Low- and middle-income countries often make the same mistake with every new technology wave. They chase the Silicon Valley model, throwing resources at private companies and hoping market forces will magically create equitable access. Now they’re doing it again with a infrastructure, and this approach will fail spectacularly.
The numbers tell a scary story about AI infrastructure concentration.
Stanford’s 2025 AI Index reveals that U.S. private AI investment reached $109 billion in 2024. That is 12 times China’s $9.3 billion and 24 times the UK’s $4.5 billion. Nearly 90% of notable AI models now come from industry, up from 60% in 2023. Meanwhile, research shows that lower-ranked universities have published an average of six fewer papers at AI conferences since deep learning’s rise, creating clear “haves and have-nots” in AI research.
Sign Up Now for more DPI insights
Many development practitioners are wringing their hands about this concentration while recommending the same tired solutions: more public-private partnerships, startup incubators, and capacity building programs.
India has quietly been building something different—and other LMICs should pay attention.
India’s Digital Public Infrastructure Gambit
India isn’t trying to beat Silicon Valley at its own game. Instead, they’re treating AI infrastructure as Digital Public Infrastructure (DPI): shared, government-controlled systems that provide foundational services.
According to a new white paper from India’s Office of the Principal Scientific Adviser, this approach deliberately positions AI building blocks as Digital Public Goods rather than proprietary assets.
The contrast with market-led approaches is clear.
Where Silicon Valley concentrates AI capabilities in private hands, India is building modular, interoperable systems that expand access.
- IndiaAI Kosh platform has already onboarded 5,722 datasets and 251 AI models from 54 entities across 20 sectors.
- IndiaAI Compute Portal operates over 38,000 GPUs and 1,050 TPUs at subsidized rates under Rs. 100/hour—half the global commercial rate.
This isn’t just about cheaper compute. As we’ve noted before, when AI and DPI converge properly, they can democratize access to high-quality public services globally. But the key phrase is “done right.”
Four Lessons Other LMICs Should Learn
India’s approach suggests they understand something about a investments most countries miss.
1. Government ownership for core infrastructure
The white paper reveals that India’s approach emphasizes public control over AI infrastructure while enabling private innovation on top. Their Telangana Data Exchange demonstrates federated data sharing without requiring central pooling. This allows for collaboration while maintaining data sovereignty.
This directly contradicts conventional wisdom that governments should get out of the way and let markets work. I’ve watched too many countries cede control of digital infrastructure to foreign companies, only to face sovereignty challenges later. India shows you can maintain public ownership while fostering innovation.
2. Modular, phased implementation
Instead of trying to build everything at once, India has developed its AI infrastructure in phases. They started with lighter-weight elements like directories and metadata standards, then added more complex features like federated data access and coordinated compute exchanges.
This phased approach matters because it allows institutional capacity to develop alongside technical systems. Too many LMIC governments try to leap to advanced solutions without building the governance frameworks to sustain them.
3. Standardization enables competition
India’s DPI approach creates common protocols and interfaces that private companies must use to access the system. This is similar to how their Unified Payments Interface processes over 12 billion monthly transactions while maintaining government control over rules and standards.
The result isn’t reduced competition—it’s more equitable competition. Startups in smaller Indian cities can now train AI models at subsidized rates, while universities without on-premise infrastructure can conduct advanced research. The playing field becomes more level, not less competitive.
4. Scale matters more than sophistication
India’s approach prioritizes broad access over cutting-edge capabilities. Their Bhashini platform hosts over 350 language AI models covering 17+ Indian languages—not because they’re the most advanced models available, but because they serve real users with actual problems.
This focus on scale and coverage rather than technological leadership makes sense for most LMICs. You don’t need GPT-5 to improve agricultural extension services or streamline social protection programs.
Implementation Challenges Are Real but Manageable
The white paper acknowledges significant hurdles. Building DPI for AI requires consistent technical standards, high-quality metadata, and sustained institutional capacity. Privacy safeguards become critical as access scales, and interoperability proves difficult in fragmented ecosystems.
But these challenges aren’t insurmountable. The US is attempting something similar with their National AI Research Resource pilot, which aims to “democratize access to critical resources necessary to power responsible AI discovery and innovation.” Early results show promise, with over 340 research projects supported across 40 states.
The key is sequencing implementation correctly. Countries should start with basic elements like national AI registries and common data formats before attempting advanced features like federated compute sharing.
The Window Is Closing
As governance experts have warned, “DPI are technologies—and as such are neither good, nor bad—but never neutral.” The orientation chosen now will determine whether AI infrastructure empowers communities or concentrates power.
Most LMICs still have time to choose their approach. But that window is narrowing as private AI companies expand their global footprint and lock in dependencies. Countries that wait too long may find themselves permanently relegated to consumer status in the AI economy.
The choice is between systems designed for public benefit versus private extraction. India’s DPI approach shows there’s an alternative to Silicon Valley dominance. But only if other countries act decisively to claim it.
