3.4.2 Leapfrogging Opportunities

Priya Devi runs a small healthcare startup in Bengaluru, India. Her company builds AI-powered diagnostic tools for rural clinics—places where doctors are scarce, equipment is minimal, and patients travel hours for basic care.

Her flagship product is a smartphone app that analyzes photos of skin conditions and provides preliminary diagnoses. It doesn't replace doctors. But in villages where the nearest dermatologist is 200 kilometers away, it's transformative.

The app runs on cheap Android phones. It works offline. It's trained on images from Indian patients, so it recognizes conditions common in India that Western diagnostic tools miss. And it costs less than $5 per month—affordable for clinics operating on tiny budgets.

Priya didn't build cutting-edge AI. She used open-source models, fine-tuned them on local data, and optimized for constraints—limited connectivity, low-end hardware, non-expert users. But in doing so, she created something more useful for her context than anything Silicon Valley has produced.

This is leapfrogging: using new technology to skip over intermediate stages of development, solving problems in ways that weren't possible with older approaches. India didn't build a universal healthcare system with doctors in every village. But it might deliver healthcare access through AI-powered tools that work where traditional infrastructure doesn't exist. The question is whether the Global South can leverage AI to leapfrog developmental gaps, or whether infrastructure and resource constraints will prevent leapfrogging from happening at scale.

The Leapfrogging Precedent

Leapfrogging isn't new. It happened with mobile phones. Many developing countries never built extensive landline networks. When mobile technology became cheap and accessible, they skipped landlines entirely, going straight to mobile. Kenya's M-Pesa became a global leader in mobile payments—not despite lacking traditional banking infrastructure, but because lacking it created demand for alternatives.

AI offers similar possibilities. Countries without extensive legacy systems—bureaucratic, industrial, financial—could adopt AI-driven approaches faster than countries locked into existing infrastructure. A country without a national healthcare database could build one using AI to aggregate and analyze health data from the start. A country without entrenched financial institutions could deploy AI-driven credit scoring that draws on alternative data—mobile usage, utility payments—to assess creditworthiness for populations with no formal banking history.

This is the optimistic leapfrogging vision: constraints become opportunities, and AI enables development pathways that bypass the slow, expensive buildouts that industrialized countries required.

The Small AI Revolution

The scenario described above is made possible by what researchers call "Small AI"—applications designed to run on everyday devices without requiring major underlying digital infrastructure. Small AI doesn't need gigawatts of data center capacity, high-speed internet, or expensive GPUs. It is designed for constraints: limited compute, intermittent connectivity, low-end hardware.

It is already delivering real impact. Healthcare apps diagnose diseases from smartphone photos. Agricultural tools provide crop advice via SMS. Educational platforms deliver personalized learning on basic tablets. Financial services assess credit risk using mobile usage data. These are not cutting-edge systems—they are practical tools, optimized for the contexts where they are deployed and solving local problems with appropriate technology.

Small AI enables leapfrogging precisely because it does not require the infrastructure that developing countries lack. It works with what exists, and what exists across much of the Global South is mobile phones, not data centers. The uptake is already visible: more than 40% of ChatGPT's global traffic originated in middle-income countries—led by Brazil, India, Indonesia, and Vietnam—by mid-2025. Globally, generative AI job vacancies surged ninefold between 2021 and 2024, with one in five of those positions in middle-income countries. These figures signal genuine engagement: developing countries are adopting AI, building AI workforces, and positioning themselves in the global digital economy. Whether this adoption translates into sustained development or remains superficial is the central question.

The Infrastructure Prerequisite

Leapfrogging requires something to leapfrog with, and even Small AI has prerequisites. Unreliable power grids make digital infrastructure unfeasible at scale; solar and off-grid solutions help, but deploying them broadly is expensive. Connectivity is essential—AI applications that work offline still need periodic data updates, cloud backups, and support access, which requires at least some broadband penetration. Device access matters too: smartphones are ubiquitous in middle-income countries, but in lower-income settings, ownership rates are lower, handsets are older, and processing capabilities are limited. Finally, even well-designed applications require users with sufficient digital literacy to interact with them effectively.

The implication is that leapfrogging doesn't mean skipping infrastructure—it means prioritizing different infrastructure. Mobile connectivity over landlines. Distributed energy over centralized grids. Appropriate technology over cutting-edge. These substitutions are viable, but they still cost money and require institutional commitment. Countries at "stage zero"—without basic electricity or connectivity—cannot leapfrog. AI can help them move faster once foundations are in place, but it cannot substitute for those foundations entirely.

The Policy Imperative

Leapfrogging does not happen automatically. It requires deliberate policy choices, and the research is clear about what those choices look like. The World Bank advocates for swift integration of AI into critical low-income country domains—health, education, energy, and governance—paired with policies ensuring open access to models, affordable compute, targeted education, digital infrastructure investment, and technology transfer. With these in place, AI can enable new services and meaningful productivity gains. Without them, the technology tends to benefit those who already have access.

The specific policy requirements cluster around five areas, as summarized below.

Policy Area What It Requires
Digital infrastructure Reliable electricity, broadband connectivity, and affordable devices as a baseline
Education and training Skilled local workforces—not just AI researchers, but technicians, implementers, and end users
Regulatory frameworks Rules that protect citizens without stifling adoption; both over- and under-regulation carry real costs
Local innovation ecosystems Startups, research institutions, and government support ensuring AI addresses local needs
Open access to models and data Shared models and public datasets allowing countries to adapt existing work without starting from scratch

These requirements are achievable, but they demand political will, institutional capacity, and sustained investment—all of which are scarce in countries managing competing priorities like poverty, healthcare, and basic infrastructure. The policy challenge is not identifying what needs to be done, but maintaining the coherent, long-term effort required to do it.

The Indian Example

India is the closest thing to a leapfrogging success story. Its national compute capacity has crossed 34,000 GPUs, making it one of the most extensive AI infrastructures in the developing world. The National AI Mission, launched in March 2024, commits $1.25 billion over five years to AI research, infrastructure, and sector-wide adoption. India hosts one of the world's largest AI talent pools, with over 6 million people employed in its tech and AI ecosystem, and shows some of the highest rates of company-level AI adoption globally.

More importantly, India is deploying AI for development at scale: digital identity through Aadhaar, financial inclusion through the UPI payment network, healthcare delivery through rural diagnostic tools, agricultural advisory services reaching smallholder farmers. These are not pilots or proof-of-concepts—they are operational systems serving hundreds of millions of people. Rural startups building smartphone-based diagnostic tools, now reaching tens of thousands of patients each month across hundreds of clinics, represent one small but telling piece of this broader ecosystem.

But India's success required massive public investment, sustained policy focus, and the structural advantage of scale—a large population generating the data and demand needed to make AI systems viable. Most Global South countries lack those advantages. The question leapfrogging optimists must answer is whether the Indian model can be adapted for Congo, Bangladesh, or Nepal: smaller, poorer countries with weaker institutions and less capacity to sustain the kind of coordinated effort India has managed.

Conditions for Success

Research on successful leapfrogging via AI converges on a consistent set of enabling conditions. Technology must be appropriate to the local context—designed for it, not imported from elsewhere and expected to work. Public-private partnerships are typically essential, with governments providing policy support and infrastructure while the private sector supplies innovation and implementation capacity. Local talent must be developed to deploy and maintain systems rather than relying indefinitely on foreign expertise. Ecosystems need to remain open, with access to shared models, public datasets, and collaborative research enabling adaptation without starting from scratch. And development must be demand-driven, solving real local problems rather than replicating products built for wealthy-country contexts.

None of these conditions is impossible to meet. But meeting all of them simultaneously, in countries with constrained resources and limited institutional capacity, requires a level of coordinated effort that many governments struggle to sustain. The gap between what is possible in principle and what gets achieved in practice is where leapfrogging potential most often stalls. Leapfrogging is possible; it is happening in places. But it is not automatic, and it is not occurring fast enough on its own to close the AI development divide without deliberate, sustained, and well-resourced intervention.

Key Takeaways

  • Leapfrogging has real precedent. Mobile phones and payments demonstrated that developing countries can skip intermediate infrastructure stages when new technology is accessible and appropriate. AI creates analogous opportunities.
  • Small AI is already delivering impact. Applications designed for constrained environments—limited connectivity, low-end hardware, non-expert users—are reaching populations in healthcare, agriculture, education, and finance across the Global South.
  • Infrastructure requirements shift, but don't disappear. Leapfrogging redirects infrastructure investment rather than eliminating it. Basic electricity, connectivity, devices, and digital literacy remain prerequisites, not optional extras.
  • Policy is decisive. Digital infrastructure investment, education and training, balanced regulation, local innovation ecosystems, and open access to models and data are all necessary conditions—none sufficient alone.
  • India shows what is possible, with caveats. India's leapfrogging success at scale depended on public investment, institutional capacity, and population size that most developing countries cannot directly replicate.
  • The opportunity is genuine but not self-realizing. Broad leapfrogging across the Global South will require sustained, coordinated, and well-resourced effort—the technology enables it, but policy and institutions must deliver it.

Sources:

Last updated: 2026-02-25