AI is amplifying economic concentration: gains cluster in high-value sectors and advanced economies, widening domestic inequality and the North–South divide; institutional buffers and GVC position largely determine who captures the benefits.

Main Finding

AI is reshaping wealth distribution along two linked axes: within countries (domestic polarisation) and between countries (global North–South divergence). The direction and magnitude of AI’s distributional effects are not intrinsic to the technology but are determined by three contextual conditions—industry/task structure, institutional environment, and a country's position in Global Value Chains (GVCs). These conditions explain why the same AI exposure can produce different inequality outcomes in the United States, China, and Latin America, and why developed countries are likely to capture disproportionate gains from AI at the global level.

Key Points

• Three mediators determine AI’s distributional effects:

  • Task/sector structure: AI exposure matters, but whether tasks are routine (substitutable) or non-routine cognitive (augmentable) drives upward or downward pressure on different wages.
  • Institutional buffers/amplifiers: tax regimes, labour institutions (unions, collective bargaining), education and retraining policies, and digital infrastructure shape how productivity gains are shared.
  • GVC position: countries higher in GVCs (design, R&D, services) convert AI gains into value better than countries confined to low-skill tasks.

• Domestic evidence (three regions):

  • United States: AI tends to increase capital’s share and wage polarization. Evidence: labour income share fell from 67.8% to 58.4% (1987–2019, U.S. BLS); model estimates (Acemoglu cited) suggest AI could further raise capital’s share (~0.31% in cited estimate). Task-exposure concentrated in mid–high wage cognitive occupations; weak collective bargaining amplifies skew toward capital and top-skilled workers.
  • China: AI increases regional and urban–rural inequality. Studies find stronger effects in central/western regions and rural areas—limited industrial diversification and limited reskilling constrain reallocation. One study reports an inverted-U relation between AI and inequality but ~87.3% of Chinese cities remain in the phase where AI raises inequality.
  • Latin America: AI exposure concentrates in the formal service sector; women, younger workers, and formal employees face higher automation risk. A pronounced digital divide limits who benefits—e.g., in Mexico wealthier quintiles are far more likely to hold jobs that can benefit from AI (top fifth ~5.6× compared with the poorest fifth). Informal-sector workers are less exposed (but also less able to gain productivity benefits), promoting class stratification.

• Global evidence:

  • IMF-based projections (Cerutti et al.) indicate developed economies may realize productivity/growth gains more than twice those of low-income countries. Example projections cited: global GDP up ~4% in 10 years under high-TFP AI scenarios; U.S. +5.4%; low-income countries +2.7%.
  • Mechanisms: developed-country firms concentrate R&D, model training, and high-value decision-making; developing countries often provide low-value tasks (data labeling, moderation). AI reduces the need to transfer knowledge on-site, lowering spillovers and constraining industrial upgrading in the Global South.

Data & Methods (used in the review)

• This paper is a literature review that synthesizes empirical micro studies, task-based exposure indices, and macro simulations:
  • Occupational/task exposure indices: AIOI (Felten et al.), task-based exposure measures (Eloundou et al., Cerutti et al.).
  • Microdata studies: household and labour-force surveys (China microdata in Cai et al.; 14-country household surveys in Ciaschi et al. for Latin America).
  • Administrative and macro statistics: U.S. BLS labour income shares; IMF multi-regional macro models and productivity simulations (Cerutti et al.).
  • Structural models: multi-regional dynamic general equilibrium models to simulate cross-country TFP/growth effects.
  • Case and sector studies: firm-level and production-chain analyses (Casilli et al.) documenting outsourcing of annotation and low-value tasks.
• Key metrics discussed: sectoral AI exposure, AI Preparedness Index (AIPI), labour-income share, projected GDP/TFP impacts by country group, measures of digital access/digital divide.
• Limitations noted by the author: few studies directly quantify AI’s effect on overall inequality measures (e.g., Gini) over time; much of the cross-country forecasting depends on assumptions about diffusion, preparedness, and geopolitics.

Implications for AI Economics

Policy and modeling implications: - Heterogeneity matters. Models of AI’s economic effects must include: - Task-level heterogeneity (routine vs non-routine; complementarity vs substitution). - Institutional features (taxation, labour market institutions, education/training systems). - GVC position and knowledge spillovers. - Measurement priorities: - Better micro-to-macro linkage: estimate AI’s effect on income distribution measures (e.g., Gini) directly, not only sectoral exposures. - Track AI preparedness and digital-access metrics to predict who captures productivity gains. - Quantify spillover channels from multinational firms and AI firms to local capabilities in developing economies. - Policy levers to mitigate inequality: - Strengthen institutional buffers: collective bargaining mechanisms, progressive taxation targeting capital rents, social insurance for displaced workers. - Invest in digital infrastructure, education, and retraining to raise AI preparedness and reduce the digital divide. - International cooperation to preserve channels of technology transfer and create mechanisms for shared gains (e.g., capacity-building, negotiated data/technology access). - Research directions: - Longitudinal and causal studies quantifying AI’s net effect on wealth inequality (within and between countries). - Simulations that jointly model domestic redistribution mechanisms and cross-country GVC positions. - Empirical work on how export controls, geopolitics, and platform governance affect global distribution of AI rents.

Concise takeaway: AI’s inequality effects are context-dependent. To predict and influence whether AI exacerbates or mitigates inequality, economic analyses must couple task-level exposure metrics with institutional, regional, and GVC-position variables—and policymakers must act on education, labour institutions, infrastructure, taxation, and international cooperation.