Artificial intelligence is arriving in developing countries as both a promise and a risk simultaneously. The World Development Report 2026 from the World Bank is the most comprehensive synthesis document to date on this tension: it quantifies it, locates it, and identifies the conditions for managing it. Its conclusion can be summed up in one sentence: the framing note structures priorities around empowering entrepreneurs, strengthening workers’ capabilities, and building trust through governance and institutions.
The Essentials
- AI can amplify productivity gains in developing countries, but only if reliable electricity, vocational training, and quality institutions combine in the same place at the same time.
- In rich countries, 14.2% of jobs are exposed to AI automation; in poor countries, only 4.5%, not by luck, but due to lack of access to tools.
- 16.2% of jobs in developing countries have potential for productivity gains thanks to AI, according to the World Bank report published on August 4, 2026.
- Without targeted public intervention, unregulated AI deployment reproduces the logic of previous waves of globalization: concentrated gains, structural losers.
- The indicator to watch through 2030 is less the rate of AI penetration than the effective combination of three levers in economies attempting this wager.
A Report Taking Stock of the Moment
The World Bank publishes an annual World Development Report centered on a major question. Since 1978, these reports have guided billions of dollars in public policy and conditional lending. Choosing artificial intelligence as the 2026 subject says something about the state of the debate: the World Bank Group, an institution with 189 member countries, considers AI a general-purpose technology capable of transforming many development determinants, particularly jobs, productivity, and public services.
The report is a collective work of the Bank’s research team, coordinated by its chief economists. It draws on original data, case studies from about forty countries, and a review of recent economic literature. Its ambition is explicitly twofold: to measure the exposure of global labor markets to AI, and to identify the conditions under which this exposure transforms into opportunity rather than vulnerability.
The Thesis: An Amplifier, Not a Leveler
The report’s central idea can be formulated as follows: AI is an amplifier of what already exists. When the necessary complements are present, AI can substantially improve productivity; the report does not support the idea of a general tenfold increase. Where these conditions are lacking, it passes by without leaving much behind, or worse, it widens the gap with those benefiting from its gains.
This thesis rests on two series of figures. In rich countries, a significant share of jobs is exposed to AI automation, meaning they are susceptible to partial or total replacement. In poor countries, a more limited share of jobs faces this risk. Low exposure to automation is primarily explained by the more manual structure of jobs; parallel deficits in AI access also limit the potential for increases.
Automatable tasks require digital infrastructures and skills that many developing countries have not yet deployed at scale.
A significant share of jobs in these same developing countries presents potential for productivity gains through AI, when AI amplifies certain tasks without replacing them and increases value added per worker. But this potential is conditional. The report emphasizes the need to invest coherently in infrastructure, skills, data, institutions, and the business environment. Bringing these complements together in low-income economies remains difficult.
The Trap of Absent Conditions
By identifying required complements, the report also maps their absence in many countries. This mapping describes pre-existing inequalities.
Electricity first. Much of sub-Saharan Africa and South Asia still suffers from frequent blackouts. Reliable power supply and connectivity are important for broad and intensive AI diffusion, but certain lightweight applications can be designed to function in resource-limited contexts. The data centers needed to train models are even more demanding. A reliable power supply is essential for broad and productive AI diffusion, but certain adapted tools can be used in resource-limited contexts.
Training next. Gains depend on usage skills and the quality of work organization, but AI can also help less experienced workers by giving them access to expertise that was previously rare. This requires a level of digital skills that the education systems of many developing countries do not yet produce at the required scale. Training for AI jobs remains largely to be built, including in advanced economies, so the challenge is structural, not simply quantitative.
Institutions finally. The report considers institutions, rules protecting against harm, regulatory capacities, infrastructures, data, and skills as important complements to AI. The report indicates that concentration of market power can benefit AI more to capital holders than workers and calls on governments to address harms when markets do not sufficiently protect people.
The Rodrik Test: When Conditions Are Industrial Policy
This logic of complements is not new in economic literature. It is at the center of Dani Rodrik’s work on localized industrial policy. In his recent work, notably Shared Prosperity in a Fractured World, Rodrik argues that unregulated free trade produced structural losers precisely because it assumed homogeneous market conditions where economies were deeply heterogeneous. Globalization’s gains concentrated where institutions, infrastructure, and human capital were already present to capture them.
The World Bank report applies this diagnosis to AI with new precision. The report recommends public action to develop AI complements, support local adaptation, and regulate risks; it does not explicitly qualify this strategy as localized industrial policy in Rodrik’s sense. Without adapted policies, AI can concentrate gains, create dependencies, and amplify certain inequalities; the note does not explicitly assimilate it to previous waves of globalization.
Economists from the market liberal stream, including Philippe Aghion, emphasize the risks of public intervention: resource misallocation, regulatory capture, slowdown in innovation. His work on Schumpeterian growth emphasizes the role of competition and creative destruction.
Economies that successfully completed their industrial transformation in the twentieth century (South Korea, Taiwan, Singapore) combined public investment in infrastructure and training with progressive integration into global markets. Public intervention set the conditions for the market rather than replacing it.
Points the Report Leaves Open
Intellectual honesty demands: the report diagnoses with rigor, but leaves several implementation questions open. The first is that of resources. Building reliable electrical networks, reforming education systems, and strengthening institutional capacities costs enormously and takes time. The report examines the potential and conditions for its realization. It mentions the role of multilateral donors and public-private partnerships, but without quantifying what a credible effort should represent.
The second question is that of AI governance itself. Dominant models are developed in the United States and China. Developing countries are, for the most part, consumers of technologies designed elsewhere, for other markets, in other languages. The report addresses the question of local adaptation of models, but without precisely measuring what technological dependence on these few actors represents, a geopolitical dimension that global concentration of semiconductor production makes even more acute.
The third question, and perhaps the most delicate, is that of speed. AI evolves at a pace that public policies struggle to follow even in rich countries. For developing countries that must simultaneously build infrastructure, train workers, and reform institutions, the risk is that conditions are met when tools have already mutated, and advantages captured by others.
Plausible Trajectories Through 2035
In the medium term, AI could contribute to reducing global productivity inequalities or amplifying them. The data allows us to sketch two plausible trajectories, without it being possible to favor one with certainty today.
In the first, countries that invest in necessary complements can strengthen their productive base and better absorb successive AI evolutions. Potential productivity gains can be realized progressively when supported by coherent public policies. Development financing institutions play a catalytic role by conditioning their loans on the implementation of three levers rather than on tool deployment alone.
In the second, stronger AI adoption in economies with necessary complements can widen productivity gaps with less well-equipped countries. Low-income countries find themselves in the situation of textile producers facing mechanization in the nineteenth century: exposed to competition from more productive producers, without having had time to equip themselves with the tools to compete. In the short term, direct exposure to AI replacement is lower in economies where employment is predominantly manual, but risks of disruption, particularly for exports, are not insignificant. The way industrial value chains reconfigure under pressure from industrial policy gives an idea of what this scenario can concretely produce.
The effective combination of necessary complements in middle-income economies constitutes an element for assessing these trajectories. It would rather be the effective combination of necessary complements in middle-income economies. If within five years data shows that countries like Senegal, Bangladesh, or Colombia have simultaneously reduced their electrification deficits, developed large-scale digital tools training programs, and strengthened their regulatory frameworks, then the optimistic trajectory becomes defensible. If these conditions remain disconnected—a few megawatts here, a few trainings there, fragile institutions everywhere—the pessimistic trajectory will gain plausibility.
The report clearly indicates that the answer depends on political choices being made now, not on a technological dynamic that would impose itself.
The Report’s Interest
The World Development Report 2026 is first addressed to public decision-makers, donors, and development economics researchers. But it deserves a broader reading for two reasons.
The first is that it offers broad analysis of AI’s implications for development, including employment, without establishing itself as the most rigorous global mapping available of labor market exposure. Figures by region, by job type, and by national income level are a solid working base for anyone wanting to reason on this subject without relying on intuitions.
The report examines AI’s effects on economic catch-up and consolidation of acquired advantages. The report indicates that these effects depend on choices made.
For a French-speaking reader seeking to understand AI stakes beyond use cases in advanced economies, this report is a solid entry point. It does not say everything will be fine. It says that specific, identified, financeable things can make a difference. That is already much.
Bibliographic Information
Title: World Development Report 2026: The Promise of Artificial Intelligence Author: World Bank Group Publisher: World Bank Publication date: August 4, 2026
Sources
- World Bank, Press Release, August 4, 2026: “AI Offers a Lifeline to Developing Economies”
- Dani Rodrik, Shared Prosperity in a Fractured World: A New Economics for the Middle Class, the Global Poor, and Our Climate, Harvard Kennedy School, https://drodrik.scholars.harvard.edu/publications
- ILO (International Labour Organization), studies on employment and AI, 2025-2026 (without URL: consult the ILO website)
- OECD, Skills Outlook 2025 (without URL: consult the OECD website)
