Part 1 of 2: Why India’s AI future may run through the edge, not the frontier

At the India AI Impact Summit on February 19, 2026, Prime Minister Narendra Modi unveiled a vision he called MANAV. The acronym is elegant: moral systems, accountable governance, national sovereignty, accessible and inclusive technology, and valid and legitimate systems.

It is also a quietly ambitious bet. For MANAV to mean anything beyond aspiration, one letter will do most of the heavy lifting. The N. National sovereignty. Not just of data. Of intelligence itself.

That distinction matters more than it may seem. And closing the gap between the two is the central challenge India faces as it positions itself as the Global South’s AI anchor.

India’s Inclusion Strategy Is Real and It Is Working

The country’s AI infrastructure story is genuinely impressive. Bhashini is extending AI translation and speech services across India’s linguistic diversity, making governance and public services accessible in citizens’ own languages. The IndiaAI Mission is building affordable compute capacity so that startups and research institutions can access GPU infrastructure that would otherwise remain out of reach.

At the summit itself, Modi’s speech was translated into eleven languages in real time, including sign language, through AI. That is not optics. That is exactly what inclusion-first design looks like in practice.

The government’s instinct is consistent with what built UPI and Aadhaar: design public infrastructure that serves everyone, not just the elite. In a country of 1.4 billion people across hundreds of languages and deep digital inequality, this approach is not idealism. It is necessity.

The Investment Surge Is Significant. But It Is Not What It Appears to Be.

The numbers from the summit are staggering. Microsoft has committed 17.5 billion dollars over four years to expand cloud and AI infrastructure in India. Google has pledged 15 billion dollars over five years, including its first AI hub in the country. Amazon has committed 35 billion dollars by 2030. India is seeking up to 200 billion dollars in data centre investment in the coming years.

These figures are being read as validation of India’s AI ambitions. And they are, partly. Global hyperscalers do not commit capital of this scale to markets they do not believe in.

But there is a distinction that is easy to miss. These are infrastructure investments. Data centres, cloud capacity, distribution rails. They are the pipes through which AI flows into India. They are not investments in building the intelligence itself.

India is attracting enormous capital to host and distribute AI. 

The question MANAV’s sovereignty pillar must answer is whether India will also shape and own the AI being distributed.

Owning the Interface Is Not the Same as Owning the Intelligence

Here is the scenario worth examining carefully.

Citizens interact with AI in their native languages. Public services run on Indian platforms. Startups build applications tailored to Indian realities. The infrastructure is Indian. The interfaces are Indian.

But the core reasoning engines, the foundational large language models, the agent systems that are increasingly making decisions in finance, healthcare, and governance, come from elsewhere.

In that world, India owns the interface. Someone else owns the intelligence.

This is not full sovereignty. It is localisation layered on imported cognition. And the problem is not abstract. When a model is trained primarily on Western internet data, its assumptions travel with it. The way it handles caste, defines family structure, interprets gender roles, weighs individual rights against community values, none of these are neutral parameters. They are baked into training data and reflected in model behaviour. If foundational intelligence remains external, India’s governance of AI becomes reactive. India responds to defaults it did not set, rather than shaping them.

Modi himself noted at the summit that “whose data, his right” defines the sovereignty pillar. That is an important starting point. But data sovereignty without model sovereignty is incomplete. India would own the raw material while someone else runs the factory.

Why India Is Not Leading Frontier Model Development Yet

This is not a failure of talent. India has the engineers. It has the data. The IndiaAI Mission is building the compute access. Three Indian companies even presented their own AI models at the summit, which Modi rightly highlighted as evidence of domestic momentum.

The gap is structural, not aspirational. Frontier AI development requires sustained compute at a scale that compounds over time, long investment horizons with uncertain returns, and a continuous loop of training, deployment, learning, and retraining. Nations and companies that began this loop earlier now benefit from compounding advantage. The frontier is not a single breakthrough. It is an accumulative process that rewards early starters.

India’s incentives have naturally favoured utility and inclusion. Local language models, agricultural advisory tools, public health AI, and legal access platforms deliver immediate, visible public value. They are defensible to voters and policymakers. Frontier model development demands massive investment with returns that are diffuse and long-term. Democracies tend to fund what they can point to.

This is rational. It is also strategically incomplete.

The Urgency Is Real

Catching up in AI is not like catching up in manufacturing, where you can replicate processes, license technology, and scale. It is closer to catching up in network effects. The model that has more data learns faster. The company with more users generates better feedback. Early advantage compounds in ways that make the gap harder to close with each passing year.

DeepSeek’s emergence from China demonstrated that the frontier is not permanently closed to latecomers. But it also demonstrated how much sustained state and institutional commitment it requires to break in.

India has the talent pipeline, the data diversity, and increasingly the political will. What it needs now is clarity on the path.

And that path may not look like what most people assume.

Chasing frontier model development the way OpenAI or DeepSeek have done it is one option. It is expensive, uncertain, and requires India to win a race already in motion. But there is another path, one that is more aligned with India’s democratic architecture, its linguistic diversity, and its genuine structural advantages. One that could let India leapfrog the wrapper economy without replicating the concentrated model of AI development it is trying to escape.

That path runs through something smaller, and paradoxically, more sovereign.

In Part 2, we examine what a federated ecosystem of regional small language models could mean for India’s AI future, why it may be a more durable form of sovereignty than a single national LLM, and why India is better positioned than any country in the world to make this model work.

END OF ARTICLE