Should India Build Its Own AI Foundational Models?
Analytics India Magazine (Ankush Das)
AI is headed to be the cornerstone of digital transformation worldwide. But, India faces a pressing question: invest in building its own foundational AI models or continue adapting global ones for local use? The debate revolves around sovereignty, cost, innovation, and cultural identity.
The subject was put to test at the Cypher 2025 hosted by AIM in Bengaluru.
The session brought together Jason Joseph, chief information security officer at mPokket, Ashwini Patil, EVP and Head of Product Design at Lentra, and Manish Kumar Purwar, Global IT Head for Sales & Service Technologies.
With India eyeing developed nation status by 2047, the conversation was not just theoretical, it carried undertones of national strategy.
The Case Against Immediate Investment
Joseph struck a cautious note, pointing out that building models from scratch demands immense compute power, data, and resilient infrastructure.
“Do we have the infrastructure that helps us build such models at scale? I would say not yet,” he said.
India, historically, has caught up by improving global technologies rather than reinventing them, he argued. “Let us build on what is existing… and in time make a smooth transition.”
Patil agreed that cost and datasets remain barriers. “One of the numbers said that the Indic dataset that they are using was less than 0.01%. But we cannot rely on those types of datasets.”
However, she disagreed with Joseph’s description of neighbour countries as potential adversaries. She leaned towards adapting what exists for India’s needs, suggesting to work with global partners: “If they have already done some work, we will reuse some of it and build what we want to build on it.”
The Push for Sovereignty and Innovation
Purwar, on the other hand, underscored the strategic risks of overdependence on foreign models. He said he would be doubtful about the success of the industry “if we don’t build the foundation of AI layers.”
He cited the UPI infrastructure, Aadhaar, to stress upon India’s capacity to scale digital systems, adding, “that has put us into the upper quadrant of the world in digital transformation. Why not AI?”
The debate touched upon India’s diversity, with references to Krutrim LLM and Sarvam AI, initiatives that prove local languages and contexts can outperform global models in specific tasks.
For Purwar, the argument stretched beyond technology to identity. “If our model doesn’t support our cultural diversity, probably after a few years everybody would call us an Indian, not the diversity that we have.”
A Middle Path
A middle path emerged as Patil spoke of “pragmatic nationalism”, a strategy for India to invest in foundational models without ignoring current challenges.
Taking a measured approach, she quizzed if the country is ready to act now or wait till developing foundational models is an accessible and economically feasible idea?”
Public-private partnerships surfaced as a potential solution, especially in sensitive sectors like healthcare.
Joseph noted: “It should definitely be a PPP model… there could be enough oversight, a way to safely build models without abuse.”
Need for Sovereign, But Balanced Approach
The debate reflected the duality of India’s AI journey: an immediate need for innovation balanced against the long-term ambition of sovereignty.
While the speakers differed on timing, they converged on one point: that India cannot ignore the question of investing in foundational models forever.
Whether through cautious adoption or bold investment, the country’s AI future will likely demand a mix of global collaboration and home-grown strength.
The post Should India Build Its Own AI Foundational Models? appeared first on Analytics India Magazine.
Generated by RSStT. The copyright belongs to the original author.