AI and Economic Growth: BIS Study on Advanced vs Emerging

AI and Economic Growth: What the BIS Study Reveals

The rapid proliferation of generative artificial intelligence since late 2022 has sparked intense debate about its macroeconomic consequences. While tech optimists project transformative productivity gains, sceptics warn of job displacement and uneven benefits. A rigorous new working paper from the Bank for International Settlements — BIS Working Paper No. 1321 by Gambacorta, Kharroubi, Mehrotra, and Oliviero (December 2025) — provides some of the first empirical evidence on how generative AI is already affecting economic growth across 56 countries and 16 industries.

The study’s central finding is striking: AI’s short-run growth impact depends critically on the interaction between an industry’s exposure to AI-automatable tasks and a country’s overall AI preparedness. Advanced economies with strong digital infrastructure, skilled workforces, and supportive regulatory environments are positioned to capture significantly larger gains from generative AI than emerging markets with weaker fundamentals. For policymakers, business leaders, and investors navigating the AI transformation, these findings offer concrete, data-driven guidance on where the benefits will materialise first — and where urgent investment is needed to prevent a widening global divide. Explore the interactive analysis library for more AI research breakdowns.

Research Design: Industry AI Exposure Meets Country Preparedness

The BIS study employs an elegant empirical strategy inspired by the influential Rajan and Zingales (1998) framework, adapted for the AI age. The core idea is to test whether industries with higher inherent exposure to AI tasks grow differentially faster in countries that are better prepared to adopt AI technology. This interaction — industry-level AI exposure multiplied by country-level AI preparedness — forms the key explanatory variable.

Industry-level AI exposure is measured using the AIIE (AI Industry Exposure) index developed by Aldasoro et al. (2024), benchmarked to US occupational patterns across 16 aggregated sectors. The US benchmark assumption means that the measure captures the technological potential for AI automation based on task composition, rather than actual AI adoption, which varies widely across countries.

Country-level AI preparedness is measured using the IMF’s AI Preparedness Index (AIPI), published by Cazzaniga et al. (2024) for 2023. The AIPI is a composite of four subcomponents: digital infrastructure, human capital and labour market policies, innovation and economic integration, and regulation and ethics. The final sample covers 56 economies (29 advanced, 27 emerging), 16 industries, and focuses on the 2022–2023 period — the window of rapid generative AI adoption following the launch of ChatGPT. Value-added data comes from the Asian Development Bank’s multiregional input-output tables in constant 2010 prices, giving 875 country-sector observations.

Which Industries Are Most Exposed to Generative AI

Not all industries face the same degree of AI transformation. The BIS study’s sector exposure rankings reveal a clear pattern: industries dominated by cognitive, language-intensive, and analytical tasks are most exposed to generative AI, while those characterised by physical labour and manual operations are least affected.

The three highest-exposure sectors are Finance, Education, and Information/IT — all industries where AI can automate significant portions of existing work through text generation, data analysis, code writing, research synthesis, and customer interaction. These sectors employ large numbers of knowledge workers whose core tasks overlap substantially with the capabilities of current large language models and generative AI tools.

At the other end of the spectrum, Agriculture, Transport, and Construction have the lowest measured AI exposure. These sectors rely heavily on physical tasks, spatial manipulation, and real-world interaction that current AI systems cannot effectively automate. While robotics and specialised AI applications are making inroads in these industries, the generative AI revolution primarily affects cognitive work — and these sectors simply have less of it. This sectoral heterogeneity is crucial because countries’ economic structures differ dramatically: an economy dominated by finance and services will experience AI’s impact very differently from one anchored in agriculture and manufacturing. The IMF’s AI research programme provides complementary analysis of these sectoral dynamics.

AI Readiness Index: Gaps Between Advanced and Emerging Economies

The IMF’s AI Preparedness Index reveals a substantial and concerning gap between advanced and emerging economies. The median AIPI score for advanced economies is 0.71, compared to just 0.54 for emerging market and developing economies — a gap of 17 points on a scale where differences translate directly into differential growth outcomes.

Within advanced economies, the range spans from approximately 0.58 (Croatia, Greece) to 0.78 (Denmark), reflecting variation in digital infrastructure quality, education system adaptability, and regulatory sophistication. Among emerging markets, the dispersion is even wider: from approximately 0.35 (Nepal) to 0.80 (Singapore), with Singapore’s score actually exceeding most advanced economies — demonstrating that income level alone does not determine AI readiness.

The four AIPI subcomponents each tell a distinct story. Digital infrastructure captures broadband penetration, internet speed, and cloud computing availability — areas where emerging markets often face the largest gaps. Human capital and labour market policies measure education quality, STEM skills availability, and labour market flexibility. Innovation and economic integration assess R&D spending, patent activity, and trade openness. Regulation and ethics evaluate data protection frameworks, AI governance, and intellectual property regimes. Interestingly, the BIS study finds that the regulation/ethics subcomponent shows particularly strong statistical significance in driving AI growth gains (coefficient 3.079, significant at 5%), suggesting that regulatory frameworks may be at least as important as raw technological capacity.

Key Empirical Findings: AI Growth Impact by the Numbers

The study’s baseline regression results provide concrete estimates of AI’s short-run growth impact. In the preferred specification with both industry and country fixed effects (Table 1, Specification III), the coefficient on the interaction term (AI exposure × standardised AIPI) is 3.043, statistically significant at the 5% level (standard error 1.394). This means that for a one-standard-deviation increase in a country’s AI preparedness, sectors with higher AI exposure experience meaningfully faster value-added growth.

To illustrate the magnitude: comparing sectors at the 90th percentile of AI exposure (e.g., Finance) with those at the 10th percentile (e.g., Agriculture), a one-standard-deviation increase in AIPI translates to approximately 2 percentage points of additional value-added growth for the high-exposure sector relative to the low-exposure sector. This is an economically significant effect for a two-year window (2022–2023).

The model’s explanatory power is reasonable for a cross-country, cross-industry specification, with R-squared values of approximately 0.19–0.20 in the full models. The negative coefficient on the initial share of total value added (-0.186, marginally significant) suggests a convergence effect: sectors that were already large relative to total GDP grew somewhat more slowly, consistent with diminishing returns at the sector level.

Across alternative specifications — removing industry fixed effects, adding controls for robot stock, using AIPI subcomponents instead of the composite — the core interaction remains positive and significant. This robustness across specifications strengthens confidence that the finding reflects a genuine AI-driven growth effect rather than a statistical artefact. Research published in the Libertify interactive library includes additional AI impact analyses for comparison.

Robots, Automation, and AI: Separating the Growth Effects

A critical question for interpreting the results is whether the measured AI effect is truly distinct from broader automation trends, particularly the ongoing expansion of industrial robotics. The BIS study addresses this directly by including robot stock (robots per employee, normalised) as a control variable in robustness specifications.

The results are clear: robot stock is positively and significantly associated with sectoral growth (coefficient 0.304, standard error 0.111, significant at the 1% level), confirming that physical automation continues to drive productivity gains in exposed sectors. However, the AI exposure × AIPI interaction remains positive and statistically significant even after controlling for robots, demonstrating that generative AI operates through a different mechanism — cognitive task automation — that is not captured by traditional robotics.

This distinction has important implications for economic policy. Countries that have invested heavily in industrial robotics (Germany, Japan, South Korea) may not automatically lead in AI-driven growth unless they have also developed the complementary digital infrastructure, AI skills, and regulatory frameworks captured by the AIPI. Conversely, economies with strong AI readiness but limited manufacturing robotics (several service-dominated advanced economies) may benefit disproportionately from generative AI precisely because their economic structures are more exposed to the cognitive tasks that AI automates best.

Country Winners and Losers: Simulated AI Growth Impacts

The study’s country-level simulations (Table 3, Specification III) translate the sector-level findings into aggregate growth effects by weighting each country’s sectoral AI exposure by its actual economic structure. The results reveal substantial heterogeneity both between and within the advanced/emerging economy groups.

On average, advanced economies are predicted to experience a short-run increase in real value-added growth of approximately +0.6 percentage points relative to the lowest-impact country in the sample (Cambodia). Emerging market and developing economies average approximately +0.4 percentage points by the same measure. While this 0.2 percentage point gap may appear modest, when compounded over multiple years it translates into meaningful divergence in living standards.

Within each group, the variation is striking. Among advanced economies, Luxembourg and Hong Kong SAR are singled out for particularly large predicted gains, reflecting their heavy concentration in financial services and information-intensive sectors — exactly the industries with highest AI exposure. Among emerging markets, Singapore stands out as a potential “AI growth champion,” with an AIPI score (0.80) that exceeds most advanced economies and an economic structure heavily weighted toward AI-exposed sectors.

At the lower end, Nepal and Cambodia face the smallest predicted AI growth gains due to a combination of low AIPI scores and economic structures dominated by agriculture and low-exposure manufacturing. These countries risk falling further behind unless targeted investments in digital infrastructure and human capital can shift their AI readiness trajectories. The OECD’s AI Policy Observatory tracks country-level AI readiness indicators that complement the BIS study’s findings.

Why AI May Widen Global Income Inequality in the Short Run

Perhaps the study’s most sobering implication concerns global inequality. The combination of three factors — higher AI preparedness in advanced economies (median AIPI 0.71 vs. 0.54), greater concentration of AI-exposed sectors in richer countries, and the multiplicative interaction between exposure and preparedness — creates a powerful mechanism for widening the income gap between rich and poor nations in the short run.

The arithmetic is straightforward: if AI primarily benefits countries that already have strong digital infrastructure, educated workforces, and sophisticated regulatory frameworks, and if these same countries also have economic structures tilted toward AI-exposed service sectors, then the growth dividend from generative AI will be disproportionately captured by wealthy nations. Emerging economies, many of which rely heavily on agriculture, manufacturing, and extractive industries with low AI exposure, will benefit less — or may even face competitive displacement as AI-augmented competitors in advanced economies capture market share in tradeable services.

This dynamic echoes previous waves of technological change where early adopters reaped outsized benefits while latecomers struggled to catch up. However, the authors note an important nuance: AI preparedness is not fixed. Countries that make strategic investments in the four AIPI dimensions — digital infrastructure, human capital, innovation capacity, and regulatory quality — can substantially improve their position. Singapore’s example demonstrates that an emerging economy can achieve AI readiness scores that rival or exceed advanced economy averages, turning what might appear to be a structural disadvantage into a competitive opportunity.

Policy Priorities: Boosting AI Readiness in Emerging Markets

The BIS study’s findings translate into a clear policy agenda, particularly for emerging market policymakers seeking to capture AI’s growth potential. The four subcomponents of the AIPI provide a natural framework for prioritising investments.

Digital infrastructure — including broadband connectivity, cloud computing access, and data centre capacity — is the foundational requirement. Without reliable, high-speed internet access, AI applications cannot be deployed at scale regardless of other readiness factors. Many emerging economies face significant infrastructure gaps that limit AI adoption not just in cutting-edge sectors but across the economy.

Human capital and labour market policies represent the second critical pillar. The study’s findings suggest that countries with stronger STEM education systems, more flexible labour markets, and better lifelong learning infrastructure are better positioned to absorb AI’s disruptive effects while capturing its productivity benefits. Investments in AI-specific skills training — not just for technology specialists but for workers across AI-exposed sectors like finance, education, and information services — can directly enhance a country’s growth capture from the AI transition.

Innovation and economic integration — measured by R&D spending, patent activity, foreign direct investment flows, and trade openness — determine how quickly AI technologies diffuse through an economy. Countries with stronger innovation ecosystems and more international economic connections can adopt AI tools faster and more effectively, amplifying the growth benefits.

Finally, the regulation and ethics subcomponent — which showed particularly strong statistical significance in the BIS study — highlights the importance of governance frameworks that encourage AI adoption while managing risks. Data protection laws, AI safety standards, intellectual property regimes, and ethical guidelines create the trust and predictability that businesses need to invest in AI deployment. For additional AI policy research, the Libertify interactive library offers comprehensive analysis across multiple BIS and IMF publications.