North America’s AI adoption rate sits at 31.65%. The Global South averages 14.1% [Ragenaizer]. That’s nearly a 2:1 ratio, and it’s getting wider, not narrower. Fresh 2025 data confirms what many working in this space suspected: the divide is accelerating. With 5G rollout compounding the infrastructure advantage in wealthier nations and major AI governance frameworks taking shape now, the policy window to course-correct is narrowing fast.
The Growing AI Adoption Gap
Photo by The numbers are stark. AI adoption in the Global North sits at approximately 23%, compared with 13% in the Global South [Campus]. The spread looks even wider at the regional level: North America clocks in at 31.65%, while the Global South averages just 14.1% [Ragenaizer].
The country-level leaderboard tells a more nuanced story. The UAE ranked first globally at 59.4% AI tool usage among working-age adults, followed by Singapore at 58.6%, Norway at 45.3%, and Ireland at 41.7% [Campus]. The US sits at a comparatively modest 26.3%. China is at 15.4% [Campus]. The gap isn’t a simple West-versus-rest narrative. It tracks infrastructure density, policy environment, and capital access.
The AI Preparedness Index (AIPI) makes the compounding nature of this divide clearer. Advanced economies average an AIPI score of 0.68, emerging markets land at 0.46, and low-income countries score just 0.32 [ICTworks]. That bottom tier isn’t simply “behind.” It’s operating in a fundamentally different reality.
Over 80% of global AI investment flows to the US, China, and the EU [ICTworks]. The remaining nations split what’s left. The capital required to build foundational AI infrastructure, including GPU clusters, training data pipelines, and MLOps tooling, simply isn’t reaching the places that need it most. Each funding cycle widens the gap further.
Why the Divide Keeps Widening
Cloud providers often pitch AI as democratized: anyone with a browser can access a frontier model.
Photo by The reality is harder. Mobile data costs 20% of per capita income in Sub-Saharan Africa versus 1% in North America [ICTworks]. Even where 81% of the population lives within mobile broadband coverage, only 30% actually use the internet . You can’t deploy an AI-powered supply chain tool when users can’t afford the data plan to reach it.
The bottlenecks stack up in predictable ways:
-
Infrastructure: Reliable cloud compute, low-latency connections, and stable power grids remain scarce across much of the Global South. Fine-tuning a model on a connection that drops every 20 minutes isn’t realistic.
-
Policy vacuum: Fewer than 10 low-income countries have published national AI strategies. Without regulatory clarity, international partnerships stall and private investment stays away.
-
Talent drain: Skilled ML engineers and researchers from developing nations consistently relocate to higher-paying roles in the US and EU. The pipeline builds capacity abroad, not at home.
-
Language and data gaps: Most foundation models are trained predominantly on English-language data. Deploying them in Swahili, Bengali, or Tagalog contexts requires localization work that few organizations fund.
These aren’t independent problems. They form a self-reinforcing loop: weak infrastructure discourages investment, which limits talent development, which slows policy formation, which keeps infrastructure weak. Breaking the cycle requires hitting several pressure points at once.
What This Means Going Forward
The World Bank puts it plainly: developing countries may experience the notable effects of generative AI (GenAI) faster than the productivity benefits, precisely because of digital infrastructure gaps [World Bank].
Photo by Nearly half of positions that could benefit from GenAI in developing countries are held back by a lack of digital technology access [World Bank].
That’s the core tension. The nations most vulnerable to AI-driven labor displacement are the least equipped to capture AI-driven productivity gains. Agriculture, healthcare, and education, sectors where AI could deliver outsized impact in low-income settings, remain largely untouched by tools already shipping in the Global North.
”The area where the world is most lacking is in regulation and ethics.” — IMF Managing Director Kristalina GeorgievaThat point lands differently when you consider that the nations writing AI regulations are overwhelmingly the ones already benefiting from AI deployment. The Global South risks being governed by frameworks it had no hand in shaping, rules optimized for economies that look nothing like its own.
Paths Toward Closing the Gap
The most promising near-term lever is open-source AI.
Photo by Meta’s LLaMA, Mistral’s models, and a growing ecosystem of open-weight alternatives mean a research team in Nairobi or Jakarta doesn’t need a $100K/year API budget to experiment. When the alternative is a $20/seat/month enterprise license in a market where average monthly income is $200, open weights aren’t just convenient. They’re often the only viable path.
Beyond tooling, three structural shifts could meaningfully narrow the divide:
- Targeted infrastructure investment: 5G and edge compute deployments in underserved regions, funded through multilateral development banks rather than purely private capital.
- Capacity-building at scale: Programs like UNESCO’s AI competency framework and Google’s AI Opportunity Fund have begun training learners in developing nations. Scaling existing efforts matters more than launching new ones.
- Inclusive governance: The UN’s AI Advisory Body has called for binding inclusion requirements for low-income nations in global AI policy negotiations. Whether that translates into real seats at the table remains an open question.
The AI adoption gap between the Global North and Global South is measurable, accelerating, and consequential. The architecture of who benefits from this technology is being decided right now, as major AI policies crystallize through 2026. Open-source models, targeted investment, and inclusive governance represent the most realistic intervention points. The window is open. It won’t stay that way.
