and is quoted below.

The question

Is the current wave of AI adoption widening economic inequality?

The approach

Synthesises recent work from labour economics, macroeconomics and development economics, splitting inequality into wages, income, labour share, market and wealth concentration, and the gap between countries, plus a cross-country comparison of R&D spending and the Gini index for 131 countries.

The findings

The picture is mixed: generative AI can raise productivity and narrow novice-expert gaps within firms, while AI-related skills, capital, profits and computing capacity concentrate in certain workers, firms and nations.

How to read it: review. A synthesis of existing studies — breadth over novelty; quality depends on what it includes and excludes.

Limitations and counterpoints AI-written

A single-author synthesis; the cross-country analysis uses R&D spending rather than a direct measure of AI adoption.

Abstract

The rapid spread of advanced AI, especially large language models, has renewed the debate over who benefits from technological change. This paper examines whether the current wave of AI adoption is widening economic inequality. To answer it, I pull together recent empirical work from labour economics, macroeconomics, industrial organization, and development economics, focusing specifically on frontier AI systems used across labour markets, firms, financial markets, and national economies. The evidence base here draws on the National Bureau of Economic Research, the IMF, Stanford's Digital Economy Lab, MIT Sloan, PwC's Global AI Jobs Barometer, UNCTAD, Oxfam International, the International AI Safety Report 2026, and a handful of other academic and policy sources. Rather than treating "inequality" as single number, I break it down into several moving parts: wage polarization, income inequality, labour's share of national income, capital accumulation, market concentration, wealth concentration, and the growing gap in AI capability between rich and poor countries. I also conducted a cross-country study using the most recent data from the World Bank World Development Indicators for 131 countries for which data were available for both indicators (aggregates and regional groupings excluded), covering research and development expenditure (percent of GDP) and the Gini index. What emerges is a mixed but fairly consistent picture. The evidence is a mixed bag at different levels of the economy. At the firm level, generative artificial intelligence can increase worker productivity and, in some settings, narrow the performance gap between novices and experts. At the same time, skills, capital, profits and computing capacity related to AI are becoming concentrated in certain workers, firms and nations. At the worker level, however, the evidence is mixed with some studies finding productivity gains for less-experienced workers and others finding growing demand for AI-related skill

About this publication

Preprint. A preprint — posted publicly before or without journal peer review; findings may still be revised.

Published: 2026 (recently)
Citations: 0 (OpenAlex)