人工智能对新兴市场经济体的经济影响

Economic impact ofAI in emergingmarketeconomies Leonardo Gambacorta,Enisse Kharroubi,Aaron Mehrotra andLivia Pancotto 17 February 2026 BIS Bulletins are written by staff members of the Bank for International Settlements, and from time to timeby other economists, and are published by the Bank. The papers are on subjects of topical interest and aretechnical in character. The views expressed in them are those of their authors and not necessarily the viewsof the BIS. The authors thank Iñaki Aldasoro, Giulio Cornelli, Blaise Gadanecz, Gaston Gelos, PabloHernández de Cos, Benoit Mojon, Tommaso Oliviero, Dan Rees, Frank Smets and Hyun Song Shin forcomments and suggestions. We are grateful toYui Ching Liu and Alper Yildirimfor excellent analysis andassistance, and to Danielle Ritzema for administrative support. The editor of the BIS Bulletin series is Hyun Song Shin. This publication is available on the BIS website (www.bis.org). ©Bank for International Settlements 2026. All rights reserved. Brief excerpts may be reproduced ortranslated provided the source is stated. Economic impact of AI in emerging market economies Key takeaways •The productivity and growth effects of artificial intelligence (AI) vary widely across countries, reflectingdifferences in sectoral composition and in the capacity to adopt and deploy AI. While advancedeconomies (AEs) are generally better positioned to reap the benefits of AI in the near term, substantialheterogeneity exists within emerging market economies (EMEs).•AI preparedness – covering digital infrastructure, skills and institutional capacity – is a key determinantof overall gains, amplifying productivity effects where it is strong and constraining them where gapspersist, particularly in many EMEs.•Closing AI preparedness gaps can support long-term convergence, as stronger infrastructure, humancapital and institutions would enable EMEs to harness AI more effectively, help mitigate labour marketrisks through reskilling and retraining policies, and narrow growth differences with AEs. Artificial intelligence (AI) is emerging as a transformative general purpose technology with far-reachingimplications for real economic activity. While early evidence points to sizeable micro-level productivitygains and labour market effects, the magnitude of these effects at the aggregate level remains uncertain.Cross-countrydifferences in sectoral composition and in preparedness to adopt and deploy AItechnologies shape how strongly AI affects output and employment. As a result, the near- and medium-term growth effects of AI are likely to differ markedly between advanced economies (AEs) and emergingmarket economies (EMEs). Productivity gains and labour market effects The effects of AI on real activity stem partly from its impact on productivity and labour markets. Earlyevidence from empirical studies using micro data suggests that generative AI (gen AI) could bringsubstantial productivity gains, especially by automating parts of non-routine cognitive tasks. Micro studiesgenerally suggest large productivity gains of between 10 and 65%, with strong improvements in coding,consulting tasks and professional writing (Graph 1). Moreover, evidence suggests that AI tends to equaliseworkplace performance by raising the productivity of less experienced employees relative to those withgreater seniority (Graph 1, filled vs empty dots). For example, in software development and coding, juniordevelopers experienced productivity increases of 21–67%, while senior developers saw more modest gainsof 7–26%. However, the equalisation effect refers to performance within narrowly defined tasks (egcoding). Across broader job roles, junior workers may remain more exposed to automation if their jobinvolves a higher share of routine or AI-substitutable tasks, potentially reducing entry-level opportunities. The extentto which these micro-level gains translate into higher economy-wide total factorproductivity (TFP) remains an open question. Aggregate outcomes depend on reallocations across firmsand sectors, as well as on the degree of misallocation in labour and capital markets (Hsieh and Klenow(2009)).In addition, productivity gains in AI-exposed sectors may be partly offset by labour shiftingtowards lower-productivity activities with limited scope for automation, consistent with Baumol-typeeffects (Baumol (1967)). Finally, differences in adoption speed and complementarities with skills andorganisational practices can further dampen measured aggregate productivity gains. Consequently, the estimated magnitudes for economy-wide gains vary widely (Graph 1, bars).Acemoglu (2025) reports an annual increase in TFP of 0.07%, one of the lowest estimates of AI’s benefitsfor macroeconomic productivity. By contrast, Aghion and Bunel (2024), Bergeaud (2024) and Filippucci etal (2024) estimate larger TFP gains, about one order of magnitude larger (0.3–0.9 percentage points peryear), in part due to higher estimates of industries’ A