At the Fortune Innovation Forum in Kuala Lumpur on November 17, 2025, Mahesh Uttamchandani, the World Bank’s Regional Practice Director for Digital in East Asia & Pacific and South Asia, delivered a message that resonated across emerging markets:
Stop trying to compete with Silicon Valley’s largest AI companies—and start building “small AI” systems that solve local problems at scale.
His argument challenges the prevailing global narrative, which often equates AI progress with massive models, billion-parameter architectures, and trillion-dollar tech companies. But according to Uttamchandani, the future of AI in developing economies lies not in building the biggest model—but in building the right one.
This shift in perspective could redefine how nations invest in technology, train talent, and build digital public infrastructure for the next decade.
The Rise of “Small AI”
For years, AI discussions have centered on size:
- The biggest models
- The largest compute clusters
- The highest training budgets
- The most complex architectures
But this race is dominated by a handful of companies—OpenAI, Google, Anthropic, Meta, and a few well-funded Chinese groups. Competing directly with them is not just difficult; it is economically unrealistic for most nations.
Uttamchandani argues that developing countries should instead focus on systems that are:
- Lightweight
- Task-specific
- Affordable
- Deployable on local infrastructure
- Compatible with public-sector needs
These “small AI” tools might not generate viral chatbots, but they can transform governments, businesses, and communities far more rapidly than chasing the frontier of frontier models.
Why Competing With Tech Giants Is a Losing Battle
1. The Cost of Building Frontier AI Is Astronomical
Training a state-of-the-art model today requires:
- Thousands of high-end GPUs
- Massive datasets
- Multibillion-dollar budgets
- Teams of world-class machine learning experts
For most nations—especially in Asia, Africa, Latin America, and the Middle East—this is not feasible.
2. The Talent Concentration Is Unmatched
Top AI scientists overwhelmingly work for a few companies. Replicating that talent density is extremely difficult without offering similar compensation or research opportunities.
3. The Global Compute Shortage
Even wealthy governments face difficulty acquiring cutting-edge chips. Developing nations would be pushed to the end of the supply chain.
4. Frontier Models Solve Global Problems, Not Local Ones
Large models are designed for maximal generality. But nations need tools built for:
- Local languages
- Local agricultural conditions
- Local business regulations
- Local infrastructure constraints
In this context, smaller, specialized models outperform giant general models.
What “Small AI” Can Actually Do
Uttamchandani highlighted that small AI systems can deliver massive economic impact with a fraction of the cost.
1. Transforming Government Services
Smaller models can power:
- Citizen service chatbots
- Automated permit processing
- Land registry verification
- Local tax compliance
- Social welfare eligibility checks
In countries where bureaucracy is slow, this is revolutionary.
2. Supporting Farmers and Rural Communities
Localized AI tools can analyze:
- Soil health
- Weather patterns
- Pest risks
- Market prices
And they can deliver advice in local dialects. This is more valuable than a giant foundation model speaking 40 languages but knowing little about regional crops.
3. Boosting Small and Medium Enterprises (SMEs)
AI tailored for SMEs can help with:
- Inventory forecasting
- Invoice classification
- Micro-loan assessments
- Retail pricing optimization
These are the backbone tasks of emerging economies.
4. Enhancing Digital Public Infrastructure
Many countries are adopting digital identity, payments, and health systems. “Small AI” is better aligned with:
- Privacy requirements
- Sovereignty demands
- Infrastructure constraints
- Local deployment
It strengthens digital sovereignty without ballooning costs.
The Geopolitical Dimension: AI as a Development Equalizer
According to Uttamchandani, the global AI race risks deepening inequality if countries try to chase technologies they cannot afford or maintain. Instead, “small AI” can level the playing field by:
- Enabling local tech ecosystems
- Creating local AI jobs
- Reducing dependency on foreign models
- Accelerating government efficiency
- Boosting productivity across traditional sectors
This turns AI from a global arms race into a development strategy.
How Nations Can Implement a ‘Small AI’ Strategy
1. Build AI on Top of Digital Public Infrastructure
Countries that already have digital ID, payments systems, or health platforms are well-positioned to integrate small AI models to automate and enhance services.
2. Train Local Talent in Practical AI, Not Exotic Research
Instead of chasing frontier AI research, governments should train:
- Data engineers
- Model fine-tuning specialists
- Domain-specific AI developers
These skills create immediate value.
3. Invest in Low-Cost Compute and Open-Source Frameworks
With open-source models evolving quickly, nations can deploy efficient systems without massive investment.
4. Create AI Policy That Encourages Innovation
Regulations should:
- Give clarity
- Protect users
- Enable startups
- Allow public-private partnerships
- Promote experimentation
5. Incentivize SMEs to Adopt AI
Providing grants, tax rebates, and simplified digital adoption pathways can accelerate AI usage across the economic base.
A Turning Point for AI in Developing Economies
Uttamchandani’s message reflects a wider shift happening globally:
AI is no longer only about building the biggest model—it’s about building the most useful one.
For emerging markets, “small AI” represents:
- A realistic strategy
- A cost-effective approach
- A path to digital sovereignty
- A tool to increase competitiveness
- A means to improve public services
As global AI giants compete for scale, developing nations have an opportunity to redefine AI success by focusing on impact, not size.
