In the summer of 2023, a programmer using GitHub Copilot — an AI coding assistant trained on billions of lines of open-source code — could write working software two to four times faster than without it. A radiologist at a well-resourced hospital using FDA-approved AI diagnostic tools could review more scans with higher accuracy. A lawyer at a major firm using AI document review tools could process discovery materials in a fraction of the time.
In a rural clinic in Malawi, there was no AI diagnostic tool. On a construction site in Bangladesh, there was no AI productivity assistant. In a smallholder farm in Kenya, there was no AI-optimized precision agriculture system. In a school in rural Mississippi, there was no AI tutoring system.
The AI revolution is real. Its benefits — productivity gains, diagnostic accuracy, scientific acceleration, educational personalization — are genuine and, in the aggregate, very large. But they are not uniformly distributed. They are, to a striking degree, concentrated in the populations and geographies that were already most advantaged.
This is not an accident of technology. It is a predictable consequence of how AI systems are built, where they are deployed, who controls them, and who can afford access to them. Left unaddressed, the AI productivity gap between high-income and low-income contexts could become one of the most significant drivers of global inequality in the coming decades — more durable and harder to close than most previous technological divides.
The Three AI Divides
The digital divide in AI is not a single gap but three overlapping ones.
The infrastructure divide: AI systems require computing infrastructure — data centers, high-speed internet connectivity, reliable power, and technical support ecosystems — that is deeply unequally distributed globally. A frontier AI model inference run requires data centers with millions of dollars of specialized hardware. Using AI tools via the cloud requires internet connectivity of adequate speed and reliability. Training large AI models requires computing capacity that is currently concentrated in a handful of US and Chinese cloud providers. The countries with AI development capacity are a small subset of the world’s countries.
Approximately 2.6 billion people — one-third of the global population — do not have internet access. Of those who do, connection speeds and reliability vary enormously: 5G service in a Korean city versus satellite internet in a Tanzanian village represent categorically different AI access environments. The countries with the weakest infrastructure — which are disproportionately in sub-Saharan Africa, South Asia, and the Pacific — are precisely the countries where AI-powered improvements in healthcare, education, and agricultural productivity would have the highest marginal impact.
The data divide: AI systems are trained on data, and the quality and quantity of data available reflects existing patterns of documentation, digitization, and resource allocation. High-income country healthcare systems have decades of digitized patient records; low-income country health systems often have paper records or none. Languages with more online text — English, Mandarin, Spanish — have better-performing AI language tools than languages with less online representation — Swahili, Amharic, hundreds of indigenous languages. Agricultural AI systems trained on temperate-zone farming data perform poorly when applied to tropical smallholder farming.
The result is that AI tools built on global data distributions inherit and amplify the informational advantages of already-advantaged populations and contexts. An AI medical diagnostic system performs better for patients who resemble the dataset it was trained on. An AI language learning system offers richer experiences in major world languages than minority ones. The data divide is not merely an access problem; it is a representation problem that shapes what AI tools can do for underrepresented populations even when access exists.
The skills and complementarity divide: AI tools amplify the capabilities of people who have the skills to use them effectively. A programmer who knows what to ask GitHub Copilot to write gets dramatically more value from it than a programmer who does not. A doctor who understands AI diagnostic outputs and knows how to integrate them with clinical judgment extracts more value than one who does not. An educator who can design AI-augmented curriculum gets more value than one simply handing students AI tools without pedagogical integration.
The skills that make AI tools most valuable — technical literacy, critical evaluation of AI outputs, ability to prompt effectively, domain expertise that allows intelligent interpretation of AI assistance — are unequally distributed in ways that correlate with existing educational and economic advantage. AI is a capital good that multiplies the productivity of skilled workers more than unskilled ones. In labour markets, this amplifies income inequality between skill levels.
The Productivity Gap: How It Compounds
Productivity gaps, once established, compound through time.
An economy in which businesses, researchers, and workers can use AI tools to accelerate output, reduce errors, and solve problems faster will grow faster than one without equivalent access. That growth generates investment, infrastructure improvement, and skill development that further expands AI capability. Meanwhile, the economy without AI access grows more slowly, falling further behind.
This is not theoretical. It is the historical pattern of technological diffusion: countries that adopted industrial technology early built lead times that took generations to close. Countries that adopted information technology early captured the productivity gains of the 1990s and 2000s that contributed to widening global income gaps during that period.
If AI generates productivity gains of the scale that optimistic projections suggest — Goldman Sachs estimated that AI could add 7% to global GDP, concentrated in countries with high AI adoption — and if those gains are as concentrated by geography and income level as current access patterns suggest, the result would be a significant widening of global inequality at precisely the moment when climate change, demographic pressure, and political instability make that inequality more dangerous.
What Could Be Done
The digital divide in AI is not immutable. Several interventions could meaningfully alter the trajectory.
Open-source AI development: Large, capable AI models released under open licenses — like Meta’s LLaMA series — reduce the barrier to AI access by removing per-use pricing. They allow researchers, developers, and organizations in low-income countries to run and adapt AI models on local infrastructure without paying per-query fees to US cloud providers. The trade-off is reduced safety oversight and the potential for misuse, but the access benefits are real.
Data generation for underrepresented populations: Deliberate investment in creating high-quality training data in underrepresented languages, health contexts, and agricultural environments would improve AI performance for populations currently underserved by AI trained on skewed data distributions. This is the kind of public good that markets systematically underprovide and governments or foundations need to fund.
AI for development design: Rather than waiting for AI tools built for high-income contexts to eventually diffuse to low-income ones, targeted investment in AI applications specifically designed for development contexts — disease burden, agricultural conditions, educational infrastructure — in partnership with local institutions can produce tools actually useful to underserved populations.
Compute access programs: International initiatives to provide cloud computing access for researchers, universities, and health systems in low-income countries at subsidized or free rates — analogous to how the internet was made available to academic institutions before commercial pricing took hold — could meaningfully expand the population with the infrastructure to use AI.
AI governance with equity mandates: AI regulatory frameworks that require impact assessment for distributional effects, that mandate accessibility, or that condition market access on equity commitments would build distributional considerations into AI deployment decisions rather than leaving them as afterthoughts.
None of these will close the AI divide overnight. The infrastructure deficits, skill gaps, and data representation problems that underlie it took decades to create and will not resolve quickly. But the time to begin is not when the gap has become permanent — it is now, before AI-driven productivity compounds into an economic structure that makes the gap structurally self-reinforcing.
The promise of AI is that it could accelerate human capability at a speed and scale previously impossible. The risk is that it does this primarily for those who already have the most capability, in a world that cannot afford to leave the rest further behind.
Also explore:
AI in Healthcare: Revolution, Risk, and the Patient Privacy Dilemma (published August 13)
Africa’s Strategic Moment: The Continent the World’s Powers Are Fighting Over (published August 10)
AI Regulation: Who Controls the Controllers? (published June 23)
Related Articles
- Algorithm Political Polarization 2026: Unveiling the Digital Divide Shaping Society
- Navigating AI Art Ownership: Who Really Owns Digital Masterpieces?
- Understanding AI Art Copyright Issues
Weekly Geopolitical Briefing
Go Beyond the Headlines
Analysis of the forces shaping the world - delivered every week. No noise, no bias, just depth.
Subscribe Free ?No spam. Unsubscribe anytime.
About the Author
António Monteiro
Engineer by profession, geopolitical analyst by conviction. I believe responsibility for the planet's future doesn't belong only to governments and institutions - it belongs to all of us. Knowledge about geopolitics, international conflicts, and the forces shaping the world is the most powerful tool for becoming more conscious, informed citizens. You don't need to be a diplomat to understand what's at stake - you just need to want to go beyond the headlines. At Outside The Case, I analyze conflicts, power dynamics, and global trends with rigor and accessible language, so you can understand what's really happening in the world.
Read more about the author →