AI readiness is a policy choice: evidence from 24 overperforming countries

Rwanda has one of the lowest per capita incomes in the world. It also has a dedicated Responsible AI Office, privacy protection laws aligned with international standards, and an AI readiness score that places it ahead of countries with far greater resources. It is the only low-income country to qualify as an AI overperformer in a new study, and its example raises an important question: what is holding other countries back?

The global artificial intelligence (AI) divide is real. The International Monetary Fund’s 2023 AI Preparedness Index (AIPI) measures countries’ readiness to adopt and integrate AI across four pillars: digital infrastructure, human capital, regulation and ethics, and innovation. The index reveals stark disparities, with high-income countries dominating the upper end of the distribution while most low- and middle-income countries trail far behind. Yet within this landscape, a number of countries are quietly defying expectations, outperforming what their economic structure alone would predict.

Understanding how and why these countries overperform is the focus of the paper. It proposes a replicable, data-driven methodology to identify “global” and “local” AI overperformers by benchmarking AIPI scores against each country’s level of economic complexity, captured through a composite index integrating trade and research data. The Economic Complexity Index (ECI) reflects the diversity and sophistication of what a country produces and exports, serving as a proxy for its broader knowledge intensity and institutional capacity. Countries with more sophisticated economies should, on average, be better prepared for AI adoption. Deviations above this expected relationship reveal the fingerprints of deliberate policy choices.
 

Figure 1. Global Distribution of AI Preparedness (2023)

^{Country-level AIPI scores in 2023, mapped across income groups. The map illustrates the scale and geographic concentration of the global AI divide, with high preparedness scores clustered in high-income countries and the majority of low- and middle-income economies falling significantly below the global median. Source: Taken from Mandon (2025), derived from IMF AIPI Data Mapper.}

Who Are the Overperformers?

The study identifies 24 overperformers in total. Among high-income countries, 10 “global overperformers” emerge: Australia, Denmark, Finland, Hong Kong SAR (China), Japan, the Republic of Korea, the Netherlands, New Zealand, Norway, and Singapore. These are already highly sophisticated economies, with strong ECI scores. What makes them overperformers is that their AI preparedness exceeds even that high baseline: their observed AIPI scores are statistically significantly above what their economic complexity predicts. By contrast, countries such as the United States and Switzerland, with similarly high or even higher complexity scores, are excluded because their AI readiness aligns with rather than exceeds expectations.

Among upper-middle-income countries, seven “local overperformers” emerge: Albania, China, Costa Rica, Indonesia, Kazakhstan, Malaysia, and Ukraine. Among low- and lower-middle-income countries, another seven clear their respective threshold: Ghana, India, Morocco, Rwanda, Sri Lanka, Tunisia, and Viet Nam.

For policymakers in developing economies, the more instructive comparison is not with Singapore or Switzerland – it is with a structural peer that is outperforming them. A country ranked 80th globally may have little to learn from the top tier, but a great deal to learn from a comparable country that has found ways to outpace expectations within similar resource constraints. That reframe is one of the study’s most practical contributions.
 

Figure 2. AI Preparedness and Economic Complexity: Identifying Overperformers by Income Group

^{Population-weighted scatter plots relating AIPI scores (y-axis) to the composite ECI Trade-Research index (x-axis), shown separately for high-income countries (left panel) and low- and middle-income countries (right panel). The regression line represents expected performance given economic complexity. Countries plotted above the line and above their income-group threshold are highlighted as overperformers. Source: Taken from Mandon (2025), derived from IMF AIPI (2023); OEC Economic Complexity Index (2021–22).}

What Drives Overperformance? Regulation First, Then Context

One finding cuts across income groups with unusual consistency: regulation and ethics is the top driver of overperformance everywhere. Overperformers systematically outpace non-overperformers on this pillar even after controlling for economic complexity. The message is that institutions seem to matter enormously for AI readiness, and that building adaptive legal frameworks and ethical governance structures is not a luxury reserved for wealthy nations. Ghana and Rwanda have each established dedicated Responsible AI Offices and enacted privacy protection laws aligned with international standards, positioning themselves as regional leaders in AI governance despite their resource constraints.

Beyond this universal driver, the relative emphasis on other pillars shifts with income level. In low- and lower-middle-income overperformers, human capital and workforce development follow closely behind regulation, reflecting the priority of skill-building in resource-constrained environments. Digital infrastructure receives the least emphasis, limited by fiscal realities. Upper-middle-income overperformers increasingly prioritize infrastructure, as their growing capacity allows for larger investments: China’s massive 5G rollout and Kazakhstan’s construction of Central Asia’s most powerful supercomputer facility are emblematic. High-income overperformers present the most balanced profiles, with regulation, infrastructure, innovation, and human capital all playing meaningful roles.
 

Figure 3. Key Pillars Driving Overperformance Across Income Groups

^{Waffle chart comparing the relative contribution of each AIPI pillar (digital infrastructure, human capital and labor markets, innovation and integration, regulation and ethics) to total AIPI scores for the three groups of overperformers: low- and lower-middle-income, upper-middle-income, and high-income. Each symbol represents 0.5% of the group’s weighted average AIPI score. Source: IMF AIPI (2023).}

Three Coordination Models — and Why They Cannot Simply Be Copied

The paper goes beyond identifying who overperforms to examine how. Drawing on national innovation systems theory, three distinct coordination models emerge from a comparative analysis. The first is a state-led model, characterized by strong central government direction and specialized institutional architecture. Singapore exemplifies this: the Infocomm Media Development Authority coordinates AI strategy across multiple ministries, while the Personal Data Protection Commission ensures ethical compliance. Rwanda demonstrates that this model is not exclusive to wealthy city-states, deploying its Responsible AI Office and Centre for the Fourth Industrial Revolution as analogous coordination instruments under far tighter resource constraints.

The second model is market-responsive facilitation, best illustrated by Malaysia and Kazakhstan, where the state establishes enabling conditions through regulation and public infrastructure while private sector actors drive implementation. Malaysia’s statutory levy system, which requires private operators to fund workforce training, is a particularly instructive mechanism for countries seeking to build AI skills without relying entirely on public budgets. The third model, distributed innovation, characterizes large and institutionally complex countries like India, where multiple ministries, state governments, and private actors operate semi-autonomously within broad national frameworks, enabling regional experimentation but requiring sophisticated coordination to maintain coherence.

A critical caveat runs through all of this: these models are not freely transferable. Singapore’s regulatory approach rests on decades of government-industry trust. China’s infrastructure achievements depend on centralized planning mechanisms unavailable in most of other contexts. The lesson is not to copy specific policies, but to understand the underlying coordination logic and adapt it to one’s own institutional setting.
 

What Policymakers Can Do

The overperformers’ experiences point to three practical priorities for governments in emerging markets and developing economies looking to close the AI readiness gap.

1. Start with governance, not infrastructure. Regulation and ethics is the consistent top driver of overperformance across all income groups. Establishing a dedicated AI governance body, even a lean one, and adopting a privacy framework aligned with international standards are achievable steps that do not require large capital outlays. Ghana and Rwanda are proof of concept.
   
2. Invest in skills early and deliberately. Among lower-income overperformers, human capital and workforce development are the second most important driver. Mechanisms that mobilize private sector funding for training, such as Malaysia’s levy model, offer a way to scale skill-building without placing the full burden on public finances.
   
3. Look sideways, not just up. The most useful benchmarks are structural peers, not global leaders. Countries should actively identify comparable economies that are outperforming them and examine the specific policy and institutional choices behind that gap. The methodology in this study provides one replicable way to do exactly that.
   

The Divide Is Real — But Not Inevitable

The 24 countries identified as overperformers demonstrate that deliberate policy choices, particularly in regulatory governance, workforce development, and strategic coordination among state, private, and civic actors, can generate AI readiness levels that exceed what economic complexity alone would predict. Rwanda’s example shows that even the most resource-constrained governments have more room to move than the headline rankings suggest. For policymakers in emerging markets and developing economies, that is an invitation to look carefully at what their most comparable peers are actually doing, and to ask whether the same logic can be made to work at home.

Source : World Bank